About this episode
Gerry Marletta, SVP of Delivery at Union Square Consulting with nearly 20 years of go-to-market operations experience, joins Eddie Reynolds to explore how Salesforce Headless 360 and AI are fundamentally changing whether sales reps ever need to log into Salesforce again. Marletta's background spans eight years running GTM ops at Argyle (growing the org from 20 to 300 people) and nearly a decade at Union Square, making him uniquely positioned to discuss how AI tools like Claude and ChatGPT can now read from and write to Salesforce directly through the Model Context Protocol (MCP).
The core insight is that you do not need a clean Salesforce to start delivering AI-driven value to your go-to-market team. Rather than requiring months of data cleanup before deploying AI, organizations can now provide guardrails and instructions to Claude that tell it which fields matter, where to find data across 20+ systems simultaneously, and how to synthesize insights for sales reps without ever touching the Salesforce UI. The real shift is moving from "Salesforce is the source of truth" to "Salesforce is one spoke in a data hub orchestrated by AI."
This episode unpacks the difference between reading data (where you can take the training wheels off quickly) and writing data (where guardrails are critical), shows concrete examples like automating call prep from calendar events and extracting MEDIC fields from call recordings, and challenges the old playbook that says clean your house before you can deliver outcomes. Instead, the dinner party metaphor applies: you only need to clean the rooms your guests will visit. The episode also explains why Union Square built their own analytics on SOQL queries instead of trusting the MCP for hard numbers, and how six months of CRM cleanup can compress to six weeks when AI handles the structural work.
What we cover in this episode
-
0:45
Do we ever need to log into Salesforce again? Explores whether traditional Salesforce UI access becomes obsolete when AI can read and write data for sales reps and leaders.
-
5:07
No rep has ever wanted to update Salesforce Examines the friction between frontline sales resistance to data entry and leadership's need for clean, accurate pipeline data.
-
9:19
The 35-field mutiny that got everyone fired Real example of over-engineering Salesforce requirements that caused sales team rebellion and leadership turnover.
-
12:03
What is Salesforce Headless 360 and MCP? Explains Model Context Protocol as a standard way for AI tools to understand Salesforce without deep API knowledge.
-
16:45
Does your Salesforce need to be clean for this to work? Debunks the myth that AI requires pristine data; guardrails and field descriptions matter more than comprehensive cleanup.
-
25:07
Your data can live in 25 different sources now AI orchestration eliminates the need for data centralization; Claude can fetch from Salesforce, HubSpot, Clay, and other systems in parallel.
-
31:56
Updating opportunities from call recordings Use AI to extract next steps, decision-making criteria, and MEDIC fields from call transcripts without rep manual entry.
-
39:09
Why rigid tools kill creative selling Sales engagement platforms enforce linear workflows; AI allows reps to pivot strategy on the fly while still logging outcomes.
-
50:07
SOQL queries vs. MCP: why hard numbers need zero interpretation For board-level analytics, Union Square uses direct SOQL instead of MCP to avoid hallucination risk in executive reporting.
The lines worth sharing
You could spend entire days, maybe even weeks, without really having to log into Salesforce to get things done.
Gerry Marletta · 4:19
No sales rep in the history of sales reps has ever said, oh, I can't wait to log in and update Salesforce.
Eddie Reynolds · 5:07
Your data could live in 2, 3, 4, 5, 10, 20 different sources. We're at a point now where you could definitely spend entire days without having to log into Salesforce.
Gerry Marletta · 27:16
We've learned that garbage in and garbage out doesn't have to be true. You don't have to have perfect data in and of itself.
Gerry Marletta · 46:41
Common questions from this episode
Do I need a clean Salesforce to use AI with Salesforce Headless 360?
No. You need guardrails, field descriptions, and clear instructions for Claude on what data matters, but you do not need months of cleanup. AI can work with messy data as long as you tell it which fields to trust and where to find relevant information across multiple systems.
What is the Model Context Protocol (MCP) and why does it matter?
MCP is a standard that lets AI tools like Claude and ChatGPT understand Salesforce without requiring developers to know Salesforce's API deeply. It translates natural language prompts into Salesforce queries automatically, enabling instant data access.
Can AI automatically move deals through sales stages or close won opportunities?
Contract signatures can auto-close opportunities (this has been standard for years). For losses and stage transitions, guardrails matter. Generally, humans should make subjective judgment calls, but mechanical triggers like meeting-held or contract-signed can be automated.
How do I extract call data into Salesforce without manual rep entry?
AI can listen to call recordings, extract MEDIC fields, next steps, and decision-maker insights, then write those as opportunity notes or custom fields. Reps still validate before pushing, removing friction while preserving accuracy.
Can my sales data stay dispersed across multiple systems instead of centralizing in Salesforce?
Yes. AI can fetch and synthesize data from 20+ sources simultaneously on demand for call prep or reporting. You no longer need complex integrations to centralize everything; orchestration via AI is often faster and more flexible.
Should I use the Salesforce MCP or direct SOQL queries for board-level analytics?
For hard numbers and executive reporting, use SOQL queries directly to eliminate hallucination risk. MCPs are excellent for reading context and prep work, but deterministic data queries should bypass AI interpretation.
SEO meta description
Gerry Marletta explains how Salesforce Headless 360 and AI eliminate the need to log into Salesforce, extract call data automatically, and deliver GTM outcomes without data cleanup.
Target keywords
Salesforce Headless 360
Model Context Protocol MCP
Claude Salesforce integration
AI Salesforce automation
Gerry Marletta
go-to-market AI
headless CRM
call recording Salesforce
SOQL queries
sales pipeline AI
GTM operations headless
Salesforce no login
Read the complete conversation
Auto-generated from the episode audio and lightly cleaned. Expand to read or search inside.
Read the full transcript · 72 KB
GERRY0:00people have complained for years about the fact that our data is all over the place.
GERRY0:03that's not an issue anymore.
GERRY0:04You could let your data be all over the place. Your data could live
GERRY0:07two, three, four, five, ten, 20 different sources.
GERRY0:10As long as you construct and provide the right instruction of,
GERRY0:13let's just say, a process of where to go and pull that information from, let it live in 25 different sources.
GERRY0:18We're at a point now where you could definitely,
GERRY0:21spend entire days, maybe even weeks, without really having to log into Salesforce to get things done.
GERRY0:26going to have to engage in some way, whether that's through slack, through teams on your phone,
GERRY0:31there's going to need to be some mechanism whereby you're viewing and updating information.
GERRY0:35But there is definitely a much smaller requirement for you to have to deal with the traditional Salesforce user interface
GERRY0:42from this point forward.
GERRY0:43And I should say it's actually a lot
GERRY0:45easier for you and your organization to get to a point where that is true.
EDDIE1:13Awesome. Welcome to another episode of Go to Market Science. Today we are going to talk about the question, do we ever need to log into Salesforce again what Salesforce Headless 360 means for go to market? My guest today is Jerry Marlette, our own SVP of delivery here at Union Square Consulting. He's been here like what? Something like nine years.
EDDIE1:32Almost the entire time this company has existed. Yeah, yeah.
GERRY1:37Company was a year old, I think when I started.
EDDIE1:39If that it was almost non-existent. And previous to that, Jerry ran go to market operations at Argyle for eight years, just a really long time, growing from pretty junior to pretty senior with what, like a 300 person org Jerry.
GERRY1:57From like 20 people when I started to 300 at its max and shrank back down at one point. But yeah, yeah, we had seen for a while.
EDDIE2:06You put those two things together. You're going on 20 years of go to market strategy and operations and and yet is this the first podcast you and I have done together?
GERRY2:15I believe so I feel like we recorded one other at some point in time, but I don't think anything came of it. So it's called this the first official one.
EDDIE2:22Yeah. And so I'm excited to have you on the show back again. Maybe, I don't know, I can't remember. But so many of these we do with external guests or Rachel and I do these. But this particular topic, I'm way out of my depth. And I wanted to bring you in, Jerry, to talk to the extent you can.
EDDIE2:38And this is constantly evolving about how Salesforce is enabling AI to communicate with Salesforce and what this actually means for go to market. Why should a crow or a simple care? How is this going to affect the sales job, the marketing job, the CS job, the rev ops job? How is this going to change? Go to market. And so the elevator pitch here is that Salesforce has surrendered to Claude and ChatGPT.
EDDIE3:03Salesforce is co-founder Parker Harris, like literally one of the co-founders of the company is asking the same question why should you ever log into Salesforce again? So Salesforce recently went headless. These two things are tandem. That's like when Parker Harris asked this question at a conference as they were announcing headless, at least as far as I remember, allowing AI tools to communicate directly with the data in Salesforce, which I'm really excited because as much as I'm a Salesforce fanboy and have been for like 20 years, I don't want to log into Salesforce.
EDDIE3:31I
EDDIE3:31don't want to update my opportunities. Nobody does. So we're here on this podcast to ask that very question. So, Jerry, once again, thank you for taking the time to join me today. I'm excited to dive into it with you.
GERRY3:42Thank you for having me.
EDDIE3:45All right. So let's start with this main question. Do we ever need to log into Salesforce again. And I'll start there. I'll just I'll let you take over. You want and I'll break this question down okay. A little bit more specific.
GERRY3:56Yeah. In short, it depends on who you are. Depends on what your role is, depends on what it is you're trying to do. We're at a point now where you could definitely, depending on what you do within your organization, you could spend a good chunk of your day not ever having to log into Salesforce. You could probably spend entire days, maybe even weeks, without really having to log into Salesforce to get things done.
GERRY4:19You're going to have to engage in some way, whether that's through slack, through teams on your phone, but there's going to need to be some mechanism whereby you're viewing and updating information. But there is definitely a much smaller requirement for you to have to deal with the traditional Salesforce user interface from this point forward.
GERRY4:40And I should say it's actually a lot it's a lot easier for you and your organization to get to a point where that is true.
EDDIE4:48And so what's the benefit here? Like from my perspective,
EDDIE4:51I see two things. Number one, no sales rep in the history of sales reps has ever said like, oh, I can't wait to log in and update Salesforce. Even when I was a Salesforce, when I was a sales rep at Salesforce, we didn't want to log into Salesforce and update Salesforce, right?
EDDIE5:07Nobody wants to do that. And then from a leadership perspective, whether you're a CRL, a CMO, Rev ops, CEO board, you're relying on data that is populated in Salesforce previously manually by reps, increasingly with automation and now with AI. And so from my perspective, this is really, really exciting because we can not only go to the frontline sales rep, but we can also go to The Crow and the CEO and the board and say you will have fuller, richer data that will tell you about where you're headed as a company and what's working and not working because of this new capability.
EDDIE5:46Or you could.
GERRY5:47That's true. Yeah. You're removing a lot of the friction when you've got that sentiment from users on the team, from A's, on the team who hate logging into Salesforce, who hate clicking through things, who hate jumping through the hoops that the operational side of the business, the rev ops or the go to market ops team has decided is necessary in order to have clean and accurate information that causes friction.
GERRY6:13Because that's a lot of extra things that, like you said, sales reps don't want to have to do, and then therefore they're kind of fudging it, you know, they're they're clicking for the sake of clicking. They're not bothering to put in as much effort, and you're relying entirely on them to begin with to make the judgment call on some of that information.
GERRY6:29When you have maybe more accurate and more trustworthy sources, like call recordings, like direct email exchanges that you could rely on to glean that information from. So we're now at a point where you can you can meet the team where they're at at, have to say no and have to compromise far less often in terms of how you configure the experience for just we'll use sales reps here.
GERRY6:54You can say no and have to compromise far less often now than you ever had to before.
EDDIE6:59Yeah, I mean, this is something I've come full circle on because I remember my first implementation of Salesforce. I was part of a small sales team. I was doubling as like the Salesforce admin, building out Salesforce for our sales team, and I was butting heads with the senior people on that team that were saying we refused to log into Salesforce.
EDDIE7:17Now this was on Wall Street, different culture than B2B tech, right? But we had directors, managing directors, principals, very senior people bringing in a lot of money. For context, my firm helped to raise private equity funds, and so we were like an outsourced sales team, private equity firm would hire us. We'd go out and call institutional investors, set up meetings and help drive investments of multi-billion dollar funds.
EDDIE7:41And so you had reps. They wouldn't be called reps, but you had people that were bringing in hundreds of millions of dollars of business and millions of dollars of revenue for my company, saying, I'm not going to log into Salesforce. And previous to this, we didn't have Salesforce, and these guys were calling analysts and they would literally leave voicemails and the analysts would have to update the I'm going to use air quotes for CRM because it was Excel prior to this.
EDDIE8:08And so I was tasked with solving this problem. And I'm being told by these senior people, we're not going to log into Salesforce full stop period. I fought with them and lost that fight. I soon thereafter went to Salesforce where that was very much the culture. If you don't keep Salesforce updated, you will not have a job. If it doesn't happen in Salesforce, it doesn't count kind of thing, which is a mantra that a lot of companies live by.
EDDIE8:34But you and I have seen both sides of the spectrum, even as recently as the last couple of months, where we talked to cross on a regular basis that say, I would fire certain people on my team for refusing to update Salesforce, but then I would get fired the next day. So I can live in this utopian worldview where we fire anybody that doesn't do exactly what we want to do in Salesforce.
EDDIE8:58But there's two problems with that. Number one is how much can you get them to do? Because you and I have made this mistake where we let a customer drive us to this destination many years ago, where every time a rep got off the phone, they had to update 35 fields in Salesforce. And the result of this was like an absolute mutiny of the sales team, where all the people we worked with and us included all got fired.
EDDIE9:19We're talking sea level people got fired over this decision and we should have pushed back harder. And then fast forward to today and this is still a problem. And so for me, I'm looking at this and I'm like me personally, I don't want to log into Salesforce. I have Sarah, our executive assistant, do 99% of the updating to Salesforce only because like, we have yet to build out all the perfect AI stack to replace this.
EDDIE9:44But for years and years I have been avoiding logging into Salesforce myself while simultaneously telling our clients they need to force their reps to log into Salesforce, right? I've also said on the side like, hey, like hire a bunch of executive assistants. They're amazing. Nobody listens to me there. But now that we have AI, I think it's an incredible opportunity to not only improve the profession of sales, but also improve, go to market analytics and go to market leadership and the company as a whole.
GERRY10:10Yeah, absolutely. Again, you're removing that friction. You're getting honest, truthful information all aggregated in one place or maybe in multiple places. You know, if Salesforce is where it needs to live, if at the deal level, at the opportunity level is where you need those facts to live again, the rep may just be able to key in their updates again in a channel to a bot.
GERRY10:38They may be able to dictate it, and depending on the tool they're using. Or again, you can pull it from the first party sources so it doesn't have to filter through their brain. Or maybe you want that, right? I think it's important to mention here that not everything should be fully left up to an LLM to make a judgment call on all those decisions, but it's a lot easier to run some updates past the 80s.
GERRY11:03Hey, does this all look great to you? Let me know if there's any tweaks that need to be made, and then I'll go push it into into Salesforce under the opportunity for you, than it is for them to come up with that from scratch.
EDDIE11:12So absolutely. And so we're going to get into brass tacks here. I want to give folks listening to this some key takeaways. If you're listening to this and you're thinking how can I, you know, access sales headless sales. Sorry Salesforce headless how can I improve. You know, my analytics. How can I make life easier for my sales reps?
EDDIE11:32We're going to get into that. If you're the person that's actually trying to build this stuff out. We're going to share some some Intel here. Now obviously like the devil's in the details. And we're not going to get into like complex workflows, but at a high level we're going to touch on that. And so if you're listening to this stay tuned.
EDDIE11:46I want to get into some specifics that you can take away and take action on. But I want to start by setting a foundational understanding for anybody listening to this. In case you didn't have this, because this is something I didn't have when I hear Salesforce Headless 360, my first question is what the heck is Salesforce Headless? So let's start with that, Gerry.
EDDIE12:03And then we'll kind of peel the layer, the onion back so that people listening to this can have like a foundational understanding of what this is and what's possible. So what is headless 360.
GERRY12:13Yeah. So headless 360 is really just a of label for a concept of how to approach, you know, systems interacting with Salesforce and the people then also interacting with Salesforce. As anybody who's listening may know, Salesforce has had an API available basically since day one. And that API is a let's just say it's a dictionary of how one machine can talk to the Salesforce machine, how a, you know, a third party tool, something that you buy off the shelf or something that you may create on your own, can interact with the Salesforce.
GERRY12:48However, that API needs to be understood and needs to be known by the developer, let's say, who's building that connection to Salesforce to to do some of that automatic writing to and reading from Salesforce. So that's nothing that's new at all for Salesforce. What makes the the sort of headless 360 truly headless is the the rollout of the Salesforce MCP.
GERRY13:12So the MCP that's the model context protocol. What that is is that is a MCP is like a standard. It's a standard way for all of the AI and LM tools that are out there, whether it's Claude, whether it's ChatGPT, whether it's Gemini, whether it's something you're building on your own to understand basically instantaneously how it can translate things that the end user like myself using Claude, asking questions, how it can translate that into a language that Salesforce would understand without Claude having to understand Salesforce's API in depth.
GERRY13:53So Salesforce publishes this, and the MCP is a combination of resources, and it allows you, me to say, to write a quick prompt in Claude, hey, tell me about my open pipeline. And that MCP is what enables Claude to very quickly know it needs to look at opportunities, let's say, and it needs to look at opportunities where it's one of the open stages, for example.
GERRY14:17So that's the that's really the big sort of radical shift. That's the element that's made this announcement relevant and available.
EDDIE14:25Yeah. And I think that's really helpful. And one other thing that I think is really interesting is the ability for an MCP to help prevent hallucinations by giving you the raw data. Right now we're going to get into that. And I think that this is really important because again, I'm trying to frame this conversation to be a little technical but not too technical.
EDDIE14:46If a crow or a CMO is listening to this and they're thinking, why should I care about this? What can I do with this? I want to answer that question here in this podcast, and at the same time provide like a high level guidance if they want to hand that off to somebody else in their organization to say, like, let's go run with this.
EDDIE15:00Of course, you guys can always call us and talk to Jerry and hire us for your help. But the point of this podcast is to try to put out content that you can take action on yourself if you don't want to do that. And the way that I think about this is, is like, first and foremost, we need to be able to trust our data, right?
EDDIE15:17I literally just saw a V over a data rails put together an AI video and it was hilarious. I saw this right while I was prepping for the call. And in the AI video you have like a CFO of like a big company prepping the board deck, and he's in Claude and he's just asking Claude to just prep the board deck.
EDDIE15:35And then of course, like, all the numbers are wrong and he loses his job and then he becomes homeless and he's like, fighting raccoons on the street. And, you know, it's just supposed to be funny, but like, it touches on a really important point, like, would you trust claw to do your data analytics in a board meeting? And the short answer is no, you need to verify this data, but it's one end of the spectrum to say, like, let me just go into Claude and ask a bunch of questions and have it prepare my board deck for me, and then I'm going to show up in a board meeting and say a bunch of things
EDDIE16:03that I can't rely on. And it's another thing to say, well, I'm going to start in Salesforce and I'm going to build out all these custom reports, and Salesforce and Salesforce reporting has limitations. Instead of trying to give myself a leg up and we're only talking about analytics. There are many other use cases here, right? There are sales reps prepping for calls, figuring out who they should call.
EDDIE16:25There's marketing like trying to build out contact lists and things like that. So they want to maybe do targeted ads to there's all kinds of use cases here. And so my first question is like, Jerry, how do you think about getting quality data out of the system? Let's start with kind of the easy one. How important is it to have a clean Salesforce and clean architecture for this?
EDDIE16:45If we have a bunch of custom objects and a bunch of custom fields and we have redundant fields and the data is like an absolute mess. When we were prepping for this call, you talked about like having descriptions for fields, things like this. If we try to just plug the MCP in a cloud and start asking questions, what are we going to get if our Salesforce architecture is an absolute mess?
GERRY17:04Well, you're you're not going to get hallucinations necessarily. And that's a word that often gets done around when it comes to llms. It's not so much that you're going to have hallucinated information where where Claude will just stick with Claude here is making things up. Instead, what you're going to run into with scenarios where Claude doesn't necessarily understand the ins and outs, it doesn't have the institutional knowledge that your own in-house Salesforce admin may have to understand that.
GERRY17:29We'll just stick with the example you were saying with fields, with descriptions. If I log into a Salesforce org today and I take a look at an open opportunity, I might be presented with 20 or 30 fields. And hopefully those fields are somewhat clearly labeled, and I might be able to quickly get up to speed and understand how how a sales process is run for that team.
GERRY17:49When you're utilizing cloud and you're utilizing an MCP, it doesn't see Salesforce that way, whether you're using the API or an MSC, I should say. But regardless, it's not seeing things the way a human user sitting in front of Salesforce does. So that's where one of the first kind of risks you would run into is going to be that there is kind of a self-imposed that a guardrail is put in place by only showing so many fields to the user.
GERRY18:12So without giving Claude any additional contacts or any additional instructions or guidance, what you're likely to run into is every time you go through this exercise of prepping that board deck, you have to repeat yourself, and you're gonna have to correct it and say, look, don't use this. This field you're looking at called, you know, let's say you're dealing with recurring revenue.
GERRY18:32Don't use that, Mr. Field. Use this other RR field instead. Well, how would Claude know that? How would Claude know that the standard amount field isn't really relevant for some of the metrics you're trying to put together? It doesn't, because all of those fields available on your opportunity essentially created equal as far as it's concerned. It needs to basically give the same amount of weight or the same amount of importance to all of them as it's trying to figure out what you're asking it.
GERRY18:58Again, if you are a new sales rep to the organization and you go in and you look at an open opportunity and you just see one field on a screen with a dollar value in it, you can probably pretty safely assume that's the important one. That's the one I need to think about when I'm thinking about my pipeline value.
GERRY19:13So that's kind of the first big thing to consider is, is the cleaner from a build perspective, from a configuration perspective that your Salesforce org is, the less correcting and the less guardrails and less rules you would have to put in place.
EDDIE19:29Let's say that Salesforce is an absolute hot mess, right? And we're not exaggerating here. We've seen so many instances. You said 30 fields. How about 300 okay. And many of them are redundant. And some of those fields haven't been there in a long time. And let's say that they're not on the page layout. I don't want to get too deep into the weeds here, but I'm assuming that Claude wouldn't even necessarily know that it's on a page layout or not.
GERRY19:50Has it? Wouldn't have any idea. You could. That's maybe one angle. You could try to provide some guidance to say, because it can see that configuration. There are, you know, the way the page is configured lives in code somewhere. So you might instructed to only look at fields that are on a page even that's not necessarily good enough.
GERRY20:06So I think the question you're asking is what do you do in that scenario. You've got 300 fields and most of them are legacy and outdated and not relevant. Well, now you've got a compromise or you've got a situation on your hands. So you need to determine, am I going to rely on Claude to prep this board deck?
GERRY20:19That's two and a half an hour, or am I going to just have to do with the old fashioned way and invest some time after this to documentaries a little bit better? So you don't necessarily need to go in and consider this massive initiative to clean up your instance of Salesforce, just so that you can use an LLM or a tool similar to that to build that board deck going forward.
GERRY20:39But you probably need to think about having some instruction somewhere to guide it. These are the one or 2 or 3 key fields when I ask for, you know, value or you look at these fields include opportunities owned by these reps and the reps that have a role of sales rep versus the ones that have a role of a rep or something you know you want to you can put some thought in.
GERRY21:03The way I think about it is pretend you just hired an incredibly, incredibly intelligent but fresh out of college. I don't want to say intern because you hired them, but you know somebody who's really new to all of this and they have no context, but they can see it all. And given the right guardrails is a term I keep using, given the right guardrails and the right set of instructions that you could refine over time, they can they or in this case, Claude can turn out some really powerful information for you very, very quickly.
GERRY21:33But if you just set them free again, they have no context awareness, they have no institutional knowledge. They're going to make guesses at things and they're going to error on the side of appeasing you. I think that's probably one thing that's really important to also consider when you're dealing with any of these AI tools, is they want to give you a positive answer, sort of natively, unless you provide other direct instruction.
GERRY21:55They want to be yes men or yes people. So you have to be really cautious about that.
EDDIE22:00This is exactly where I was going with this, and I want to bring this back up high level because I want I want to try to thread the needle between getting in the weeds and being high level. And I want to talk about actual use cases. Right. And so, Jerry, feel free to correct me, but I'm just going to sort of think out loud here about how I would solve for this and why, let's say that I want to do I want to build something that enables my reps to do better account research and or let's say like update the opportunities in Salesforce.
EDDIE22:26Right? Okay. So let's just say that like, let's just combine these two things and say that I want to prep for calls with, you know, folks that are in our pipeline and I want to update our pipeline. So the first thing I'm thinking is, is that we need that context in those instructions. And so what I would want to do is I'd want to go and let's say that, like I'm new to the organization, I don't know anything.
EDDIE22:45I'd want to go in and look through the opportunities and figure out which fields are relevant. How do we interpret those fields? How does Claude understand what those fields are if they don't have the description, field, etc., for the purpose that we're trying to achieve here, which is we want to give our reps the information they need to get on their next call with opportunities that we have in the pipeline.
EDDIE23:04And the reason that I'm trying to get really specific here is, is there's nothing I hate more than this idea that, you know, we as Salesforce professionals are going to go in and clean up every object and every workflow and everything. And for what? Like, hey, Mr. or Miss Crowe, give me 100,000 or $1 million as an employee, a team of employees, a consultant, what have you, and your benefit is that you're going to get a nice, clean Salesforce.
EDDIE23:33Why do we want to clean Salesforce? Well, we want to drive outcomes. So I don't want to clean Salesforce. I couldn't care if Salesforce is like the biggest dumpster fire ever. And I say this having worked at Salesforce where aspects of Salesforce were a dumpster fire, but the reason they were a dumpster fire is because the organization said our priorities are to drive revenue and forecast accurately.
EDDIE23:53And these aspects of Salesforce being a dumpster fire don't prevent us from doing that. So that's not where we're going to reinvest our resources. We're going to invest our resources in driving outcomes. And so if we want to have really clean pipeline management, tell me if I'm wrong here. What I'm thinking here is, is that we need to go through and understand how opportunities and Salesforce work, as well as the related objects like accounts, contacts, activities, etc. and we need to provide the context and instructions to Claude so that it can go through the MCP and know what fields to look for, how to interpret those fields, and how to give us the information that
EDDIE24:26we want alongside the instructions to say, for example, give me like a call prep or an account research document. This is how you would do this. This is the fields you would look at. And oh, by the way, possibly using other systems, we can go do a web search. We can tap into zoom info or Clay via their MCP.
EDDIE24:45We can do a lot of different things to then feed that rep the information that they need to get on their next call.
GERRY24:51Yeah, and I know we don't want to dive deep into like prompt engineering here, but another layer you would add to that also is
GERRY24:56every day or every hour, depending on how frequently your reps calendars may shift have clogged. Go and take a look at all the invites on their calendar for the day. All the events on the calendar for the day.
GERRY25:07Look at the domains of the people on the invites, and look at the subject line and prep those those those, research those prep sheets ahead of time, rather than the AI having to remember to pull that, because there's nothing worse than being in back to back calls and two minutes before the call realizing, crap, I didn't you know, I didn't prep for this.
GERRY25:24You can have a scheduled task that goes and preps that for you and can infer who the customer is that it needs to do the prep for without even having to lay your hands on the keyboard. So you can take these things even a step further and remove the human without there being any detrimental effect. I will say, you know, there's I'm I'm overly we've talked about this before.
GERRY25:44I'm much more cautious in some ways than you are when it comes to some of these builds, especially as we've done some experiments internally. But when we're talking about reading information and not writing it, you can you can take the training wheels off a little bit more quickly. We're talking about writing to the database. You want to be a little more careful and provide a lot more guidance to begin with.
GERRY26:04I think that sort of answered your question or follow it up on your statement. Right. But.
EDDIE26:08Well, I think that there's two different things here. So you're talking about automating the account prep. And I totally agree with you. I mean, this is something I've been doing for years with Sarah, our executive assistant, where every morning I pop in, I have a single sheet of paper and it says, this is what I'm doing for the day she prepares.
EDDIE26:22She's been preparing that for years. And then there's a link to the document for each and every call, including this call. And then it says, like, here's everything that you need to know. And for years I've been walking into phone calls, especially sales calls, because we have it so dialed in where my prep is less than five minutes and these are important calls.
EDDIE26:39I'm talking to executives in large companies. You know, about the potential to work with us and pay us hundreds of thousands of dollars. And we've got this dialed in to a point where I know exactly what I need to talk about. I've got everything in one place. And whereas previously, like even when I was working at Salesforce and we had all this data at our fingertips, I was spending half an hour, an hour prepping for a call for a much smaller deal.
EDDIE27:03And so I think this is monumental. And the only gap between what we've been doing with Sarah and AI is just building this. But the concept is they're.
GERRY27:11Just making the connections. And people have complained for years about the fact that our data is all over the place.
GERRY27:16That's not a that's not an issue anymore. You could let your data be all over the place. Your data could live in two, three, four, five, ten, 20 different sources. As long as you construct and provide the right instruction of, let's just say, a process of where to go and pull that information from, let it live in 25 different sources.
GERRY27:35It may not be the most efficient utilization of your tokens with your tool, but that's okay. That's no longer a challenge. It's no longer an issue for this process.
RACHAEL27:44Quick pause. Everything we talk about on this show. Diagnosing go to market ops. Prioritizing projects for revenue. Impact processes. Metrics. Insights. Building a predictable go to market engine. We've built frameworks for all of it. They're free and undated on our website. Union square consulting frameworks. The link will also be in the show notes, so make sure you check that out.
RACHAEL28:07All right. Back to the episode.
EDDIE28:09I think this is another big issue, because when I left Salesforce, I was effectively the VP of robots at a fintech startup. I want to say small, but they were very heavily funded. And what we were doing is we had an app where Wall Street traders could trade securities in a like basically a, a hidden room so that they could not reveal the other person they were dealing with.
EDDIE28:29Because if you're selling like $20 million of credit, you don't want people to know that you're doing that because that could inhibit that could that could affect the price that somebody would pay for it, which is not good. And so on the one hand, like our Salesforce was really, like dialed in even before I went there. And we had all of our accounts and contacts and everybody like, dialed in.
EDDIE28:50On the other hand, we had this mountain of data from the usage in the app or like the people like our main goal. Like I worked there sort of at launch, our number one priority was getting people literally into these rooms in our app so that they could trade hundreds of millions of dollars of securities in a day.
EDDIE29:05And all our sales team did all day long was like, jump into Bloomberg Instant Messenger, because that was the the communication medium that, you know, was working with our clients and messaging them to try to get them to go into these rooms.
EDDIE29:19On the one hand, we had like a pretty small data set in Salesforce, and on the other hand, we had this massive data set of every single like room, entrance and indication of interest and actual trade that was going on.
EDDIE29:31I mean, thousands and thousands and thousands of rows of data every single day, even as we were just launching our product. And I made this mistake of saying like, well, everything should be in Salesforce. Sales forces are single source of truth, and then trying to push all this data in Salesforce. And then for what? Like the reps were moving too fast to access this data.
EDDIE29:50I mean, this stuff was happening in real time, right? And so like we had some other dashboards outside of Salesforce, but it just felt like one of the biggest mistakes I've ever made in Rev Ops was like buying into this idea that we had to have all the data in one place and like, for what reason? And Salesforce is not like a really great place to store large sets of data.
EDDIE30:10And you know, the only reason to have it in Salesforce is like, if you're like looking at an account or a contact and you're prepping for a call and you have relevant information sitting there that you're going to use to prepare for your call, that makes perfect sense. But having these large data sets doesn't. And now instead of doing this massive integration project, you could just say, well, why don't we just toggle two, three, four different systems with AI?
EDDIE30:33Because it's possibly even easier to do that than to centralize this data into one system, even ignoring like the implementation work, just for the rep to be able to go into Slack or Clod or whatever the tool they want to use as their interface and just say like, tell me what's going on with this customer. You provided instructions and it's like, cool.
EDDIE30:50Like, let's go check Salesforce, let's go check HubSpot, let's go check the website, let's go check this and that and the other. Let's go check Clay. We can check all these sources at once simultaneously and provide them with the information they need at their fingertips, lightning fast in comparison to what we could a couple of years ago.
GERRY31:06Yeah. And it's and it's the inventory on demand piece of that too. You don't have to have gathered all that information from the sources ahead of time, so that it's there when you are trying to do your call prep,
GERRY31:14you get it when you need
EDDIE31:16Yeah. And I think that that's really exciting. So let's talk about updating Salesforce. Right. So yeah, you and I have been talking to like literally every vendor under the sun that does some form of call recording and up and potentially updating Salesforce with that call recording. And I'm hesitant to like name names yet, but there's 1 or 2 players that we're really excited about, and this idea that AI can just listen to my call transcript, not to mention the fact that we can integrate emails and update the Salesforce record and tell me who I met with, what we talked about, what the decision making criteria is, medic fields, things like that, to not only inform
EDDIE31:56the rep, but also leadership, Koestler's, S's, etc. on what's going on with a deal, improve our forecast accuracy, etc. etc. this is really exciting to me, but and I think you're going to agree with me here, like there's a limitation to this. I personally would not want a rep having AI move a deal through the funnel. I would want them to make the ultimate decision as to this is where this deal is.
EDDIE32:23This is in stage two. Stage three, this is in pipeline, out of pipeline. I'm forecasting this as such. Maybe update the next steps or validate the next steps manually, etc. but like populating like what the decision maker said or what what they said about the decision making criteria into a field like decision making criteria on the opportunity does two things.
EDDIE32:47One, it is a massive time savings to the rep. Actually, it's three things. It prepares them for the next call. So much better. And then thirdly, it provides management a level of visibility that I think no amount of threatening to fire people will ever provide you.
GERRY33:02I mean, yeah, in terms of the the next call prep, sure. If you don't like putting in your nose from the last call to begin with. Well, now you're entirely relying on your own memory to prep for the next call for that element, because no amount of exec assistant is going to be able to to pull that up unless they were on the call and you had the recording and they're reading through it.
GERRY33:20So yeah, if you can get the notes from the call that just occurred or the exchange, whatever it was in a system to be read by something else in advance of the next one. Yeah. You've set yourself up for success there in terms of updating the deal to the pipeline. I'm going to agree with you broadly, but I'm actually going to say it might depend on your motion and your sales process, you know, without getting in the weeds, depending on what your stages are and depending on how you sort of get those, there may be services that could be people, but there's others that they should.
GERRY33:53I absolutely, I think 99.9% of the time would say, never let that thing move it to a win or a loss. That ultimately is a human judgment call. But depending on what, of course, I'm not going to be able to come up with a good example. Wait.
EDDIE34:07No, that's that one's interesting. I want to test you on that. I would argue with you on that one. Like if the dock you sign is signed, is that not a win?
GERRY34:13Okay. That's fair, that's fair. No no no you're right, you're right I doc I'm talking about from like something conversational let's say in a call or and sure.
BACKGROUND34:19Sure sure.
GERRY34:21In that regard I would not let it know. I mean, we've been letting contract signature close one opportunities for years now. That's been the first piece of automation we roll in with a DocuSign implementation, let's say. But losses, those absolutely need to be a human judgment call for a thousand different reasons, and wins for any other reason, you know, shouldn't be.
GERRY34:41But there's probably, you know, there's probably some scenarios where there might be like your stage 40 or stage five or your stage 1 to 2, maybe you could, depending on what your sales motion looks like, depending on the size of the product and the size of the customer and the prospect, there might be some scenarios where you could do that.
EDDIE34:55I think that that's an important point. Like if you just look at our own like sales process and what Salesforce looks like, I mean, we don't use leads, really, or at least I don't know if we use leads. You might have built some automation in there that I'm not aware of, but I never look at Leeds for one reason.
EDDIE35:11Just because I think it's really important to multi-threaded accounts. And so we have a stage where we also have a lot of partner opportunities. So we have a stage where it's like stage zero. Many companies have this, it's not qualified. And for us it's so zero, zero zero. It's like for me our first stage means that we haven't even set a meeting yet.
EDDIE35:31And then the next stage is that we actually set a meeting. And then both of those go before I qualified. Where qualified is like, we've had a meeting and there's something real there. And so you could not even with AI, but you could just automate and say like, well, once the meeting is sad, or once the meeting is held, it's going to be moved into these stages, right?
EDDIE35:51And then I think that that like takes away some ambiguity so that like, the reps can spend more time thinking about how to close deals and how to source more deals, rather than having to get into the minutia like, oh, like, which of these deals have I met with? And which one needs to go from this stage to that stage?
GERRY36:11The beauty of
GERRY36:13the ability to leverage AI alongside Salesforce, the ability to leverage AI in general nowadays with all of this is the elimination of of having to say no or to compromise. And so I think for somebody who's listening to this conversation and, you know, I myself, just from reading other pieces of content that are out there, it's funny because you can say, yes, we can do that so much more and still account for every unique scenario from one customers process to the next, or from one team's way of doing things internally within your own or to the next.
GERRY36:48That this doesn't have to be a one size fits all, and at this point it's maybe a bit extreme to say, but anyone who's offering a product who is trying to sell something, that is.
GERRY37:03How do I want to say this? Who's trying to appeal to everybody is inevitably going to have to put some restrictions in what can be done, whereas with a little bit of know how and a little bit of time and effort and relying on what others have successfully done, like ourselves, you can go and roll your own way of doing these things, and it's going to do it's going to do it right.
GERRY37:25And your business doesn't compromise for it. I mean, just the whole thing, though, to customers, because we've been dealing primarily with Salesforce over the last ten years here. You know this, I've shared this video, like, we often have to say, we can't do it quite like that. We can't build you a report quite like that. Or we have to buy another tool that gives us the ability to do those things.
GERRY37:47This eliminates all of that, so you truly can customize to your own unique scenario, as long as you do it intelligently and smartly and know that there are some risks. And so you want to make sure you're putting guardrails and protections in place. Yeah, more often than not, somebody in an operations role can say yes to leadership to find ways to use these tools to be more efficient without, in some cases, having to buy something off the shelf.
GERRY38:11The couple of the couple of vendors you're talking about that we're really excited about. It's funny, if you think about it. One of them, one of them can do can write updates to Salesforce around medic fields, but only their medic fields, only the ones that come with their package that you install. If you've been using medic for the last, I don't know how long.
GERRY38:31Ten years within your business and you've got your own fields to capture those data points. You're you're up a creek. You've got to move everything over into their fields and rebuild all of your reports. So that's not that's not good. It seems like you're looking at this tool can get update my medic after a good conversation. Well yeah.
GERRY38:47And now you're admin teams got a lot of rebuilding to do.
EDDIE38:50I think that that's a really, really important point because like whether we're like inputting information or we're extracting information, what we're ultimately trying to do is a go to market team is to meet our customers where they're at. And we're trying to provide the unique experience that they want that will result in the right people, in the right organizations buying from us.
EDDIE39:09Right. Like I always try to strike this balance where, you know, there's this concept of the customer is always right. And then like, no, that's not true. Like, if you have somebody that is going to end up becoming a net negative customer for you because they won't pay you enough money, or they're just going to be dissatisfied with this out of the other, then they're not.
EDDIE39:27Right. But within the context of like the right customer, that could potentially be good for you. We want to provide the experience that that turns them into a profitable customer. Like the flip side is is like, yeah, sure. Like customer is always right. I want to buy from Salesforce. I want Marc Benioff to be my personal CSM. It's like, okay, like that's not scalable, right?
EDDIE39:47But I think that AI gives you such a greater ability to get closer to that point where we can provide the unique customer experience that our customer wants, and also do that in a way that we can maintain profitability and serve that customer profitably. And when we pull that into Salesforce, I think it gives us the opportunity to get a lot more creative in the way that we think about going to like in the way we think about go to market, if just from a pure reporting perspective, because we have been hamstrung by Salesforce for so long to say like, well, Salesforce is an architect of this way, so we can't get this report the
EDDIE40:20way we want it. We could download these two or 3 or 4 reports in Excel, and we can manually match it up all together if we wanted to spend all the time doing it. But it's not practical to do that on a regular basis. Now we can go build something in cloud and say, give me this custom report that's virtually impossible to build in Salesforce, possibly give me this custom report that's not only impossible to build in Salesforce, but also pulls data from these three other sources that I don't want to build an integration for and give me this information in real time, either like on an automated schedule or whenever I request it, or
EDDIE40:52both. And now I've got this visibility that I can get in a unique way. And I'll share with you a very concrete example. I've been doing sales for, like, I don't know, 25 years or something like that. And when I ran into sales often outreach, my first reaction was, this is amazing. I tried to build something like this for myself years ago, and obviously this is a much more robust product, but then I've failed to use it.
EDDIE41:15We've had sales, loft and outreach for years that are kind enough to to give us some free licenses, and I've tried to use it again and again. But what I get stuck in is, is that I'm sitting in this rigid flow, and it tells me I need to send this email to this person. But then I realize that, like, oh, we know this, this other person.
EDDIE41:33I could ask this other person for an introduction or I didn't think that they're like actually like on Twitter. And I could just jump into Twitter and comment on their tweet. And then I want to remind myself in two days to follow up, or maybe tomorrow, or maybe in three hours to see if they responded to that tweet, or I know that they're going to go to this conference, and I should go reach out to them and see if they want to, like, or maybe they're in Pavilion and I could jump into the pavilion slack.
EDDIE41:55And there's a million things that are these like creative ways that I've thought about trying to go after this customer. And it inhibits me from using sales, their outreach. And I end up just back in old school Salesforce with this list of people, because I'm saying I just need to know who to reach out to, and then I might get creative with element to reach out, because in outreach, it's 17 different clicks for me to like add the task to do the thing and it just becomes prohibitive.
GERRY42:21And you could set up your instructions with an LM to say, when we're thinking about the next thing I need to do here, just remember, these are all the different avenues I have available. So if you're like me and you have a terrible memory, and it's nice to just have to think about that once, and then when it comes time for the next step, you don't have to do a bunch of those clicks, and you don't have to remember that you have 17 different methods in which you could reach out or follow up.
GERRY42:45It can even say to you. Hey, Eddie, don't forget, you could leverage any of these channels. And you go, oh, that's right, I forgot. I could leverage the Pavilion slack channel as a as a potential avenue to follow up.
EDDIE42:54Well, and it's even more specific than that. Like, the challenge I run into is like, okay, step one. Like, let's jump into LinkedIn and see who comments on LinkedIn or post on LinkedIn. Let's comment on their LinkedIn post. That's a great idea. Everybody says, go do that. But then the problem is, is 99% of prospects aren't posting on LinkedIn.
EDDIE43:07So what you end up with is a list of 200 people, and you go through 200 people to find the two people that actually post. That's incredibly inefficient. And then you have to go through like now there's like six steps where I like, comment on the post and follow up two days later, etc. and I'm just giving like a, like a very random example.
EDDIE43:24But there's many examples like this. Whereas I could just tell like Claude, like, hey, they don't post on LinkedIn, skip to like step seven, you know, and it's just really hard to do in these rigid tools. And I'm not trying to poo poo all over outreach and sales loft, but because they've built this system to be scalable across like hundreds of thousands of sales reps, they have built a system that's so rigid that you can't get creative.
EDDIE43:45And I'm excited about Claude's ability to allow anybody in go to market, not just salespeople, but also marketing and CS to get creative.
GERRY43:54Yeah. And and that means that you can deliver ROI to your organization, or we can deliver ROI to a customer organization so much faster. Going back to the messy org like we used to have to say, like if your house is a mess, you know, we can't do anything with it until we clean up the house. And it's like, well, that's I don't want to spend the next six months cleaning up things, but if you have a machine, you can go in and it just knows where everything is, regardless of the clutter.
GERRY44:24On day one, we're delivering insights versus on day 366, we can say, okay, we've cleaned up your org. We're ready to start taking a look at the data itself now, and try to figure out how we can help you make better decisions.
EDDIE44:38Well, and this is a real thing that we face. So like if somebody's listening to this and they're like a VP of Rev ops or even a crow, or trying to work with their head of Rev ops, I think that we've probably even made this mistake in the past, where there's this inclination to want to like get everything clean and then we get this pressure from cross.
EDDIE44:55And nobody's saying this explicitly, but it's kind of like, hey, I just started this job. I'm probably going to get fired in 18 months. I don't have six months to sit around and clean house before I can show results. I need results yesterday and over the course of ten years of doing this with customer after customer, and I think you get a little bit more leniency when you're like in a full time role, and maybe your boss isn't like new and they're on day one.
EDDIE45:17We don't have that luxury, and we've gotten so much pressure from customers that like, we need to ship results yesterday. And for me, the lens that I've looked through with this and feel free to like, test me on this since you're the one actually like doing this work and leading our team. Doing this work is like, I don't care how messy the house is, if my job is that I need to host dinner, to use a metaphor here, then all I care about is where are our guests going to go?
EDDIE45:40There. Going to come to the living room and the patio, and they're going to be in the kitchen. Okay, we need to clean up the living room, the patio and the kitchen. We need to figure out what we need to do to get them in and out of the house to have this amazing dinner. If the third guest bedroom is like an absolute mess.
EDDIE45:54It doesn't matter. Our job here is to put on the dinner. And so if our job here is to figure out how are we going to improve outbound, we need to think through the lens of who do we need to contact, how do we prep, how do we get them the best messaging, etc., etc. and we don't need to go back in and like clean up, like all the workflows and automations that don't impact that.
GERRY46:15No, it doesn't care. The LM machine. The AI machine doesn't care how messy it is as long as you give it guidance early on. We are just as guilty as because a lot of other people of using that that concept of like garbage in and garbage out. And what we've learned is that doesn't have to be true. We do not have to have a perfect, you know, structure to our database.
GERRY46:41We do not have to have perfect data in and of itself, necessarily, yet either. All we have to do is to take a lot of the knowledge we already have and can infer and spend that time and energy once, and iterate on improving how that tool goes and collects and does its thing. And you can even use AI and those tools to improve the quality of the data to improve the structure of your database.
GERRY47:09So maybe I'm getting ahead of before you wanted to go to this conversation, but it's worth mentioning rather than spending six months cleaning up your database, you know that could be scaled way down to six weeks instead of months. Because not only can these AI tools deal with the data, the metadata, the actual structure of the databases it can offer, and even build out better ways of doing those things in a way that no 1 or 2 or team of, let's say, admin developers could do in the same amount of time.
EDDIE47:39Oh yeah, I've already run like clawed coworker to do like fuzzy duping of like records. And you're just like, oh my God, this is amazing. Like we used to hire an outsource company to do this and and now, like maybe it's not perfect, but my God, it's pretty good.
GERRY47:55Yeah. I mean, when we started engagement with the customer, one of the first things we do alongside other elements of our diagnostics, just on the technical side, we're going to want to go in and see what is your CRM structure look like, what what tables you have, what are the fields, who has permissions, all of that. And it's you know, it's the process doc for that is a labor laborious, I should say step by step thing that mostly the human has to do.
GERRY48:18It doesn't have to be that way any longer. We can now do that in five minutes. But it used to maybe take five hours and we can validate recommendations. We can review and say, actually there's there's some really big wins we could make here to the structure of how things are working for you and find uncover things that are meaningful sooner rather than later.
GERRY48:37So not just in the data, but in how the way you capture the data is stored. You know, the the example we were talking about with this 300 fields, we don't have to necessarily go through and read what each of those 300 fields are. We can get a first, best guess from the machine saying, hey, it looks like 150 or so of these probably aren't even relevant because they haven't been updated by anyone in months or years.
EDDIE49:00Yeah, that's kind of the cool thing. Like, you know, you could easily run a prompt to say, show me each field the last time it was updated, how many of the records we have are populated? I mean, these are the things that we do as humans, like we being Union Square Consulting anyway. It's like, yeah, okay, like we've got this, these 300 fields on here, but two, 200 of them haven't been updated in like five years.
EDDIE49:22And we have 40,000 records and they're only on like 2000 records. Well, like, it's pretty easy for me to imagine Claude like, doing that lightning fast. Right. And now we have like, zone like zoomed in really quickly to figure out what the context and instructions are for Claude to say, this is what you need to look for within this object or set of objects.
EDDIE49:40But we're running up on time, and I want to take us home, because I don't want to turn this into a two hour podcast. This is not the Joe Rogan Experience.
GERRY49:46I want to keep going.
EDDIE49:47I know you do. Don't worry. We got a couple more minutes. But I want to talk about hallucinations and MCP versus social queries. Right. So for anybody not familiar with social queries, I'm probably going to boss this explanation. It's basically just like a way to run a report in Salesforce via like a code command, like a SQL query.
EDDIE50:07But it's so cool. Instead of like building out a report with the UI and Salesforce. And so this is going to be a really, really reliable way of getting data because you're literally communicating with Salesforce and you're saying, give me these objects, give me these fields, filter based on this criteria in a very, very like direct and specific way, as opposed to like Claude, like tell me like my open deals.
EDDIE50:30So help me understand, like what's the difference between a social query and MCP? And why have you Jerry built our own analytics tool starting first and foremost with social queries.
GERRY50:41Yeah, absolutely. So yeah. So cool. Query like you said, is like a SQL query stands for sale. Salesforce Object Query Language or Salesforce Oriented Query Language. Probably should couple. Check me on that. It's Salesforce version of a very, very, very explicit statement of the information you want to extract. Like you said, there's no compromise. You're telling it the exact fields, you're telling it off of what table you're telling it, how to filter, that there's no room for interpretation unless you just wrote it wrong.
GERRY51:10And if you wrote it wrong, you're still going to get what you wrote. It's just not going to be what you expected.
EDDIE51:15Oh, yeah. And this is like something I'll just touch on because, like, I have to teach everybody we hire this if they don't have like any Salesforce background people assuming like the reports are wrong, I'm like, the report's not wrong. Your filter criteria is wrong, right? The reason you're not getting what you want to see is because you put the wrong filters in.
EDDIE51:30It's that simple as that.
GERRY51:32yeah. The the error. The problem is between the chair and the keyboard. So with the the MCP, the MCP is just a way for.
GERRY51:44And I need to double check this. You might want to don't hold me to this. Basically helping I'm going to stick with Claude here. It's helping Claude infer your ask and how to turn that into something like a social query. But it's that inference that you get in trouble. That is the example we talked about, probably at the top of this, where if you've got seven different fields that can hold a currency amount on your opportunities, if you're not very, very crystal clear with Claude, which field it is that it should be pulling to get that dollar amount.
GERRY52:21If you forget to mention, like in our use case where not every open opportunity is actually something we consider in pipeline, we don't use our stage heroes or our stage ones. You know, there's a way to infer if it's considered actually in pipeline or not. It doesn't know those things. So it's going to it's going to take what you said.
GERRY52:37It's going to use the MCP to do some translation, and then it's going to it's going to run a query that it thinks is correct. Whereas if I document the exact circle query and that includes things like filter on deals that have the specific fields with these specific values, there is no chance that it's ever going to infer or guess how it should pull the data the wrong way.
GERRY53:01So the example you were sharing, some of the the analytics that we run, we started with that. We started with just documenting the social queries. And it lives in a set of instructions. And Claude knows that should never guess how to pull the data. It uses that to pull the data. And if you're asking for something that those queries don't provide, it doesn't guess.
GERRY53:23It just tells you I don't have that available. I can't pull that information.
EDDIE53:28And so interesting. Oh, sorry.
GERRY53:31I was just gonna say we've done that to protect the idea that from one day to the next, if you ask Claude, tell me about my open pipeline. You could get subtle variations every time. Even not even every day, two minutes apart. You could in the same chat, you could ask it the same question twice, and you could get two different results.
GERRY53:48And when it comes to numbers and pulling data in that regard, there's there's no room for that kind of problem. Those subtle differences are okay if you're doing a call prep and you're pulling information at that sort of level, that sort of zoom out. But when we're talking about hard numbers, you can't allow for
EDDIE54:05Yeah, absolutely. I mean, I think there's a really big difference between, like prepping a board deck and prepping for a sales call where you were on the call three days ago. You remember most of it. You're filling in some gaps in your memory, and you know, the AI is telling you, like, here's all the information you need to know.
EDDIE54:23And then if you're doubting something like, let me go check that. Let me check that report. Let me go double check this out. I mean, that's something I've been doing with Sarah for years, is like provide all the hyperlinks that I can, like, reference like the key, like the original source. But it's also interesting. I'm excited to, like, get more time to experiment with the MCP to see, like, if and to what extent it hallucinates.
EDDIE54:43I asked Claude today as I was prepping for this podcast. Like how like risky is it for the MCP to hallucinate? It basically says there's zero hallucination risk because you're pulling the raw data itself. If you ask, show me all open opportunities. Claude calls the Salesforce API, gets the actual records and reports them the opportunity. Names, amounts, stages, close dates those come directly from Salesforce Cloud.
EDDIE55:05Cannot hallucinate what Salesforce returns. However, what it says is the low but non-zero hallucination risk is synthesis, calculation and interpretation. For example, aggregations. What's the total value of my open pipeline? Claude might miscalculate if there are many records, and it's doing math in context rather than using a dedicated query filtering logic. If you say show me stale deals, cloud interprets stale and applies it's own judgment of what that means.
GERRY55:34Truncated results. Concept of open. The concept of open doesn't mean the same for everybody. Yeah, there is a mechanical way that that's done standard, but we know that our our terminology of open is not the same as what is documented, let's say within Salesforce.
EDDIE55:49Yeah. And like tell me if I'm wrong here. But my thought on this is, is like, you're not going to get like what you see on these LinkedIn posts of like, we built this over the weekend and like we just replaced our entire XYZ team. That's an MVP and it's going to be far, far from perfect, right? And so my thought is, is if you tap into the MCP and you start spitting on all these reports, it's probably going to be pretty wrong.
EDDIE56:12There's going to be all these issues and so be it. That's fine. And then you update the context and the instructions until you get it right. You double check your your stuff and you keep running it again and again and again until it's good enough. And then depending on the use case, you train your team to validate the things that are important to validate.
GERRY56:30That's right. It's a highly capable, highly intelligent, but very green to your business employee
GERRY56:36that can work faster than any human can. So but but know that if you're if you're providing feedback and you're providing coaching and instruction, right. It's going to remember that it's not going to forget the next time. As long as you provide clear instructions, it's going to remember.
GERRY56:51Oh yeah. Open means stage two or greater. So there's there's a there's a lot of strength in that. But you have to remember that it is it's it's not that it's a toddler necessarily. It's just that it needs to understand things. It can't interpret what's living in your head.
EDDIE57:09Cool. Well, where do we go from here, Jerry? So I typically ask our guests, where can they find you? It's cherry. Where can people find you?
GERRY57:21So I live in southeastern Vermont. Now. You can email me, Jerry, at Union Square consulting calm. I'm on slack. I don't have as much of a presence on LinkedIn as I probably should in other places, but you know, we'll work to improve that.
EDDIE57:35But cool. I wanted to let you answer that, but I was like, kind of just queuing it up for myself. Like, if anybody's listen to this and you guys want some help with this stuff, we are not going to say that. Like, we know everything there is to know about AI, and I don't think anybody else does. I think we're all learning this together.
EDDIE57:50And I think the really important point here is like, if you want to build something over a weekend, like, don't call me please. Like we're not going to do that. Like we're trying not to work weekends. But if you want to build something and iterate on it continuously over a sustained period of time to get it to a point where you can have your sales reps spending more time in front of customers and prospects instead of spending time updating Salesforce or prepping for calls.
EDDIE58:15If you want to improve the way that marketing is targeting your audience and probably generate more meaningful opportunities with the right companies, if you want to improve the way that CS is looking at expansion opportunities, prepare for board meetings better, have better visibility, better forecasting, and you want to use AI and specifically the MCP with Salesforce to do that, reach out to us and Jerry and our team will work with you.
EDDIE58:41Like I said, over that extended period of time, to not only build that MVP over a weekend or preferably over a Monday to Wednesday, but then to iterate on it till we get it right for you and then to continue to iterate on it as your business continues to evolve, as the tools evolve, as the data evolves, and as the ability to get more and more out of that stack evolves.
EDDIE59:01So if you guys want to help, reach out to us and if not, check out our content on Union Square Consulting or a podcast. Go to Market Science, which you're listening to right now. Unless you happen to be watching this on YouTube and run with this, this is exciting. Like people can build some really cool stuff with this and I'm excited about it.
EDDIE59:23And if you guys want to build that alongside us, we'd love to help. Anything to add to.
GERRY59:28That, Jerry? You know, I just want to say you've done a fantastic job of keeping me on on point. If anybody wants to get really in the weeds and way more technical, I'm your guy for that as well. So we tried to keep this relevant and high level, but there's whole more. Instead of layers of technical side of this that I'm happy to talk shop about.
EDDIE59:48Cool, man. Thanks, Jerry.
GERRY59:50Yeah. Thanks, Abby. Talk to you soon.
EDDIE59:52Thanks for listening to the episode. If this resonated. Please give us a five star rating and a follow. It helps us reach more people, and you get our latest and greatest content without having to search for it. And if you're looking for hands on help and go to market strategy and or rev ops, please reach out to us.
EDDIE60:06We help our clients with everything from annual planning to improving processes and go to market, implementing systems to support those processes and go to market AI. We're always happy to offer a free consultation to help you identify the best opportunities to improve your go to market engine, with or without our help. You can find us at Union Square consulting.com and the info will be in our show notes.
