Your CRM is probably a mess.
But it's been that way for a while, and it never caused much trouble because a person was always around to catch the weird stuff before it mattered.
Now AI is doing a lot of that work instead, and it doesn't catch the weird stuff.
That's why forecasts are coming back wrong, leads are landing with the wrong reps, and renewal emails are going out that make no sense.
So what does it take to fix it?
Let’s find out.
Your CRM, your forecasts, and every AI tool you're running right now all sit on top of your data model.
Most of the time, data models are invisible. And that’s a problem.
For example, most people look at a forecast and take the number at face value. They don't think about what's behind it. But if there are problems in the data feeding that number, AI won't notice them. It just uses what's there and moves on.
That’s not good…here’s why: if a person looked at your pipeline and noticed closed-won meant three different things depending on who logged the deal, they'd probably pause and ask about it before bringing a number to leadership.
AI doesn't do that.
It looks at whatever data exists, picks a pattern, and reports it back like there's nothing to question.
A few things cause this in almost every CRM:
AI is only as good as the data model underneath it. And bad data isn't a small problem as poor data quality costs organizations an average of $12.9 million annually, yet many teams are feeding that same data directly into forecasting models, enrichment tools, and AI workflows. Fix the data model, and forecasts improve without much extra work. Leave it messy, and every new tool just finds another way to be confidently wrong.
We covered what a data model is in general. A B2B data model is the same idea, just specific to how businesses sell to other businesses. It covers what objects exist in your CRM, like contacts, companies, deals, products, and subscriptions, how those objects relate to each other, and how data moves between systems across your GTM stack.
A well-built B2B data model reflects how the business actually makes money.
Want to see what a modern revenue data model actually looks like? Check out our RPX class on the Revenue Performance Model! 👇
But most don't. They reflect how the CRM happened to get set up a few years ago, and nobody's gone back to fix it since.
Most B2B data models were built for a simpler version of selling (e.g. one pipeline, one product, a single close date). But that setup doesn't hold up for companies running expansion revenue, product-led motions, multi-stakeholder deals, or some mix of all three, which is most B2B companies at this point.
A modern B2B data model needs a few core pieces in place:
So those are the core pieces. But having the right objects in your CRM isn't actually where most B2B data models fall apart. The bigger issue is how those objects connect to each other.
You can have perfectly clean contact records, and if those contacts are linked to the wrong accounts, your data model is still broken. Clean data in the wrong structure is still broken data.
This is where AI tools start to struggle.
To do anything useful, like scoring a lead or forecasting a deal, AI has to follow the connections between these objects. If those connections are missing or inconsistent, AI doesn't know which one is correct. It still picks one and moves forward, so the answer you get depends on whichever connection it happened to follow, not necessarily the right one.
This also looks different depending on how a business sells. A company doing one-time deals can usually get by with a simpler setup, since most of the important work happens before the deal closes. A company running on recurring revenue doesn't have that luxury. It needs to track what happens after the close just as closely as what happens before it, since renewals, expansions, and churn all depend on that same structure holding up over time.
Most companies aren't running one system. They've got a CRM, an email tool, a billing system, and probably something for customer success too. Each one got added at a different time to fix a different problem, and nobody really asked if these tools would agree with each other later.
Each tool works fine on its own, but when you put them together, they tell different stories. The CRM has its own idea of what a customer is. Billing has another. The result is data silos that create competing versions of the truth across the business. In fact, employees can spend as much as 12 hours per week dealing with data silos instead of acting on insights, fixing a problem that only gets worse when AI is pulling information from multiple disconnected systems.
So when you hear "one source of truth," that’s a decision somebody actually has to make. One system holds the real number. Everything else pulls from that system.
Here's where this breaks down the most:
This didn't used to matter much. Someone would notice the numbers were off, ask around, and fix it.
But AI doesn't do that.
If a tool pulls from your CRM and your billing system and gets two different revenue numbers, it won't stop to check which one is right. It just picks one. Sometimes it splits the difference, which sounds careful but is still wrong.
Disconnected systems used to be something ops teams cleaned up. But now they're feeding straight into decisions AI makes, often with nobody checking the work. If your revenue data doesn't have one clear home, you're not giving AI anything solid to work with. So don't be surprised when it guesses.
Everything we've looked at so far gets even more important once you're running a recurring revenue model.
The data behind this kind of model has to stay accurate for a lot longer than it would for a one-time sale, because forecasting, revenue recognition, and retention all depend on those same records staying correct well past the day the deal closed.
With a transactional deal, the data mostly matters up until the close. Once the deal is done, it's done. But with recurring revenue, the deal isn't done. It keeps going, and so does the need for the data behind it to stay accurate.
A few things make recurring revenue data harder to manage:
Most CRMs aren't set up to capture this well. The fields that matter most here are often missing, inconsistent, or just not configured properly:
And this can get worse with AI involved.
A small mistake in how renewal dates are logged doesn't stay small. If an AI tool is forecasting ARR six months or a year out, that small error compounds the further out the forecast goes. What started as one wrong date turns into a forecast that's significantly off.
Net revenue retention is a good example of how this plays out. It's one of the most important numbers in a recurring revenue business, and one of the hardest to calculate cleanly, because it needs several different pieces of data to all be accurate and connected at the same time. When AI is asked to calculate or predict NRR using a CRM that isn't built for that, the output still looks precise. It just isn't correct.
But the same CRM gap barely matters for a company selling one-time deals. It's a minor annoyance. For a recurring revenue business, that same gap means churn can't be properly attributed, retention numbers can't be trusted, and any forecast built on top of that data is wrong from the start.
Everything so far has been about AI giving you wrong information, a bad forecast, a confusing number, or an answer that looks right but isn't. That's frustrating, but someone usually catches it before real damage is done.
But AI agents are a different story, because agents don't just answer questions. They take action.
An AI assistant looks something up and tells you what it found. But an AI agent actually does something with that information. It enrolls a contact in a sequence, routes a lead to a rep, updates a record, or fires off a playbook on its own. A wrong answer from an assistant is annoying. A wrong action from an agent, built on bad data, is a real risk to your revenue and your relationships with customers.
Here's the kind of failure mode that shows up when agents are working from a bad data model:
None of these are hypothetical in the sense of being far-fetched. They're the kind of thing that happens and usually gets noticed by a customer before it gets noticed internally.
This also gets worse with scale, not better. One bad field in a smaller company is something a rep notices and fixes. In an agent-driven motion running across hundreds or thousands of accounts, that same bad field can trigger the same wrong action across every account that field touches, all at once.
None of this means companies should slow down on adopting AI agents. But it does mean there's a real list of things that need to be true about your data before an agent gets to act on its own:
Get those right, and agents become a real advantage. Skip that work, and you're handing action authority to a system working off bad information, and your customers are usually the ones who notice first.
Thinking about AI agents before fixing your data model? Check out or RPX class on AI Readiness for GTM Teams! 👇
Data model work isn't a project you finish and check off, it's something you maintain. Most teams treat it like a one-time migration, clean it up, move on, forget about it. And that's exactly why the same problems keep coming back every year or so.
A clean B2B data model really comes down to four things.
Although RevOps teams have been pointing out these exact problems for years, AI is now making the consequences visible to everyone.
A reasonable starting point is to pick one of these four and audit it this quarter. Most teams find the field definitions are the fastest place to get a quick, visible win, since it's mostly a documentation and training exercise rather than a system overhaul.
Not sure where your data model is breaking down? This RPX class walks through the Revenue Engine Diagnostic framework used to audit systems, map revenue flows, and uncover the gaps that create reporting and AI issues. 👇
A data model is the structure that defines what information a business captures, how it's organized, and how different pieces of data relate to each other. In a GTM context, a data model determines what objects live in your CRM, how those objects connect, and how data flows between your systems. It's the structural foundation everything else runs on.
A B2B data model is a data model built specifically for business-to-business commercial operations. It defines how contacts, companies, deals, products, and subscriptions are structured and related in your CRM and GTM stack. A well-built B2B data model reflects how the business actually generates and retains revenue, including the full lifecycle of a customer relationship, not just the initial sale.
A B2B revenue model describes how a business makes money from other businesses. Common B2B revenue models include one-time transactional sales, recurring subscription revenue, usage-based pricing, and hybrid combinations. Your data model needs to be built to match your revenue model. A business running a recurring revenue model needs a CRM structure that captures subscription data, renewal logic, and expansion triggers, not just closed deals.
A recurring revenue model is a business structure where customers pay on an ongoing basis rather than through one-time purchases. SaaS, managed services, and subscription businesses all run on recurring revenue models. These models require more data precision than transactional models because revenue recognition, forecasting, and retention metrics all depend on the same underlying records being accurate and up to date over time.
AI tools pattern-match on the data they're given. When that data is inconsistent, incomplete, or structured differently across systems, AI produces outputs that look confident but reflect the errors in the underlying data. The more AI is used in a GTM motion, especially AI agents that take autonomous action, the more visible and consequential those data problems become.