How revenue intelligence is moving into the real world

I've spent the last ten years building go-to-market engines in B2B SaaS. At Moss, a European spend management solution, I built the outbound engine as we scaled from $20 million to $50 million in ARR. I then ran pipeline generation at Pendula before we were acquired, and now I lead pipeline generation and revenue intelligence at GoodFit.

And in all that time, across every company I've worked at or consulted with, one thing has stood out to me consistently. We spend enormous amounts of money and time building the best execution engines we possibly can. The best salespeople. The best tools. CRMs, sales engagement platforms, power dialers, call recorders, meeting schedulers, CPQ tools, forecasting software. The list goes on.

But we spend a fraction of that on the intelligence that actually guides all of it.

The intelligence-execution mismatch

When I talk about intelligence, I mean two specific questions. What accounts do we target? And how do we allocate resources against them? Meaning, what accounts do we work, in what order, and across what channels?

At GoodFit, we recently went through a rebrand with an ocean theme, so bear with me on this analogy. Our revenue intelligence is like a small tugboat pulling a massive execution engine. The tugboat is so much smaller than what it's guiding, but it determines the direction of everything. That's the intelligence-execution mismatch. The best execution is wasted if the intelligence guiding it isn't up to the job.

I've seen this play out at every company I've worked with. Exceptional salespeople, rigorous training, top-tier tools. But working from rudimentary intelligence. And to understand how we got here, you need to look at where that intelligence data came from in the first place.

Phase one: three data points and a lot of guesswork

When I started out in B2B SaaS, we had access to basically three data points. Employee count, country (typically where the company's HQ was located), and industry, which was, frankly, unreliable. Industry fields are often miscategorized, and they rarely match the internal definitions companies actually use when deciding who to sell to.

If you were lucky, you had all three of these fields populated in Salesforce for every account. Often, the data was patchy and unstandardized. Some accounts had all three, some had none.

So what could you do with that? You'd try to approximate your ICP using those filters. At GoodFit, our true ICP is B2B SaaS companies in EMEA and North America with an established outbound sales team. In phase one, the best we could do to approximate that was to pull a list of businesses in EMEA and North America with 50 to 1,000 employees, and maybe layer on an industry filter, though B2B SaaS is notoriously hard to capture through legacy industry fields.

Think about those filters for a second. Businesses in EMEA and North America with 50 to 1,000 employees. That's the same filter a company selling HR software would use. Or finance software. Or sales software like ours. It's such a broad approximation of who we actually sell to that it almost doesn't mean anything.

The game at that point was to find the data provider with the biggest list of accounts, filter it down into something that broadly matched your ICP, and then have your reps manually run through accounts to qualify them.

We'd track something called disqualification rate, which measured how many accounts a rep was assigned before they'd even reached out to one, because they'd done desktop research and found it didn't match what we were actually looking for. That number was telling. And a lot of companies are still operating this way today.

Phase two: more data, better approximations

Over time, the breadth and depth of data available for revenue intelligence expanded considerably. More publicly available information on companies meant more detail about how they actually operate.

We moved from three fields to something much richer. You could now see the size of specific functions within a company, not just overall headcount. You could see granular web traffic data, funding round information, technographics, and more.

This made a real difference. At GoodFit, we could now use industry subcategory fields to get much closer to a list of true B2B SaaS companies. We could approximate an established outbound sales team by looking at actual sales team size rather than overall company size. The variance in our account lists went down. Disqualification rates dropped. We were getting closer to reality.

Territory planning got more sophisticated too. With more fields, you could segment your market more granularly and build rudimentary grading systems. For example, if our data analysis showed that companies with $20 million or more in funding and a sales team of 20 or more reps and two or more RevOps people had twice the ACV of our average account, we could start to prioritize accordingly. It wasn't perfect, but it was directionally accurate.

But here's where we hit what I'd call the database fallacy.

The database fallacy

No matter how granular your data gets, databases and tables can only provide pre-configured data points. You'll always have to filter those data points to create lists of accounts. That's just how databases and tables work.

And filters are hard-coded rules that apply to every account. So you end up making rules like: if a company has five or more sales reps, it has an established outbound sales team. But in reality, there's no fixed rule that tells you whether a company has an established outbound motion. The answer is hidden in their team structure, their tech stack, their job listings, their website. And critically, every company needs to be evaluated on its own evidence, case by case.

When you use a rule-based database to answer a complex qualitative question, you get variance. False positives and false negatives are created by edge cases. And when you're making decisions that define the growth of your business, like your business plan, your territory plan, your ICP definition, your account prioritization, you want to minimize variance as much as possible.

Databases and tables, by their design, create a proxy layer between you and reality. And that proxy layer has real costs.

Moving beyond the database

Here's where things get genuinely interesting. We now have the capability to bypass that proxy layer entirely and go straight into the deep data sources that have always existed underneath it. By combining agentic systems, reasoning models, and those data sources, we can make case-by-case judgments about accounts at scale.

Instead of trying to find the data points that help us approximate an answer, like how many salespeople does this company have, we can ask the question directly. Does this company have an established outbound sales team? And get a reasoned answer, with evidence, for every account in our market.

Let me give you a concrete example. On the filter-based approach, our rule for an established outbound sales team is: sales team size of five or more. Simple. Applied to every account the same way.

With case-by-case intel, we're looking at the actual breakdown of roles within the sales team. SDR roles, AE roles, how titles are categorized, what the tech stack suggests about their sales motion, whether the company mentions outbound sales explicitly anywhere.

For one account, that analysis might give us very high confidence that yes, they have an established outbound team. For another account that also passed the five-plus filter, we might find there's no sales leader, two of the salespeople don't carry quota, and the company is still early stage and bootstrapped. Very low confidence.

Two accounts that both passed our filter. Vastly different in reality.

Here's another example from my time at Moss. Moss is a spend management platform, so the more a company spends, the more valuable they are as a customer. We wanted to understand how much a company spends, and ideally how that spend breaks down across areas like travel, subscriptions, and marketing.

To approximate marketing spend using filters, we'd look at the percentage of web traffic that's paid, and the size of the marketing team. Reasonable proxies. But the actual picture of marketing spend is spread across a whole patchwork of data sources.

How many ads does this company have live in their Meta ad library? What's their paid web traffic, and how does that compare to their total traffic volume? What types of marketing channels are they running, based on job titles within the team? If they have an events function and a paid media function, they're likely spending more. How many events did they actually sponsor or run in the past year? Does their team structure suggest they're using a paid media agency?

No rule-based system gets you to that level of understanding. But when you bypass the database and reason across those data sources directly, you get something that's genuinely closer to the truth.

Why LLMs alone aren't the answer

If you're thinking that this sounds a lot like what you can do with ChatGPT or Claude, you're not wrong that LLMs are a key part of the picture. The way we interact with LLMs is actually a good model for how revenue intelligence should work. You don't ask ChatGPT to find you accounts with five-plus salespeople. You ask it which of these accounts have an established outbound sales team.

But LLMs alone aren't sufficient for revenue intelligence, for three reasons.

First, LLMs don't have the data access you need. If you ask ChatGPT to find every company in EMEA with an established outbound sales team, it'll tell you it can't. It doesn't have access to a universe of accounts to work from, and it doesn't have access to key proprietary data sources like Crunchbase for funding information, SEMrush for web traffic, or historical job postings that may no longer be publicly available.

Second, LLMs don't have validation loops. To accurately reason across multiple data sources and draw reliable conclusions about accounts, you need first-party training data. Verified examples from your own experience that tell the model what good looks like.

Third, LLMs struggle to perform these reasoning tasks at scale in a cost-efficient way. Answering nuanced qualitative questions about thousands, or tens of thousands, or hundreds of thousands of accounts, at a low operating cost, isn't something an LLM can currently do on its own.

So the revenue intelligence category isn't going away. But selling pre-configured data via databases and tables is increasingly the approach of the past. When you're evaluating revenue intelligence vendors, the right question to ask is: can you provide accurate, case-by-case intel about every account in my market?

What you can actually do with this intel

Let's get practical. Once you have this kind of intelligence, what changes?

ICP definition and account capture

Instead of approximating your ICP through filters, you can ask the actual question of every account in your market. Your CRM starts to look like every account in it has been manually qualified by a rep. You know exactly how many workable accounts you have, which means better planning. And your reps aren't burning time qualifying accounts before they even reach out.

Territory planning

With case-by-case intel, you can segment your market based on specific use cases, pain points, and business models, not just geography and company size. Reps can specialize in the use cases they sell to, the business models they understand, the pain points they know how to address. That specialization compounds over time.

Account prioritization

This is where it gets particularly interesting. Take every account you've won and lost in the last 12 months. Categorize them with key pieces of intel that you believe correlate with account quality. Things like: does this company have a large addressable market? Does their ICP definition rely on niche data points that legacy providers struggle to surface? Do they spend heavily on outbound execution? Does the BDR function report into sales or marketing?

Then look at the difference in win rates between accounts that had those characteristics and accounts that didn't. This is similar to how you'd build a grading model, but instead of filtering on pre-configured data points, you're answering specific qualitative questions.

You can identify the factors that genuinely move win rates. Group accounts into clusters with meaningfully different win rates. And then make sure your team is working the accounts where you're most likely to win.

That's a significant win rate improvement from doing nothing more than working a better set of accounts. Before you've even changed your messaging or your outreach approach.

Sales messaging

And once you know specific things about an account, like whether their BDR team reports into sales or marketing, or whether they're expanding into new international markets, you can use that to tailor your outreach. Better messaging in your sequences, more relevant paid ads, more specific conversations. The intel feeds everything downstream.

Where this leaves us

Going back to that tugboat analogy. We're finally reaching a point where revenue intelligence can do justice to the execution engines we've spent so much building. The tugboat is getting a new motor.

The two key questions haven't changed. What accounts do we target? How do we allocate resources against them? But our ability to answer those questions accurately is changing fundamentally. We can now ensure that every account our team works is a genuinely qualified account.

We can ask the complex, nuanced questions that identify our highest-value prospects. And we can direct our execution engines with the kind of precision that was simply out of reach when all we had was employee count, country, and industry.

The shift from database filtering to case-by-case intelligence is happening now. The teams that move first will have a real advantage, both in the efficiency of their operations and in the quality of the accounts they're working. And in a market where everyone has access to the same execution tools, that intelligence layer is increasingly where the edge lives.