A contracting system said one number. Product telemetry said another. Both numbers were real, and for years, people managed the difference in their heads.
A client signed a contract for roughly a thousand locations. That number lived in forecasting, in revenue planning, and in every calculation of locations under contract, including those that made it into a board deck. It was the number that made the account look like one of the largest in the portfolio.
Only about eight hundred of those locations ever went live. Finance only billed and recognized revenue against what was actually live, so that number lived somewhere else: Billing records, product telemetry, the systems that track what a customer is actually using, not the number in the contract.
The difference between the two numbers wasn’t a mystery. Some of it was timing: A location under contract that hadn’t launched yet, and eventually would. Some of it was permanent: A client had decided, deliberately, never to launch part of what the contract covered. Both were real. Neither was going away on its own.
People saw the gap. Customer-facing teams knew, more by instinct than by report, that the account was smaller in practice than the number in the contract. Finance had both figures too, the contracted total and what was actually being billed, but didn’t always trace why the two didn’t match. On the largest accounts, the gap was obvious enough that everyone accounted for it. On the long tail of smaller accounts, it accumulated quietly, unnoticed in aggregate even when it was visible one account at a time.
None of this broke the business. People managed through it, mostly by repeating the same conversation. Someone would explain, again, why an account rated for a thousand units was only generating revenue on eight hundred, and the room would nod and move on until the next person needed the same explanation.
Add AI to that same situation, and the compensating mechanism disappears. The systems don’t intuit anything. They act on whatever data is available to them, and that data isn’t always consistent with what the system next to it has.
Two Numbers, One Account
A thousand-unit account and an eight-hundred-unit account are not the same strategic problem. They carry different revenue risk, different renewal conversations, and a different amount of work required to retain and grow what the customer is actually using, not what the contract lists.
The contract said a thousand. The product said eight hundred. Both were technically true, and neither one was useful on its own.
This is what data integrity actually means in practice. Not only whether your systems are accurate. Whether they agree.
It also determines how the account gets resourced. A CS team staffing against a thousand-unit account looks different from one staffing against eight hundred. A forecast built on contracted scope tells a different growth story than one built on live usage. Whoever picks which number to use, even by accident, is making a strategic call about the account without realizing they’re making one at all.
What AI Does With Two Different Truths
People managed the old version of this gap through memory and repetition. Someone had to notice the mismatch, and someone had to explain it, every time it came up.
What changes once AI enters the picture is confidence and the time it takes to get to the wrong answer. An AI model doesn’t have the instinct a customer-facing rep builds up over a hundred renewal conversations. It doesn’t ask which number is real. It picks a source, and it states a result with total confidence, whether that source was the right one or not.
An account health score, an expansion propensity score, a churn-risk flag: Each one looks precise on the dashboard. Each one was built on whichever data source happened to feed it, contracted scope or live usage, and the two people looking at that score usually have no idea they’re comparing answers built on different foundations.
The same pattern shows up in forecasting and territory planning. A pipeline model scoring whitespace off contracted scope will recommend expansion motions the customer isn’t remotely ready to pursue. The same account, scored off live usage, might not show up as an expansion opportunity at all. Both models will be confident. Only one of them reflects what the customer is actually doing.
This is exactly the kind of gap our AI Revenue Risk Assessment is built to surface, before a strategic account gets managed off the wrong number: see how it works →
What Actually Fixes This
The fix isn’t a new dashboard, and it isn’t a smarter model that quietly reconciles the disagreement for you. Both of those just add a third opinion to a room that already can’t agree.
What fixes it is deciding, in advance, which system is the source of truth for which fact about an account, and requiring that anyone using that fact, human or AI, checks against it first. Contracted scope can live in forecasting. Live scope can live in billing and product. Both are allowed to exist, as long as everyone downstream knows which figure applies and why.
One rule that actually works: No AI-generated score gets treated as an account’s real size until it’s checked against whichever system your company has already agreed owns that number. That agreement doesn’t require new technology. It requires someone with the authority to say, out loud, which number wins when the two disagree, and then requires every tool touching that account to respect the answer.
Where This Leaves the ROI Question
Every post in this series has pointed at some version of the same blind spot. This one is the most basic version of it: You can’t prove the ROI of an account when you can’t even agree how big it is.
Pull your ten largest accounts. Compare contracted scope against live or delivered scope in whichever system tracks what customers are actually doing. Count how many of the ten match.
This is the fourth post in an eight-post series on why most B2B SaaS companies can’t answer the AI ROI question, and what closes the gap. If you want a scored, five-minute read on where your own organization stands, the AI Revenue Risk Assessment gives you the diagnostic: take the assessment →
Andrea Mulligan is a B2B SaaS executive and advisor with 30 years of experience building Customer Success, Professional Services, and GTM organizations. She works with PE-backed and growth-stage companies on CS transformation, revenue retention strategy, and post-sale model design. Start a conversation →