The short answer

Ebsta, a vendor selling CRM capture software, analysed 655,000 opportunities worth $48 billion with Pavilion. It found 44% of the contacts sellers interact with are never recorded in the CRM, and a quarter of those missing people are senior decision-makers. New business in that dataset took eight stakeholders on average. Read the number knowing who produced it, and note that no independent study has tested it. A quarter of what the record fails to capture is the part of the account that decides.

The usual complaint about CRM data is that the reports look messy. Messy reports are the cheapest symptom available.

What is actually missing from a typical CRM?

What the data saysFigureSource
Contacts sellers interact with that never reach the CRM44%Ebsta and Pavilion, 655,000 opportunities. Vendor-produced
Of those missing contacts, the share who are decision-makers26%Same dataset
CRM contacts carrying a stale job title or phone number17%Same dataset
Stakeholders engaged in an average new-business deal8Same dataset

Read the first and last rows together. If a typical deal involves eight people and something close to half of the participating contacts are never recorded, the record of that deal holds a bare majority of the humans who touched it.

That is an aggregate rate rather than a guarantee about any single opportunity. It is still the number your pipeline review is built on.

Why is my CRM out of date?

Because nobody is paid to type. Capture is a tax collected at the exact moment of lowest attention, which is the twenty minutes after a call when the next call is already starting.

The people best placed to record who was on the call have the least reason to. The people who need that record most are two levels away and will not discover the gap until the quarter closes.

Discipline is what gets blamed here. Design is what actually decided the outcome.

What breaks first?

The forecast, and it breaks quietly.

A forecast is an aggregation over records. If the record for a $200,000 opportunity lists two contacts, and the deal is in fact being decided by eight people of whom two have never been recorded and one changed jobs in March, then the confidence attached to that opportunity is an opinion wearing a percentage sign.

The 17% figure for stale titles and phone numbers matters here in a specific way. A wrong title is worse than a missing one. A missing contact is a known gap. A wrong title is a false input that the model, the routing rule and the account plan will all treat as fact.

What else does a stale record cost?

Four things, and none of them show up as a line item.

Handover. When a representative leaves, the deal knowledge that was never written down leaves with them. The successor restarts relationships that already existed.

Routing. Rules fire on fields. A contact nobody entered cannot be routed, scored, suppressed or sequenced, so it receives whatever treatment the absence of data implies. That is usually the wrong one.

Coverage. If a quarter of the missing people are decision-makers, then the specific thing your record is worst at holding is the part of the account that decides. Coverage gaps concentrate where they hurt.

Proof. In Gartner's survey of 227 chief sales officers, 31% named difficulty proving the return on AI tools as a top challenge for the year. You cannot prove a gain against a baseline you never measured. The same survey found that 72% reported low reinvestment of the time AI had saved them, which is what happens when nobody can show where it went.

Can automated capture fix this?

Probably. No independent study has tested it, so treat the size of the effect as unknown and the direction as likely.

The mechanism is straightforward. Contacts who appear in email threads and calendar invitations can be written to the record without a human deciding to type them. Titles can be re-checked on a schedule rather than when someone notices. The work is mechanical, which is precisely the profile of work that automates well.

What is missing is the measurement. No independent study I can find quantifies how much automated capture improves data completeness, or what that improvement is worth in closed revenue. Vendors publish their own figures. Vendor-commissioned economic impact studies are marketing documents with a methodology section attached.

So the grade is promising. Build it, and measure the completeness yourself.

ClaimGradeWhat is actually known
Records are missing a large share of deal participantsDocumented44% of contacts absent, 26% of those decision-makers, on 655,000 opportunities
Missing and stale records degrade forecastingStrongly impliedThe arithmetic is direct, the effect size is unmeasured
Automated capture improves completenessPromisingNo independent effect study
Automated capture increases revenueUnprovenNo independent effect study

What the fix costs

The tooling is not incidental. Vendr's executed-contract data puts Gong at a median of $54,950 per year, with a range from $11,218 to $204,025. LeanData runs $22,908 and Chili Piper $13,500. Microsoft 365 Copilot lists at $30 per user per month.

Against those numbers, "our data will be cleaner" is a hope with an invoice attached.

How to measure this before you buy anything

Take twenty opportunities that closed in the last two quarters, won and lost. For each, open the email threads and the calendar entries and count the distinct people at the customer who participated. Then count how many of those people exist in the CRM record for that deal.

The ratio is your completeness rate. It takes an afternoon and it is the only number in this article that describes your company rather than someone else's.

Then check the titles. Pick fifty contacts on live opportunities and verify each against a current public source. If your stale rate lands anywhere near the 17% in the benchmark data, every model and routing rule downstream is running on inputs that are wrong one time in six.

Run the count first. Buy the tool second, if the count justifies it.

See how Atrium builds this →