Perfect CRM data, and it is not close. An AI sales assistant is only as good as the records it reads. Clean and score your accounts first, then let AI write, and only on the records you trust.
I'm Michael Thomas Zajac, a sales operator in Hermosa Beach. I have carried the number and built the systems under the reps, and I would take clean data over a smarter writer every single time.
What did 222 sellers just say about their CRM data?
On September 9, 2026, Firmable published a survey of 222 B2B sales professionals, from frontline reps to senior leaders. PPC Land broke down the numbers, and they are worth a slow read:
- Sellers estimate that 32% of the contact and account data in their CRM is inaccurate, incomplete or outdated.
- 75% would choose perfect prospect data over the best AI sales assistant on the market.
- Only 23% use AI to score or prioritize leads. The same 23% use it to clean CRM data. The most common use is drafting outreach, at 45%.
Firmable ran a companion survey of 220 sellers the same day. Only 14% said an AI agent could act on their current sales data without cleanup first. Just 16% would let an agent take a sales action without reviewing it. And 58% had discovered bad data only after contacting a prospect.
To be fair to the skeptics: PPC Land points out these are sellers' own estimates, not audited records, and Firmable sells prospect data. Both true. I still believe the direction, because I have lived it.
Source: Firmable surveys of 222 and 220 B2B sales professionals, published September 9, 2026, via PPC Land.
Why would sellers pick data over the best AI assistant?
Because AI does not fix a bad record. It scales it. Point a great assistant at a CRM where a third of the records are wrong and it writes beautiful emails to people who left, to titles that changed, to accounts that were never a fit. Faster than any rep could. With your name on every one.
PPC Land put it plainly: "AI, in other words, is used about twice as often to write messages to a database its users consider unreliable as to repair it."
Reps already know this. They are the ones who find out on the call. That 58% is not a data stat. It is a trust stat.
What was actually broken?
Not the writing. The writing was never the hard part. The hard part is deciding who deserves the writing in the first place.
Most teams spread effort evenly across the whole database, so the best accounts get the same thin attention as the junk, and nobody can afford to keep every record clean. Cleanup turns into a boil-the-ocean project that never ships. Then an AI tool shows up and gets pointed at all of it.
What did I build instead?
I was the first commercial hire at SuperSend, and I built the GTM engine under the reps. Every account got an ICP score from 1 to 5 before anyone spent a dollar or an hour on it.
- Humans worked only the 4s and 5s. That is where the relationships and the brand risk live.
- Clay enrichment ran only on high-fit accounts. The 1s and 2s did not burn credits. The records we worked hardest were the records we paid to get right.
- 3s waited for a signal. A hand-raise or a real trigger first, then automation.
Here is the unlock: you do not need perfect CRM data everywhere. You need perfect data where your people and your AI are going to spend. Score first, and cleanup shrinks from the whole database to the top of the list. I walked through the full score gate, and what it produced, in Should an outbound agent work your pipeline for weeks without a score gate?
Why does speed make bad data worse?
At Intertrust, where I ran sales ops and BD for 17 salespeople across four continents, we took speed-to-lead from 300 minutes to under 5.
I love that number. And it taught me that speed is a multiplier. It multiplies whatever is sitting in the record. Fast on a clean, scored lead is pipeline. Fast on a wrong record is just a faster mistake. AI is the biggest speed multiplier sales has ever had, which is exactly why the record underneath it matters more now, not less.
Where should AI go first?
Where the survey says almost nobody puts it: scoring and cleaning. Point AI at the unglamorous work before you let it write. Catch job changes. Merge duplicates. Flag missing titles. Score fit.
Then let it draft, only on clean, scored records, with a human approving anything that carries your brand. Only 16% of sellers would let an agent act without review. Honestly, that instinct is right until your data earns more trust.
The order of operations: clean, score, gate, then let AI write.
What should you do Monday?
Here is the Monday checklist:
- Pull your top 100 accounts by fit. Not by who is loudest in the pipeline. If you have no fit score, building one is step one.
- Audit 20 of those records by hand. Right person, right title, still at the company, working email. Count the misses. Now you have your own number instead of a survey's.
- Fix the top tier first. Clean and enrich the 4s and 5s. Do not spend enrichment credits on the bottom of the list.
- Point AI at cleanup and scoring before writing. Job changes, duplicates, missing fields.
- Gate the writing. AI drafts only on scored, cleaned records. A human approves anything customer facing.
- Re-audit in two weeks. Records decay. A job change turns an accurate record into an outdated one overnight.
Take the perfect data. Then let the AI earn its seat.