Clean data beats expensive AI

Hey Everyone,

An AI forecasting tool scores whatever your reps typed into the CRM. Validity surveyed 602 CRM users and admins in 2025, and 76% said less than half of their CRM data is accurate and complete. Feed a model records like that, and you get a precise-looking number built on missing inputs.

Today we're covering where AI forecasting helps and what to fix in your CRM.

What AI sales tools can and can't do

Gartner puts median forecast accuracy at 70–79%, and only 7% of sales organizations reach 90% or better. AI tools can raise those numbers by removing human bias, but they cannot fix poor data discipline.

What they can do

• Eliminate emotional bias: Reps forecast deals they hope will close. AI scores them strictly on historical facts.
• Standardize deal scoring: The model judges every opportunity by the exact same rules every single time.
• Track objective metrics: AI instantly analyzes deal age, rep activity, time in stage, and days since the last update.

What they can't do

• Fix dirty CRM data: The algorithm only knows what is inside the system. It cannot analyze missing close dates.
• Normalize inconsistent processes: If pipeline stages mean different things to different reps, the AI learns from bad records.
• Invent missing information: No algorithm can fill in fields or log loss reasons that your team never entered.

Your data sets the limit

Machine learning tools require high volume and deep history before they can outperform simpler forecasting methods. Artificial intelligence learns by identifying patterns across hundreds of historical deals—both won and lost—alongside the specific activity logs behind them.

A small sales team closing only a few dozen deals per quarter simply does not generate enough data history for an algorithm to beat a well-maintained, weighted pipeline. At this stage, your business will gain far more value from fixing manual logging habits than from purchasing predictive software.

5 operational fixes for accurate forecasting

Write down your stage definitions and enforce them: Every stage needs a one-line definition explaining exactly what must be true for a deal to sit there. Make key fields mandatory before an opportunity can advance. If a rep can move a deal forward without a close date or a named decision-maker, your pipeline data will remain unreliable.

Review forecasts against actuals every week: Compare what you projected against what actually closed, and document the root cause of every miss. The same errors will resurface continuously, pointing you directly toward the parts of your process that need fixing.

Log a reason for every lost deal: Select these reasons from a short, fixed dropdown list so the answers stay consistent and comparable. Without documented loss reasons, neither you nor an AI model can learn from the deals you lose.

Track win rates by segment: An enterprise deal in stage three closes at a completely different rate than an SMB deal in that same stage. Track them separately—in your spreadsheet today and in your forecasting models later.

Clean stale deals monthly: If an opportunity sits untouched for 30 days, force reps to either update it or close it out as lost with a reason. A pipeline cluttered with dead deals inflates every single forecast you run.

Carl Eschenbach, former partner at Sequoia, told founders that "it takes many quarters to gain credibility and only 90 days to lose it." He was talking about forecasts. Your board plans around the number you bring them, and so do your own hiring and spending decisions. A forecast that comes from an unmaintained pipeline risks your credibility every quarter.

Try this today

Open your CRM and pull every deal with no activity in the last 30 days. Add up their total value and divide it by your total pipeline. That percentage tells you exactly how much of your forecast depends on deals that nobody is actually working on.

Next, pull your 10 largest stale deals and have each owner update the stage, close date, and next step, or close the opportunity out as lost with a reason. Repeat this exercise monthly, and your forecast accuracy will improve long before you spend a single dollar on new tools

Go deeper

👉 Validity: The State of CRM Data Management in 2025 – the survey behind the 76% figure, and what drives poor data quality.

👉 Sequoia Capital: Leading in Uncertain Times (PDF) – Sequoia's guidance on forecast accuracy and credibility, including the Eschenbach line.

👉 Demand Gen Report: Transforming Sales Forecasting with AI – Gartner's accuracy benchmarks and where AI helps.

👉 Cognism: Half of C-suite CRM data goes stale within two years – research on how fast contact data decays, which is the root of much forecast error.

Coming up tomorrow

Tomorrow, we'll cover how the best capital allocators stress-test a request before they fund it.

P.S. On a scale of "spotless" to "we don't talk about the CRM", how clean is your pipeline right now? Let us know.