subinternal Talk to us

ResearchSales, forecasting

Why sales forecasts miss

Only 7% of sales teams reach 90% forecast accuracy. The data under a forecast lives in systems that were never joined, and AI on top of it learns the gaps faster.

An engraving of a quiet weather office: two barometers and a recording barograph drawing its line on a long bench, rolled charts, a telegraph key, a chart table with dividers, and through tall windows a rooftop wind vane and anemometer.

How often does a sales forecast land?

Rarely close. Gartner finds only 7% of sales teams reach 90% forecast accuracy or better, the median team sits between 70 and 79%, and 69% of sales operations leaders say forecasting is harder than it was three years ago.1

Vendor surveys say the same in their own words. In Xactly’s survey of 405 sales and finance leaders, four in five had missed a quarterly forecast in the past year, and 52% of sales leaders said their forecasts were off by 10% or more.2 Confidence was already low in 2020, when Gartner found 45% of sales leaders had high confidence in their forecast.3

FigureWhat it measuresWho was asked
7%of sales teams reach 90% forecast accuracy or betterGartner1
27%of a company’s applications are connected, on averageMuleSoft, 1,050 IT leaders4
40%of a seller’s week goes to sellingSalesforce, 4,050 sales professionals5
51%of sales leaders using AI say disconnected systems slow itSalesforce, 4,050 sales professionals5
60%of sales and finance leaders are unsure where pipeline data comes fromXactly, 405 leaders (vendor research)2
76%of CRM users say less than half their CRM data is accurate and completeValidity, 602 CRM users (vendor research)7

Why does the data disagree with itself?

Because a forecast is assembled from systems that were never joined. The average organization runs 957 applications, and 27% of them are connected.4 The deal sits in the CRM, the contract in billing, the warning signs in support tickets and product usage, and the forecast in a spreadsheet that copies from all four.

Sales leaders feel it. In Salesforce’s 2026 State of Sales, 51% of sales leaders using AI say disconnected systems slow their AI work.5 Clari Labs, which sells forecasting software, found 55% of enterprise revenue leaders see conflicting pipeline signals from disconnected sources.6 And 60% of the sales and finance leaders Xactly asked were unsure where their pipeline data comes from.2

Isn’t the CRM the source of truth?

Only as true as what people type into it. In Validity’s survey of 602 CRM users, 76% said less than half their CRM data is accurate and complete.7 Sellers spend 40% of their week selling; the rest goes to everything else, data entry included.5

Some activity never reaches the CRM at all. In older Salesforce setups, emails captured by Einstein Activity Capture are not CRM records and do not appear in standard reports.8 “Activity capture is on” does not mean a forecast can read the activity. It shows in results: in a Gartner survey, 84% of sales leaders said analytics influenced performance less than expected, with poor data quality among the top barriers.9

Does a manager’s call make it better?

Sometimes, and not in the direction people expect. The standard study of judgmental adjustments, over 60,000 forecasts at four supply chain companies, found large adjustments often helped, small ones often hurt, and upward ones were less accurate, with a general bias toward optimism.10 That was demand forecasting, not a sales pipeline, but the pull is the same one a commit call feels.

On a sales pipeline itself, a study of 25,578 opportunities at a global B2B consulting firm found a machine learning model beat the win probabilities salespeople entered in the CRM.11 And 60% of chief sales officers told Gartner their revenue number is largely driven by things outside their control.12

What does AI change, and what doesn’t it?

AI reads more of the signal than a person can. Salesforce’s deal health scoring, for one, already joins opportunity scores with email, call recordings and service cases.13 When the systems are joined, that is the point: support and usage show a deal in trouble before the stage field does.

When they are not, AI makes the gap faster. Gartner put it plainly: “If those systems are fragmented, the agents will scale the fragmentation.”12 Gartner also predicts organizations will abandon 60% of AI projects that lack AI-ready data.14 A forecast model trained on half-right CRM data learns the half-right part well.

What does a forecast you can trust need?

Our read, from the work we do: four things, in this order.

One customer across systems. CRM, billing, support and product usage keyed to the same account, so a renewal at risk in support is a renewal at risk in the forecast.

Numbers defined once. Pipeline, bookings and ARR computed by code from one definition, not rebuilt in each spreadsheet. A semantic layer does this: change a definition and it changes everywhere it is used.15

Judgment on the record. Every override kept beside the model’s number, so the quarter shows whose call was closer.

Scored when the quarter closes. A forecast is one of the few things in business with a right answer that arrives on a date. Score every forecast against it and keep only the changes that score better. That is how our learning harness improves: it keeps a change only when a check confirms it.

References

  1. Gartner, “The Role of Artificial Intelligence (AI) in Sales.” Topic page, read November 2026. gartner.com
  2. Xactly, “New Sales Forecasting Research from Xactly Uncovers Need for Cross-Team Collaboration for Accurate Forecasting.” July 2024, via Accesswire. finance.yahoo.com
  3. CIO Dive, “Sales leaders grapple with inaccurate forecasting. Data quality might be the culprit.” February 2020, reporting Gartner’s State of Sales Operations survey. ciodive.com
  4. MuleSoft (Salesforce), “2026 Connectivity Benchmark Report Insights.” February 2026. blogs.mulesoft.com
  5. Salesforce, “Salesforce Announces State of Sales Report for 2026.” February 2026. salesforce.com
  6. Clari Labs, “New Clari Labs Research Reveals 87% of Enterprises Missed Revenue Targets in 2025 Despite Record AI Investment.” January 2026. salesloft.com
  7. Validity, “The State of CRM Data Management in 2025.” validity.com
  8. Salesforce Help, “Using Emails Without Sync Email as Salesforce Activity in Einstein Activity Capture.” Read November 2026. help.salesforce.com
  9. Gartner, “Gartner Survey Finds Sales Analytics Has Less Influence on Sales Performance Than What Leadership Expected.” February 2024. gartner.com
  10. Fildes, Goodwin, Lawrence and Nikolopoulos, “Effective forecasting and judgmental adjustments.” International Journal of Forecasting 25(1), 2009. econpapers.repec.org
  11. Rezazadeh, “A Generalized Flow for B2B Sales Predictive Modeling: An Azure Machine-Learning Approach.” Forecasting 2(3), 2020. mdpi.com
  12. Gartner, “Gartner Predicts AI Agents Will Outnumber Sellers 10 to 1 by 2028, Yet Fewer Than 40% of Sellers Will Say Agents Improved Productivity.” July 2026. gartner.com
  13. Salesforce Help, “AI Solutions for Sales Pipeline Visibility and Forecasting.” Read November 2026. help.salesforce.com
  14. Gartner, “Lack of AI-Ready Data Puts AI Projects at Risk.” February 2025. gartner.com
  15. dbt Labs, “dbt Semantic Layer.” Read November 2026. docs.getdbt.com