94% of Spreadsheets Have Errors. Sales Teams Still Forecast On Them.

Only 7% of sales teams hit 90%-plus forecast accuracy, and 30% of organizations still run their primary budgeting process on spreadsheets — unchanged since 2014 — even though audits find errors in 94% of operational spreadsheets, according to a new Outreach guide aimed at RevOps leaders. The guide’s sharper finding is about who’s actually driving the forecast: across an analysis of more than 60,000 forecasts, over 90% were manually adjusted by people, and upward adjustments often moved in the wrong direction. Separately, 59% of companies don’t document loss reasons and 68% lack full touchpoint histories. Outreach’s fix is a five-step framework: standardize stage definitions around observable exit criteria, baseline current accuracy while running shadow AI models in parallel, design bounded override controls with required reason codes, maintain full forecast auditability and versioning, and restructure pipeline calls around exception management instead of number collection. It closes with a CRO quote describing accuracy improving from off by 10-20% to within 5% as board-level transformational.

The guide operationalizes a model-choice framework Outreach published a month earlier. That piece argued the real decision isn’t picking one forecasting method but choosing the right category for the org — statistical for high-volume, low-variance pipelines with mature history; AI-native where signals are weak and markets move fast; hybrid, which it called “the default answer for most revenue orgs.” It cited customer Omniplex Learning landing within 5% forecast accuracy, down from 10-20% variance, and noted Outreach customers collectively close over 2 million opportunities a month on the platform.

The through-line across both pieces is a governance problem more than a modeling one: the technology to forecast accurately already exists, but more than 90% of AI-assisted forecasts still get manually overridden by humans — meaning the bottleneck to hitting that 7% accuracy ceiling isn’t the model, it’s whether an org has built the auditability and override discipline to trust it.