Blockbuster’s data showed profit from late fees — a metric that monetized customer resentment. Netflix optimized for retention — a metric that compounded customer love. Both companies had data. Only one had honest metrics.

The same divide now runs through AI. In MIT’s 2025 review of more than 300 enterprise GenAI initiatives, just 5% of integrated pilots were extracting millions in value while the rest stalled — and the winners shared a signature: they bought workflow-embedded tools through partnerships (success rate around 67%) instead of building in-house trophies (~33%).[1] New technology, old lesson: the scarce resource was never data. It was honesty about what the data says.

The HiPPO — the term is Avinash Kaushik’s[6] — is the most expensive algorithm in business: senior, confident, and wrong at scale. The antidote is a culture where evidence outranks title.

🍕 Domino’s Pizza, 2009. Their pizza was rated worst in class. Instead of a PR spin (the HiPPO move), they used raw feedback data to reinvent the recipe — and pivoted into “a tech company that sells pizza.” Over the following years Domino’s stock famously outpaced even Amazon’s and Google’s over comparable windows.

When official channels lose to opinions, employees vote with their logins. MIT found a thriving “shadow AI economy”: only about 40% of companies have licensed an enterprise AI subscription, yet employees at over 90% of surveyed organizations regularly use personal AI tools for work.[1] The HiPPO doesn’t stop adoption — it just pushes it off the books, ungoverned and unmeasured.

There is a subtler trap hiding inside the 95% statistic: many pilots reported “no measurable P&L impact” because nobody set a baseline before launch — the absence of measurement read as the absence of value.[1] The fix is boring and decisive: pick the number, record it before you build, review it after. McKinsey’s 2025 survey lands in the same place — 88% of organizations use AI somewhere, yet only 39% can point to any EBIT impact, and the strongest single correlate of impact is redesigning the workflow around the tool, which only about 21% have done.[2]

Before Target can predict customer behavior, it needs clean, unified purchase history. Losers try to implement AI on top of messy spreadsheets — and then blame the model.

The raw material is getting harder, not easier: industry estimates put 80–90% of enterprise data in unstructured form — documents, images, logs, video — growing far faster than the tidy tables dashboards were built for.[3] And budgets keep aiming at the wrong target: most GenAI spending chases visible sales-and-marketing pilots, while the clearest returns in MIT’s data sat in unglamorous back-office automation.[1]

Governance is not bureaucracy — it is the cheapest accuracy you will ever buy. Verify at the source, or pay a hundredfold at the decision.