The Insight Engine Isn't a Cost. It's the Opportunity You Already Lost.

Nobody sends you an invoice for the two weeks you waited. You pay it anyway

By Elisabeth Virgilio, Founder, Maison Virgilio · November 2026

It is Tuesday morning, nine o'clock. Your monthly business review is about to start. On the screen in front of you: last month's performance, yours and your competitors'. The numbers raise one question only you can answer, but answering it properly means waiting on your team to pull the full analysis together, which takes two weeks.

Two weeks later you have your answer. Competition has not waited. They have moved on price, on shelf space, on the customer segment you were still deciding whether to chase. The decision you finally make is correct on paper. It is also two weeks and several market share points too late. Nobody sends you an invoice for that. It still cost you millions in lost opportunity, you simply never saw the bill.

I learned to put a number on that cost the hard way, at General Mills, where the same meeting happened in market after market every month. Someone pulled the sales data, someone else pulled the scanner data, and between gathering it and presenting it, two to three weeks went by, leaving next to no time to act before the next cycle began. So I built a system to cut those two to three weeks down to seconds, certain that speed alone would fix it. It helped. It did not fix it.

What actually protected the opportunity was not the speed of the number. It was two things nobody budgets for when they build one of these systems: getting the people around that monthly table to trust it enough to act differently because of it, and getting someone to approve paying for it before it had existed long enough to prove itself.

What an insight engine actually is

I call a system built to do this an insight engine: something that answers one specific, recurring business question directly from a company's own data, on demand, instead of restarting the analysis from scratch every month. The General Mills review above was my first one. At a later employer, a large industrial manufacturer, I orchestrated a second: an AI enabled view of existing customers built to surface exactly where growth was hiding inside accounts the sales teams already owned, so upsell and cross sell conversations could be prioritised on evidence instead of on whoever argued loudest in the pipeline review.

What is different now is that AI lets the system generate the actual answer, not just the dashboard it would once have been drawn on. That is a real step change from the scanner data and self service BI most companies already have, and it is also why so many now reach first for breadth, a platform that promises to answer everything, instead of depth, a system that gets one recurring decision right every month.

The part that isn't technical: getting people to act on it

A faster number is not automatically a different decision. At General Mills, the review only started changing what markets did once we attached a specific next step to specific movements in the data: if a competitor's price or share crossed a given threshold, a named person owned raising it in that same meeting, not a separate one. The system did not just get faster. The habit around it changed, deliberately, at the same time.

That is change management, and it is usually harder than the build. People do not trust a number because it arrived quickly. They trust it once they have watched it be right a few times, once someone senior visibly acts on it in front of them, and once using it is simpler than falling back on the spreadsheet they already know. Skip that work and the engine still gets built. It just does not get used.

Built to prompt action, not just answer a question

An insight engine that stops at the answer is only half built. The industrial manufacturer example above did not just surface where growth was hiding inside existing accounts. It was designed to name the accounts worth calling that week and point to who owned the call, so the output was a next step, not a research finding someone still had to translate into one. The engines that changed outcomes always had that last link built in from the start, not bolted on once the analysis proved interesting.

The part nobody budgets for: getting it approved

BCG's 2026 survey of CPG and retail executives, run with the Consumer Goods Forum, found that roughly 75% of CPG companies are still in pilot or exploration mode on AI, with only 18% scaling it to where it actually changes results.¹ Retail looks only slightly better. The report calls this the pilot trap: initiatives that work cleanly in a controlled test, then stall once they have to compound at scale.¹

Most of that stall happens before the system is even built, at the point where someone has to approve paying for it. A platform is hard to size and nobody wants to own its budget line. A named decision is easy to size: it already has a cost, every month, in the hours spent producing a number that is stale the moment it lands, and in the market share and margin lost while everyone waits for it.

The pitch that gets approved does not ask finance to believe in a platform. It names the decision the engine serves, names who owns acting on it once the answer arrives, and puts a number on what the current two week wait is already costing in lost opportunity, not on what the new system might someday save. That is a business case built the way a CFO already thinks, in opportunity cost, not in software cost, and it is one finance can sign off in a single meeting. A platform pitch, however elegant, usually gets sent back for a pilot, which is exactly where BCG found three quarters of these efforts are still quietly sitting.¹

Translating between the boardroom and the data team, convincing stakeholders the analysis would change what they decide, securing the actual funding, then getting people to genuinely change how they worked because of it, was the real job for twenty years before I started Maison Virgilio. Building the system was never the hardest part of it.

Elisabeth

The beauty of Data. Designed for Growth.

References

1. Boston Consulting Group with The Consumer Goods Forum, “AI in CPG and Retail: How Winners Are Pulling Ahead” and accompanying press release, 24 June 2026. Survey of 39 senior CPG and retail executives; figures are directional given the sample size. CPG: ~75% still in pilot/exploration, 18% scaling significant impact, 220 to 350 bps of potential cumulative EBIT from scaling AI across relevant initiatives. Retail: ~45% scaling, ~40% barely begun, 180 to 360 bps potential EBIT. https://www.bcg.com/press/24june2026-cpg-retail-leaders-ai

Next
Next

Nobody ever grew a brand by approving invoices faster