Most AI budgets get spread thin across pilots that never compound. A disciplined 90-day audit finds the two or three workflows where AI actually pays for itself, and gives you the ROI model to defend the spend.
Almost every enterprise we meet is busy with AI. There are pilots in marketing, a copilot trial in support, a data-science team fine-tuning something nobody in operations has heard of. Activity is everywhere. Return is almost nowhere.
The numbers bear this out. Adoption is now near-universal, but value is rare, and that gap is exactly what a disciplined audit is built to close.
The problem is rarely a lack of ambition or talent. It's that the spend is spread a centimeter deep across fifty use cases instead of concentrated where the economics are real. A focused 90-day audit fixes that, not by adding more pilots, but by finding the few that deserve to become systems.
Start with the work, not the technology
The fastest way to waste a quarter is to start from the model and hunt for somewhere to apply it. We start from the opposite end: the work. Which processes consume the most senior time? Where do handoffs stall? Which decisions get made hundreds of times a day on incomplete information?
Map the actual workflow, the steps, the systems, the people, the exceptions, before anyone says the word “model.” The map is where the opportunity becomes visible, and where the cost of the status quo finally gets a number.
Quantify before you build
Every candidate use case gets the same treatment: volume, cost per unit of work, error rate, cycle time, and the realistic share AI can absorb or accelerate. That gives you an expected annual value and, just as importantly, a confidence band around it.
And don't let perceived cost stop you from starting. As Andrew Ng put it at ScaleUp:AI, the economics of getting going have largely flipped:
“Don't worry about the price of LLMs to get started.”
Where ROI actually concentrates
Across audits, the winners cluster in the same places: high-volume document and data work, first-line support and triage, internal knowledge retrieval, and the long tail of “someone manually moves this between two systems” tasks. Unglamorous, repetitive, expensive, and exactly where AI compounds. The macro data agrees: value sits in the core of the business, not its edges.
Where that value shows up splits cleanly by function. Cost savings concentrate in service operations; revenue gains concentrate in marketing and sales, the single most common place firms report a top-line lift from AI. A good audit reads those patterns against your own cost base and prioritizes accordingly.
What a good audit hands you
You should walk out with three things: a prioritized shortlist of two or three use cases, an ROI model for each with the assumptions written down, and a build-vs-buy recommendation that accounts for your data and your stack.
That's it. Not a fifty-page strategy, not a platform RFP, a defensible decision about where to put the next dollar, and the evidence to back it. From there, the work is execution, which is a different discipline entirely.