Three years into the applied AI cycle, the pattern is clearer than it was at the start. The businesses that compounded with AI did not adopt it earlier or harder than their competitors. They adopted it where it changed the economics, and they ignored it everywhere else.
That is the discipline. It is harder than it sounds because almost every internal voice will argue for the opposite. The product team will want it in the product. The marketing team will want it in the funnel. The board will want it in the deck. Most of those deployments will produce nothing that compounds.
Where applied AI actually changes the economics
There are three places. Operators who recognise them early move ahead of their category. Operators who do not, spend two years deploying tools that do not move the operating curve.
The first is decisions at scale. Anywhere the business runs a decision more than ten thousand times a year, and the decision can be specified well enough to instrument, an applied AI layer changes the economics. Pricing, routing, classification, ranking, support triage. The unit economics of each decision shifts, and the cumulative effect compounds.
The second is production at low marginal cost. Anywhere the business produces output that scales with hours, an applied AI layer collapses the cost curve. Content, code, design, analysis, transcription. The question is not whether to deploy. The question is which layer of the production stack to deploy at.
The third is operating leverage. Anywhere the operator's calendar dominates the business, an applied AI layer absorbs the routine. Reporting, monitoring, follow-up, scheduling, drafting. This is the least exciting category and the highest leverage. The operator who reclaims twenty hours a month from this layer compounds faster than any of their peers.
Applied AI does not need to be everywhere. It needs to be in the three places where it changes what the business is, not what the business looks like.
What we usually find
Most companies we engage with at this layer have AI deployed in the wrong three places. The product has an AI feature that does not move conversion. The marketing has an AI tool that does not move CAC. The leadership has an AI briefing that nobody reads. Meanwhile the operations layer, the support layer, and the decision layer are running on the same processes they ran on three years ago.
The reorganisation is not a technical migration. It is a discipline. We take the AI out of where it does not earn its place and put it where the economics demand it. The total spend usually drops. The total leverage usually rises.
The honest version
Most boards want AI in the deck because the market expects it in the deck. The companies that compound do not optimise the deck. They optimise the operating curve. The two are not the same.
The work we do in applied AI is not about being early to the technology. It is about being disciplined about where it goes. That discipline is harder to develop than the technical capability. It also matters more.
What we look for
We engage on applied AI where the operator has already tried, already deployed, and already learned that adoption is not the same as advantage. The work is not to teach AI. The work is to redirect it.
This is a narrower lane than the market suggests. It is the only one where AI actually compounds.