What your AI budget actually buys
The question a board asks about AI is almost always the same: when does this show up in the numbers? It is a fair question. You are measured per quarter, the licences are already running, and someone has promised that a pilot in service will pay for itself before the summer.
The honest answer is that it depends on which of two things you bought. Installing buys the cost base of this year. Integrating buys the rate of improvement, and a rate compounds. This article lays out the difference and the evidence behind it. A second part follows on what you do with it.
Installing buys one year
A tool dropped on top of an unchanged process gives you a saving you can book once. You buy seats, you switch on a summarisation feature, handling time drops a little, and the business case holds for about three quarters. Then the line flattens, because the work around the tool never changed. Same routing, same scripts, same handover from the bot to the human being, same reasons people call twice.
That is not a failed project. It is a purchase with a known ceiling, and it is worth knowing you bought it.
Integrating buys the slope
Integration means the process, the data, the decision rights and the roles move with the technology, so every month starts a little higher than the last. The gain is smaller in quarter one and larger in quarter six, because each improvement sits on top of the previous one.
The cost base is a number you book once. The slope is a number you keep booking.
We have seen this before
Agile transformation gave us a clean test of the same idea, and most organisations failed it. Roughly seven in ten agile transformations did not deliver. Teams got stand-ups, boards and a new vocabulary while the budget cycle, the hierarchy and the handovers stayed exactly where they were. The ceremony was installed. The operating model was untouched.
McKinsey puts the gain from a real transformation at around 30 percent. That number belongs to the organisations that changed how decisions were made, not to the ones that bought the ritual. Journey mapping went the same way. Plenty of walls covered in sticky notes, far fewer processes rebuilt around what the map showed.
The controlled counterexample
Brynjolfsson, Li and Raymond studied 5,172 support agents who were given an AI assistant instead of being replaced, and the results say something specific about design. Productivity rose by roughly 15 percent. The least experienced agents improved by about 30 percent. Escalations fell. So did attrition.
Why the number sits where it sits
The gain landed with the newest people because the assistant carried the knowledge of the best performers into their hands during the conversation. That is integration: the model sat inside the work, in the moment the work happened, with the human being still holding the customer. The escalation and attrition numbers matter to a board as much as the productivity figure, because both feed next year's cost base without appearing in this year's licence line.
What you do on Monday
- Ask of every running AI initiative which of the two you bought, and write the answer down in one sentence per initiative.
- Put a second line next to every business case: the one-off saving, and the monthly rate of improvement you expect after month six.
- Pick the process where your least experienced people do the most customer contact, and start there, since that is where the evidence says the gain is largest.
- Name one person who owns the process change, not the tool rollout, and give that person the decision rights to change routing and scripts.
Part two covers how to move an initiative from installed to integrated without stopping the quarter. If you want to check where your organisation stands before that, the CX+AI Maturity Check walks through the same distinction per domain and gives you a score you can take into a board meeting. Bekijk de materialen