The pilot and the operating model
Most teams I meet can show me a working AI pilot, and almost none can show me a process that now runs differently because of it. The demo works. The model answers. The slide says "promising results". Then the quarter ends and the customer journey looks exactly like it did in January.
That gap has a shape, and it is the same in every organisation I walk into. It sits between the way of working we built for digital projects and the way of working AI actually asks for.
The old way: the pilot as proof
Yesterday's logic was simple: prove it small, then scale it. You picked a contained use case, you ran it next to the live process, you measured a lift, and you asked for budget to roll it out. The pilot was the argument.
AI breaks that logic at the handover. IDC found that roughly 88 percent of proofs of concept never reach broad rollout. The pilot wins the argument and still dies, because the thing it proved was the model, and the thing that blocks it is the process around the model: who approves, who owns the data, who is accountable when the answer is wrong.
The new way: the journey as the unit of work
Today's logic starts with the work the model has to live inside. You take a customer journey, you decide where decisions get made in it, and you redesign those decision points so a model can carry part of the load with a human accountable for the outcome.
EXL saw the split clearly: 44 percent of leading organisations had redesigned their operating model, against 23 percent of the laggards. Same technology, available to both. The leaders changed the way work flows; the laggards added a tool to the work they already had.
A model placed inside an unchanged process inherits every limitation of that process.
Where it tips
The tipping point sits at level three. The maturity ladder runs from level 1, where AI happens ad hoc, to level 5, where systems improve themselves. Only about one in five organisations gets past level 3.
Level 3 is where you have working use cases and no system around them. Everything above it requires decisions that a project team cannot make on its own: data access as a standing arrangement, decision rights written down, a funding model that survives the pilot.
High CX maturity is associated with up to 30 points of extra NPS. That number does not come from a better model. It comes from an organisation that can act on what it knows, every day, in the same way.
What the leaders do differently
The practical difference is where the data sits. Leaders are roughly 2.5 times more likely to have data available at the point where the decision is actually made, not in a dashboard someone opens on Friday.
The return follows the structure. ServiceNow found that pacesetters realise an average of around 160 percent return, and high CX maturity is also associated with up to a third lower service costs.
The confidence problem
Almost everyone in this market believes they are already ahead. EXL surveyed 322 C-suite leaders: 76 percent believed they were ahead of the competition, and roughly 10 percent actually were.
IBM and Ecosystm found the same pattern in APAC. There, 85 percent described themselves as data-driven or AI-first, while 11 percent could demonstrate it. Measure yourself against evidence, and the conversation about priorities gets much shorter.
What you do on Monday
- Pick one journey and mark every point in it where a decision gets made. That is your map of where AI can do work.
- Check, for the top three decision points, whether the data is present at the moment of the decision. Write down the answer.
- Score yourself on the ladder from 1 to 5 using evidence someone outside your team would accept.
- Name one person who owns the outcome of each automated decision, and put it in writing this week.
- Take one stalled proof of concept and list what has to change in the process, not in the model, for it to run live.
If you want a scored starting point instead of an opinion, the CX+AI Maturity Check walks you through the five levels with the evidence each one requires. It pairs with the AI Decision Rights template, which is where most teams find out what is actually blocking their pilot. View the materials