Customer Experience & AI Integration
Point the AI at the Customer
AI rarely fails on capability. It fails on aim. Pointed at the cost center, it produces a rounding error and a press release; pointed at a specific customer outcome — and made trustworthy enough that customers and operators actually rely on it — it compounds into loyalty, revenue, and a product competitors can't match. Integration is a customer-experience decision before it is a technology one.
By Tom Hanks· July 2026· 7 min read
The most quoted statistic in enterprise AI is a warning that everyone repeats and almost nobody heeds. In MIT Sloan Management Review and BCG's study, seven in ten companies reported minimal or no impact from their AI investments — even as nearly all of them called AI a major opportunity. Years and several hype cycles later, the ratio has barely moved. If you have sat in the rooms where AI budgets are set, you already suspect why, and it is not that the models weren't good enough. It is that the AI was pointed at the wrong target.
Walk through how most AI programs are actually justified. They are aimed at cost — deflect tickets, cut headcount, automate a back office. Or they are aimed at appearance — the board wants an AI story, so there is now an AI story. Both of those aim the technology at the org chart, at internal efficiency, at the company's own convenience. And the returns come back exactly as small as the aim. The AI worked; it just wasn't pointed at anything that mattered to the person paying the invoice.
The value is where the customer is — that's not a slogan, it's the data
Here is the pattern hiding inside the disappointment. When McKinsey mapped where generative AI is actually creating value, it clustered in the functions that touch the customer — service operations, and marketing and sales — not in the places companies instinctively cut first. And when McKinsey studied personalization — which is just a technical word for getting closer to the individual customer — it found that faster-growing companies drive roughly 40% more of their revenue from it, with a typical revenue lift of 10 to 15%. The signal is consistent and it is loud: AI pays where it reaches the customer, and it fizzles where it only reaches the balance sheet.
This should reframe the entire integration question. "Where do we put AI" is not an IT roadmap decision or a cost-takeout decision. It is a customer-experience decision wearing a technology costume. Ask it that way and the answer organizes itself around three surfaces — the three places a company actually meets its customer.
Three surfaces, one aim
Operations — resolve the customer's problem before it becomes a grievance. Every company has a version of the Sunset problem I've written about elsewhere: customers on hold for thirty minutes, issues that take a day to close. The lazy AI play is a chatbot that deflects tickets so the cost line drops and the customer's problem doesn't. The right play aims the same technology at the customer outcome — resolve it fast, the first time — through triage that routes correctly, agents that surface the answer to the human before the customer has to repeat themselves, and detection that fixes the problem before the customer calls at all. Same technology, opposite target. One shrinks a cost center; the other closes the delivery gap.
Production and delivery — make the quality and reliability something the customer can feel. A defect is not an engineering statistic; it is a customer's moment of failure at the worst possible time. At AuthenticID I treated an 85% defect reduction as a trust program, not a quality program, because that is what the customer experienced — and it helped retain $15M in ARR. AI belongs here too, watching for the failure the customer would have felt and heading it off. In the field platforms I build now, the design principle is blunt: an agent that only watches is an advisor; an agent that acts — locally, in the seconds that matter, and then explains what it did — is a teammate. The customer never sees the model. They see a product that just doesn't break the way the last vendor's did.
Product — anticipate the job the customer is trying to do. This is where personalization earns its 10–15%, but only when it is built as service and not as surveillance — the product that remembers, predicts, and removes a step because it understands the customer's intent, not because it is mining them. Pointed correctly, AI lets the product do more of the customer's work for them, which is the only kind of personalization anyone has ever actually thanked a company for.
Notice what all three have in common. Each is aimed at a single, stated customer outcome, and each is measured against that outcome — not against model accuracy, not against tickets deflected, not against "AI adoption." That is the discipline that separates the seven-in-ten from the one.
The part everyone skips: AI only helps if it's trusted
There is a failure mode that turns well-aimed AI into a worse customer experience, and it is the one I see most often. AI the customer doesn't trust — or that the operator doesn't trust — doesn't get used, or worse, gets used and resented. An opaque model that curtails a customer's service with no explanation, a recommendation engine that is confidently wrong, an alert system that cries wolf until everyone mutes it — each of these is AI actively widening the delivery gap.
Trust is not a soft add-on to the integration; it is part of the integration. The U.S. government's own AI Risk Management Framework names explainability as a core characteristic of trustworthy AI for exactly this reason. In practice it means AI that shows its work — "I flagged this because it matches the last three failures; here's the evidence" — and that earns authority in order: let it observe and explain while humans act, let it act on the low-stakes high-frequency cases where being right is obvious, then widen its authority one verified category at a time. An AI the customer and the operator both trust is the only kind that improves the experience. The rest is a demo.
Why this is the hurdle most companies can't clear alone
Pointing AI at the customer sounds obvious once it's said. It is genuinely hard to do, and that difficulty is exactly the hurdle that holds most companies back — because it requires three things to be true in the same person or the same aligned team: someone who understands the customer outcome deeply enough to pick the right target, understands the operation well enough to integrate AI where it actually moves that outcome, and understands the technology well enough to build it so it's trustworthy rather than a black box. Companies usually have those three understandings in three different silos that don't reconcile, so the AI gets aimed by whichever silo owns the budget — and that is almost always the cost center.
I do this work in one seam. I've carried the customer outcome as an operator, integrated the systems as a product leader, and I architect and ship the AI myself — down to the autonomous edge agents that act and explain in the field. You don't have to take that on faith, which is the point: you can run the open-source tools I've published, and you can log into the platform I'm building now and watch the agents do it. I would rather show you a working system than a slide about one.
AI doesn't transform a company. It amplifies whatever the company was already pointed at. Aim it at your cost center and you'll get a slightly cheaper version of the same experience. Aim it at your customer, make it trustworthy, and it compounds into a product the market can't figure out how to match.
The question to carry into your next AI meeting
When the next AI proposal lands on your desk, don't start with the model, the vendor, or the savings. Start with one question, and refuse to move until it's answered in the customer's language:
What specific customer outcome does this make measurably better — and how will the customer feel the difference?
If the honest answer is "it doesn't, but it cuts cost," you've found why seven in ten of these disappoint. Send it back and re-aim it. Because the money in AI was never in the efficiency. It was, like all the money always is, in the customer — and it goes to the companies disciplined enough to point everything, including their most powerful new tool, directly at them.