Your AI has never met your customer.

I once spent 30 minutes in a Frankfurt parking garage helping a rental car associate find my reservation. The digital experience had been so frustrating that the two of us ended up piecing it together by hand, side by side.

And this company had analytics and dashboards. But nobody in that garage could see what I'd been trying to do.

I think about that garage every time I see an AI chatbot open with "How can I help you?"

It should already know.

"I see you're picking up a car in Frankfurt and couldn't find your reservation. Want help with that?"

Same model. The only difference is what it knows before you say a word.

I've spent the last eleven years talking with digital leaders about their customers. Right now, every one of those conversations comes back to AI: the mandate from the board, the pressure to show what it changed, and where it can automate action for us.

Most AI plans I see cover code changes, migrations, tools, data access, and governance. Very few ask what the AI knows about the customer. AI got very good at reasoning. What it should reason about didn't keep up.

AI keeps failing in the enterprise. Not because it can't take action, but because it doesn't know which actions it should take. People need the same thing, and it comes from context. Customer context: who the customer is and what your relationship with them looks like. Your CRM and CDP handle that. And experience context: how they're engaging with you right now. That second one is the one most companies don't have.

Not every AI project needs it. If you're moving a codebase from one language to another, you don't need to know what your customers did. But if the goal is a better customer experience, it's the first thing your AI needs to know.

The research points the same way. MIT's 2025 study, The GenAI Divide, found that 95% of generative AI pilots produced no measurable P&L impact, and pointed to AI that doesn't learn or adapt to context as a core reason.

What is experience context?

Experience context is what customers did, what happened as a result, and how they felt about it. It matters when you can see what it cost your business, and how it compares to normal. That's what lets people and AI act on it.

For most businesses, those moments happen on a website or in an app. What customers did: the campaign that brought them in, the products they looked at, the content they read. What happened as a result: the error they hit, the page that took eight seconds to load. And how they felt about it: tapping the same button over and over, giving up, or telling you in a survey or on a call.

Any one of those is just data. Put together, they tell you what someone came to do and whether they got there.

Brand context vs. experience context.

Adobe uses "experience context" for the brand rules an AI follows when it creates content: voice, approved claims, and design system. That's context about how you want to show up. What I'm talking about is what your customers actually lived through. You can have AI write perfectly on-brand copy for a checkout page that's failing customers. Enterprises will need both.

Why this is a leadership decision.

Your board isn't going to ask whether you deployed AI. They're going to ask what impact it had on your customers and your business. Experience context is how you answer: which customers struggled, what it cost, and whether the fix worked.

The first risk is working on the wrong things. You could argue that doesn't matter anymore: if AI can fix everything, why prioritize? But your engineers could fix everything, too. They don't, because the real limit has always been cost, and AI doesn't change that. Every fix costs tokens. Every fix needs someone to review it. And every change, even a small one, can break something else. When the fix wasn't worth making, you pay all three for nothing. Humans or AI, the rule is the same: don't spend time on things that don't matter.

Then there's trust. What would you let AI do on its own today? What's stopping you from letting it do more? For most leaders I talk to, it's that they can't check the work. You can only hand off what you trust, and trust comes from evidence. An AI that can show you which customers hit the problem, and what they were trying to do, earns that trust much faster than one that's guessing.

It gets harder with every agent.

We need to get the basics right. As AI systems get more complex, one agent hands work to another, which calls a tool, which makes a change. Each step has to start with the right information. Step 1: what is the customer's intent? Get that right, and steps 2 through 100 get better.

Get it wrong, and it multiplies. Ten agents working from ten partial pictures of the customer don't give you ten times the insight. They give you ten versions of the truth. Your CRM, your business metrics, your backend monitoring, your ticketing system: every one of them will say it knows what matters most. After a decade of doing this, I haven't found anything that beats what the customer actually lived through, and what it cost the business.

What it looks like in practice.

One of the largest telecom companies in the US showed us this summer. Customers with suspended accounts were trying to pay to get their service back, and their payments kept getting declined. They'd retry. Declined again. Same generic error every time.

Their engineers had built an AI agent that pulls experience context from Quantum Metric alongside their logs. It could see which customers were stuck, what they were trying to do, and where it broke. Then it traced the problem to one line of code: the app only checked that a payment was more than zero, not that it met the minimum to restore the account. It filed the bugs across three codebases and drafted the fixes. All three shipped.

Their logs saw an error. Experience context saw a customer trying to get their phone working again.

Where does your AI stand?

Ask your team five questions:

1. When a key metric moves, can our AI tell us which customers were affected, what they hit, and what it cost, without people stitching tools together?
2. Before an AI agent changes something customers touch, does it know how many customers the issue affects?
3. Does our AI know what normal looks like for our business?
4. Do marketing, product, and engineering get the same answer to the same question?
5. After we ship a fix, can we prove what it earned?

If most of the answers are no, your AI is meeting your customers for the first time, every time.

So here's my ask. Before you approve the next AI initiative, ask your team one question: does it know what our customers are trying to do?

It's the problem we built Quantum Metric to solve. AI can't fix what it can't see. Give it the customer's side of the story first.

Want to see what your AI is missing? Bring us your hardest question, and we'll answer it with your own data.