Trust Is the Product: What TM Forum Americas Taught Us About AI-Native Commerce

TM Forum’s Innovate Americas brought operators, hyperscalers and vendors to Dallas last month, all focused on building the AI-native telco. A year ago, most AI conversations at events like this were about what the models could do. In Dallas, the questions were harder and more useful.

Can we trust what the model decides? And when it decides, does anything actually happen without a person stepping in to finish the job?

Those two questions, trust and the closed loop, are tied together more tightly than most of the industry is admitting. Here’s what I took away, and what it means for anyone running telecom commerce.

The trust gap is wider than the confidence gap suggests

The number that framed the week came from TM Forum and IBM research released alongside the event. 72% of AI decision-makers at telecom operators said their AI capabilities were trustworthy. Only 14% said they could provide evidence to back that up.

Frontier models are remarkably capable. They can reason across a customer’s account, plan, device and history faster than any person could. But capability is not the same as trust. Most operators still keep AI in an advisory role. It suggests, a person decides, and a person executes.

TM Forum made the point plainly: the cost of a hallucination or a bad call climbs as AI moves from supporting decisions to making and executing them. So the checkpoints stay. AI drafts the offer and a rep approves it. AI flags the fraud risk and someone works through the review queue. AI recommends the device and the associate re-keys the order.

That caution feels like the safe choice, but those checkpoints carry their own risk, because every manual handoff is a place where context is lost, data goes stale and errors creep in. Operators end up trusting the person in the loop over the model, even when that person has less information to go on.

TM Forum’s guidance on AI is blunt: it becomes operational only when ​“every principle is supported by continuous, verifiable evidence.” That’s not a policy problem. It’s an architecture problem.

Bolt-on AI leaves the loop open

Most AI in telecom commerce today is bolted on. A copilot sits beside the POS. A chatbot sits in front of the website. A recommendation engine feeds a dashboard someone may or may not check. Each one is smart on its own. None of them can finish the job.

That’s an open loop. The model produces an output, and then the process depends on people to carry it the rest of the way.

• An associate copies the recommendation into another system.
• A manager approves an exception in a separate queue.
• Back office reconciles the activation, the inventory move and the commission after the fact.

Every one of those steps is a dependency on people and process, not on the platform. The AI is only as fast, accurate and accountable as the slowest manual handoff behind it. And because the action happens outside the system, there’s no clean record of what the AI recommended, what a person changed, and what the outcome was.

An AI-native application works differently. AI isn’t a layer on top of the workflow; it’s built into how the workflow runs. The same platform that holds the customer, the inventory, the pricing and the transaction can sense an event, decide, act and confirm the result. That’s a closed loop.

Closed loops are how trust gets earned

Here’s the connection I don’t think gets enough attention. You can’t build trust in AI that you can’t observe end to end.

In an open loop, AI makes a suggestion and disappears. Whether it was right gets decided later, by someone else, in another system. There’s no evidence trail, so there’s no way to move from ​“we believe it works” to ​“we can prove it works.” That’s exactly the 72%-versus-14% gap.

In a closed loop, every decision is tied to its outcome. Did the offer convert? Did the device ship from the right location? Did the activation complete, and did the margin hold? That evidence is what lets you widen the guardrails over time. Start with AI acting on low-risk, high-volume tasks, measure it, then hand it more.

This is especially true in telecom commerce, where a single sale touches the telco, the device, the plan, inventory, payments, trade-ins and commissions. Bolt AI onto that and you multiply the open loops. Build AI into the commerce platform itself and each of those steps becomes an observable, accountable action.

The people don’t go away. Their role changes. Instead of being the glue that holds disconnected systems together, they set the guardrails, handle the genuine exceptions and spend their time with customers.

Three questions to ask of your AI roadmap

If you’re evaluating where AI fits in your stores and channels, start with these three.

1. Where does the loop break? Trace one AI recommendation from output to outcome. Count every time a person has to copy, approve or reconcile something to finish it.
2. Can you prove it worked? If the answer lives in a spreadsheet someone builds at month-end, the loop isn’t closed.
3. Is AI in the workflow or next to it? A model sitting beside your systems will always depend on people to act. A model inside them can act, and show its work.

The operators in Dallas weren’t asking whether AI is ready. They were asking whether their architecture is ready to let AI be trusted. That’s the right question. At iQmetrix, it’s why we’re building intelligence into the commerce platform itself rather than bolting it on, so every decision is connected to the data, the transaction and the outcome.

The AI-native telco won’t be defined by the smartest model. It will be defined by the loops it can close.