Answer "is it me?" before the call

The business stake

Two numbers decide whether a wireless app earns its keep: how many customers stay, and how much it costs to serve the ones who need help. When the experience is split across legacy apps and a customer’s first move on a bad connection is to call, both numbers move the wrong way. Every avoidable call is cost to serve on a high-volume care channel, and every unanswered “why is my service bad?” is a reason to leave.

The problem

Customers had their account in one place, their network questions in another, and no quick way to tell whether a slow connection was AT&T’s network or something in their own home. So they called. The obvious fix was a status page, but a status page only says whether there is an outage nearby. It cannot tell a customer whether the problem is the network or their Wi-Fi, and that was the question behind the calls.

My role

I led four development teams and three product managers on the MyAT&T App. I owned the product strategy, roadmap prioritization, and the integration of network measurement into the app. That meant deciding what the health score had to answer for a customer, where it showed up, and what the app should do next when the answer was bad.

How we thought about it

We started from the customer’s question, not the data we had: “Is it me, or is it you?” Everything else followed from answering that in one glance and then making the next step obvious.

• Consolidate first. Folding the legacy apps into MyAT&T gave the health score one home that customers already opened to pay their bill.
• Glanceable, then actionable. The score lives in a home-screen widget, so customers see their connection health wherever they are. One tap moves from “how is it?” to “what do I do?”
• Diagnose before you fix. The app separates a network problem from a home setup problem, because the fix for each is different and the wrong one wastes the customer’s time.
• A real exit. If the guided fix does not work, the customer books a technician in the app instead of starting over on the phone.

AI and data

The health score is a model built on Amazon SageMaker. It reads two kinds of signal: QCI traffic congestion data from the network, and speed tests from the Ookla SDK, which also powers the widget’s measurement on the device. The model turns those into one score a customer can read without knowing what QCI means.

What shipped

We shipped the flow pictured at the top of this page, inside the consolidated MyAT&T App:

• A home-screen widget that shows connection health wherever the customer is.
• Tap to diagnose, which separates a network problem from a home setup problem.
• A guided fix with reset, Wi-Fi, and device steps, so the issue can be resolved in the app without a support call.
• Book a technician in the app when the guided fix does not solve it.

Outcomes

Results across the MyAT&T App during my two years on it, each paired with the business lever it moves.

• Retention up 5% Reported, the strongest shareholder-facing number from this work: customers who stay are revenue that does not need to be won back.
• 10% fewer support calls from app users Reported, which lowers cost to serve on a high-volume care channel.
• iPhone loyalty rewards engagement up 13% Reported, which supports retention at upgrade time.

What I’d do differently

I would agree on the exact definitions of each support metric with the care organization before launch, and publish them with the dashboard. Calls and tickets both moved, and having one shared definition from day one would have made the story simpler to tell and harder to argue with.