ADT now sends half its calls to AI first, and the order it did things in is the lesson

On ADT’s first-quarter call last Thursday, CEO Jim DeVries said every chat and about half of the company’s phone calls are now routed through AI first (earnings call transcript). ADT is the home security company most Americans know from the sticker on a neighbor’s window. It’s a large company with a large call center, and it gave out real numbers.

Most AI rollout stories are announcements. This one came with a scorecard, and with a sequence any company running a support team can follow.

What ADT actually said

Here’s what DeVries and his team put on the record, per the transcript.

• Chat containment, the share of chats resolved without a person, was 45% in the first quarter and 60% by April.
• Call containment was 16% in the first quarter and 25% by April.
• DeVries said he expects most of the calls in the call center to be handled through AI this year, and added that he isn’t sure the improvement will keep moving in a straight line.
• People still handle the situations where human expertise matters most, and he named emergencies as the example.
• Next on the list: transcribing and analyzing calls, modeling which customers are likely to cancel, and AI call routing within a few weeks.
• In sales, outbound calls to low-converting leads will now be made mostly by AI.
• On the engineering side, Omar Khan, president of ADT Smart Home, said more than half of the company’s committed code is now written by AI, and that this has kept engineering headcount flat while adding capacity.

CFO Jeff Likosar didn’t give a savings target. He described the benefit as many millions of dollars, more if you count fewer service trucks sent to homes.

DataShare of contacts ADT's AI resolved without a person

• Chat, Q145%
• Chat, April60%
• Calls, Q116%
• Calls, April25%

The order is the lesson

Look at the sequence. ADT didn’t start by putting AI on every phone line. It started with chat, where a customer types, the agent reads, and a wrong answer is easy to spot and correct. Chat containment is now more than double call containment. Voice came second and is climbing from a lower base.

Chat also gives you a written record of every conversation. When the AI gets something wrong, you can read exactly what it said, fix the instruction or the source document behind it, and check the next week whether the fix held. Voice is harder to review, so it makes sense to learn on chat first.

Then it measured one number, containment, and reported it by month. You can argue with containment as a metric, and I’ll get to that. But a single number that leadership watches every month is how a rollout keeps moving.

Third, it drew a clear line around what stays with people. Emergencies go to humans. For a security company, that’s the call where a mistake costs the most, so it’s the last thing you’d hand to a machine.

It also kept the rollout inside one department long enough to learn from it. Engineering adopted AI coding tools in parallel, but customer-facing AI stayed in service until the numbers were moving.

Only after service was working did ADT move AI into marketing and sales, starting with outbound calls to the leads least likely to convert. That’s a smart first sales use. If the AI does worse than a person on those calls, the cost is small, because those leads weren’t converting anyway.

TimelineADT's rollout, as described on the call

1. Q1All chats routed through AI first
2. Q1About half of calls routed through AI
3. AprilChat containment reaches 60%
4. Coming weeksAI call routing goes live
5. This yearMost calls handled through AI

One caution on the metric. Containment tells you the customer didn’t reach a person. It doesn’t tell you the customer got the right answer. Depending on how you count it, a customer who gives up and hangs up can look the same as one whose problem got solved.

ADT pairs containment with customer satisfaction. DeVries said the company’s net promoter score is as high as it has been in his ten years there. That’s the right pairing, and it’s the one I’d insist on in any support rollout. If containment goes up and satisfaction goes down, your AI is closing tickets and losing customers.

The headcount question ADT left open

ADT didn’t say on the call whether call center staffing will change. The only headcount comment came from engineering, where Khan said AI had added capacity while engineering headcount stayed flat. For most companies, that’s what AI savings look like in the first year. You take on more work without hiring at the same pace.

Your support team will read stories like this one and do the math on their own jobs. Tell them your plan before they guess at it. Be specific about three things: which request types go to AI first, what happens to the hours that frees up, and what work you want them doing instead.

There’s usually plenty. ADT’s own list points at it. Once routine questions are handled, people can spend their time on the customers likely to cancel, on emergencies and complex cases, and on reviewing what the AI said to customers last week. That’s harder work than answering “where’s my technician?” for the fortieth time, and most good support people would rather do it.

The labs are betting on companies smaller than ADT

Yesterday, Anthropic announced a new AI services company with Blackstone, Hellman & Friedman and Goldman Sachs. Its job is to help mid-sized companies put Claude into their core operations (Anthropic). The announcement names community banks, mid-sized manufacturers and regional health systems as the kind of firms that could benefit but lack the in-house resources to build these systems.

Anthropic’s CFO, Krishna Rao, said enterprise demand for Claude is “significantly outpacing any single delivery model” (Blackstone). Read that as an admission: the hard part of AI in a business is the rollout itself, with someone mapping the work, building the system and measuring it. The model is the easy part.

You don’t need ADT’s call volume to copy what it did. A 60-person company with a four-person support team can follow the same order at a smaller scale: start in chat, measure one number, keep the high-stakes cases with people, then move to the next department.

What to do this week

• Pull last month’s support volume and split it by channel. If you have chat or email, start there before phone.
• Pick the one or two request types that make up the biggest share of volume and have a clear right answer, like order status or password resets.
• Decide in writing which cases always go to a person. Start with anything involving safety, legal threats or large refunds.
• Choose two numbers to track monthly: the share resolved without a person, and customer satisfaction on those same conversations.
• Put both numbers on the leadership meeting agenda for the next three months, the way ADT put them on an earnings call.

Sources

• ADT (ADT) Q1 2026 earnings call transcript (The Motley Fool, April 30, 2026)
• Building a new enterprise AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs (Anthropic, May 4, 2026)
• Anthropic partners with Blackstone, Hellman & Friedman, and Goldman Sachs to launch enterprise AI services firm (Blackstone, May 4, 2026)