What a Year of Agentic Conversations Taught Us About Customer Intent

Spend a year reading how businesses talk about their customers and you get one story: demand is chaos, the phone is unpredictable, you just have to staff for it and hope. Spend a year reading the conversations themselves — which we did, across an anonymized, aggregate slice of the traffic on our platform — and you get the opposite story. Customer intent is not random. It arrives in a small number of shapes, at times you can forecast, and understanding that pattern is worth more than any single feature you could bolt onto a phone system.

A note on the data before anything else, because it changes how you should read it. Everything below is first-party and vendor-published: it comes from our own platform, aggregated across industries and stripped of anything identifying. That makes it directionally honest but not a peer-reviewed benchmark, so we’ve kept every figure as a range and leaned on independent research wherever a harder number was available. Treat our numbers as “what a year of real conversations looked like,” not as a law of nature.

Customer intent is narrower than anyone staffs for

The single most consistent thing we saw is how few distinct things people actually want. Across verticals as different as plumbing, dental, and auto insurance, the large majority of inbound conversations collapsed into a handful of intents: book or reschedule something, check the status of something already in motion, get a price or a quote, ask one factual question, or report an emergency. Everything else — the genuinely novel, genuinely human conversation — was the long tail, not the trunk.

That matters because most front desks are staffed as if every call were unique. They aren’t. When roughly half of what comes in is “when can you come out” or “where’s my technician,” the operational question stops being “how do we handle infinite variety” and becomes “why is our most expensive resource — a trained human — spending its day on the five questions a system can answer perfectly every time?”

Intent has a clock, and it runs after hours

The second pattern is timing. Intent isn’t distributed evenly across the day; it clusters, and a large share of it lands when the office is closed. In our aggregate, a meaningful minority of all conversations happened outside 9-to-5 business hours — and the after-hours mix skewed heavily toward high-value intent: emergencies, first-time bookings, and people shopping a decision they’d been putting off all day.

This lines up with what independent researchers have long reported about the cost of a closed door. Roughly a quarter of inbound calls to home-services businesses go unanswered, per Invoca’s industry data (vendor-published), and the well-worn speed-to-lead research from Harvard Business Reviewfound that firms responding within an hour were far likelier to qualify a lead than those who waited even a few hours. The after-hours caller isn’t a lower-value caller. Frequently they’re the opposite — and they’re the one most likely to hit a voicemail box.

The same words, different urgency, by vertical

Intent shape is stable, but its urgencyand its ideal response are wildly different depending on the business. “I have a problem” means one thing to a plumber at midnight and something completely different to a remodeler mapping a six-month project. Reading the two the same way is how good tools still produce bad experiences.

The lesson isn’t that you need five different products. It’s that a system reading intent has to know which vertical it’s in before it decides what a good response looks like. The plumbing agent that treats a burst pipe like a scheduling request fails; so does the remodeling agent that treats a $60,000 kitchen inquiry like a booking to be closed on the first message.

The most valuable signal is the one nobody logs

Here is the finding that surprised us most. The highest-value pattern in a year of conversations wasn’t in what customers asked — it was in what businesses never recorded. A missed call leaves no intent behind. A voicemail nobody returns leaves no intent behind. An abandoned hold leaves no intent behind. The conversations that never happened are invisible on every dashboard, which is exactly why the loss they represent is so easy to ignore.

You can’t manage the intent you never captured — and the intent you never capture is disproportionately the intent worth the most.

When most people who don’t reach a person the first time never call back — a pattern consistently reported across home-services research, including Invoca’s benchmarks (vendor-published) — every unanswered high-intent moment is a permanent one. The instant you actually answer those conversations, they stop being a guess and start being data: a recorded, reviewable, forecastable stream of what your customers want and when. That is the real unlock. Not that AI answers the phone, but that answering turns lost demand into a signal you can finally act on.

What an operator should actually do with this

None of this is useful as trivia. It’s useful because intent being predictable means your response can be, too. Three moves fall directly out of the pattern:

• Staff for the shape, not the average. If half your volume is scheduling and status, stop paying skilled people to do it and free them for the long-tail conversations only a human can handle.
• Cover the predictable window. You already know the nights, weekends, and rush hours where high-intent conversations arrive unanswered. That gap is a config decision, not a hiring problem.
• Capture everything, then read it.The point of answering every conversation isn’t only the booking you save today. It’s the year of intent data you build — the thing that tells you what to staff, stock, and sell next.

The businesses that treated their conversations as a data asset, rather than a cost to be minimized, ended the year knowing things their competitors were still guessing at: which hours actually convert, which questions predict a big job, which silence predicts a churned customer. That knowledge compounds. It’s the difference between running a front line on instinct and running it on evidence.