The Leak That Never Moved

Hats off to Jon Torrey and the team at SearchLight.

Since March 2022, The Data-Driven Trades has published what most of our industry keeps to itself: what contractors actually spend, what they actually close, and what's left in between. Real revenue, matched to real marketing dollars, every month. Nobody else in home services does this in public, at this depth, for this long.

So I took it seriously. I ran a meta-analysis of all 173 issues, from the first one to last week's LSA benchmarks.

Here's what four years of the data says when you read it all at once.

Channels are weather

Every channel story in the archive changed at least once.

Google LSA is today's best paid channel. Closed ROAS has sat between 6.9x and 9.2x since late 2024, and customer acquisition cost between $240 and $320. It looks like the most stable thing in contractor marketing.

It wasn't always. In 2023 the same channel swung from 18x to 3.46x. HVAC customer cost jumped from $95 to $634. In September 2024 the advice was don't depend on LSA. The calm came when the sample grew from about 20 accounts to more than 900. Averages smooth things. Individual businesses still range from 1.3x to 19.1x.

Unbranded Google search has been getting worse for four straight years. Cost per sold job went from $225 in December 2021 to $700–$800 today. Branded search is a tenth of the budget and about 40% of the revenue, which keeps the blended number looking healthy. In April 2026, the median unbranded campaign returned 0.94x. Half of them lost money.

Facebook went from "undervalued" in 2022 to a $1,614 customer acquisition cost in 2025. Organic returns dropped from 44–51x to 14–37x once you account for a change of sample. Outbound SMS posted huge numbers that shrank about tenfold as the sample grew from 3 businesses to 70.

None of this makes the data bad. It makes channels what they are: conditions, not facts. Today's best channel is a result for a given sample, a given season and a given auction.

The forecast record

SearchLight has also called customer behavior early, and said so when it missed.

In 2022, the prediction was that chat and online booking would grow from 10% of leads to 20%, and to almost half of revenue. By May 2024, non-phone share had fallen back to 21%, and the post said it plainly: I was wrong. The phone has stayed at 75–80% of leads the whole time.

I respect that line more than any chart in the archive. It's also why I'd hold today's AI story loosely. LLM-referred leads went from 0.24% of organic search to 2.15% in six months. It's real, and it's the same shape as the online-booking curve that stalled. Only 55% of contractors who get an AI lead get one again the next month. Call it a lottery that's getting bigger, not a channel yet.

The number that didn't move

Here's the finding that matters most, and it's the one nobody's chart is built to sell.

In 2022, 21% of more than 22,000 Google PPC leads became booked jobs.

In 2026, the raw book rate across all conversions is 22–24%.

Four years. A different sample, different channels, new AI answering services, new attribution products, new everything. About one lead in five still becomes a booked job. About half the revenue pipeline still never closes.

Look at where it leaks.

• Most conversions were never bookable. Only 37–40% of inbound calls and forms are real requests for work. The rest are existing customers checking on appointments, billing questions, hang-ups, and calls that never got past the phone menu. Nearly half of tracked media spend lands there.
• The bookable ones slip on the call. "Planned follow-up" is the top reason a bookable caller doesn't book: 54–70% of the misses. Roughly 60% of those people never come back.
• The booked ones stall at the estimate. Close to half of LSA revenue pipeline sits in unsold estimates, and it's worse for new customers, the ones you paid the most to get.

Every one of these is an operations problem. None of them is fixed by a better ad.

What the archive actually teaches

The industry argues about channels because channels are easy to change. You can move budget on Monday morning. You can't fix a phone tree, a follow-up habit and an estimate process by Monday morning.

So we optimize what's easy and hope it fixes what's hard. Four years of data says it doesn't. The channels swung in every direction. The leak held at the same ratio.

Here's the first-principles version: a dollar of media is worth whatever your front office does with it. A contractor booking 22% of leads and one booking 35% aren't in the same business, even if they buy from the same auction at the same price. The second can afford to outbid the first everywhere, forever.

That's the real moat in home services. It isn't the channel. It's the half hour after the phone rings, and the week after the estimate goes out.

Read it with the right eyes

One caveat, in the same spirit SearchLight writes in. The samples are SearchLight's customers, who skew mature and data-driven. Sample definitions changed often, and some year-over-year baselines were restated when they did. Several partner datasets came from the vendors being measured. Trust the direction more than the level.

That doesn't take anything away from the work. It's how you read any honest dataset. And in four years of contractor marketing data, the most honest number turns out to be the most boring one.

One in five. Still.