The Front Door Integration Test

In July 2026 I placed 55 mystery-shop calls to 48 residential service brands, HVAC, plumbing and roofing, owned by roughly 20 private-equity sponsors. Same method, same scoring rubric, every call transcribed and scored the same way. I called the number each brand’s marketing budget exists to make ring, and I presented as the caller every one of these businesses says it wants: a homeowner with a real problem, money to spend, and two or three companies on the list.

The phones mostly get answered, and the selling stops there. The layer between the answered call and the booked job does not exist as a system. What I found instead were scattered individual habits: one written recap, two human save attempts, in 55 calls. Administrative integration had shipped at most of these platforms. Customer-facing integration mostly had not, and the caller is the only person positioned to notice.

Exhibit · The cliff, observed

From forty-nine answered calls to zero referrals, in four steps.

49/5589%
answered by a live voice
The phone system did its job.
Paid software ends here
8/4916%
tried to hold the lead
At the moment I deferred. Six were machines.
1/551.8%
put anything in writing
A recap received before I hung up.
0/550%
referred a sibling brand
Including all 13 calls where I volunteered a second problem.

Observed July 2026. The first number runs on software the platform bought; the next three are workflow. Denominators vary because six calls never reached a voice, and some stages trigger only when the call creates their condition.

“We only handle hot water tanks or tankless water heaters, but I don’t have any plumbers on staff for anything else. So you’d have to call a plumber for that, unfortunately.”

CSR at an HVAC brand whose platform sells plumbing under other flagsVerbatim · Finding 01 · July 2026

The five findings

Each one reads standalone, with its denominator in view.

The cross-sell thesis lacks front-door machinery
In 55 calls, including 13 where I volunteered a second problem in the home unprompted, no human or AI agent made a live referral to a sibling brand in any phrasing. Three of the 13 got one-trip handling; the rest drew a second visit, a second fee, or a competitor's name.
0/55Observed
Same platform, four different front doors
Four sibling brands under one sponsor, shopped inside 48 hours with the same scenario, held an identical $169 diagnostic fee and diverged on everything after it. One texted a recap mid-call. One captured no contact detail at all.
n = 4 brandsObserved
Scripted agents attempt the save more consistently
At the moment I deferred, scripted AI agents made a save attempt in 6 of 13 agent-run calls. Human answerers made one in 2 of 33. Fisher's exact p = 0.004; the direction holds, the multiple carries wide uncertainty.
6/13 vs 2/33Observed
Replacement intent is not recognized at intake
Five calls presented a competitor's written $8,500 replacement quote, two with an offer to email the proposal. Zero platforms asked to see it, and soonest visits ran two to ten days for a buyer deciding that weekend.
0/5Observed
AI scales configuration, including configuration errors
One agent accepted an email address announced out loud as fake, and the appointment flow continued as designed. Another walked its own empty calendar into September without escalating to a human being.
16/55 AI-frontedObserved

On finding 01, the fair objection: plenty of sponsors keep brands apart on purpose, for pricing segmentation or licensing lines across trades and states. Where the separation is deliberate, the zero is a strategy choice and this study prices what it costs. Where it isn’t, the zero is the integration gap showing up at the front door. You know which one you are running; the caller hears the same thing either way, which is that the platform sells one trade.

The instrument

Eight stages, one call. The selling stops at stage four.

Exhibit · The eight-stage intake rubric, observed Observed

Answer
49/55
Reached a live voice. Six dead ends, all at brands with live paid media that week.
Qualify
no score
Not scored: the rubric has no clean denominator here, because what counts as a qualifying question varies by trade. Read qualitatively it was the strongest stage in the sample, though one weekend emergency still ended without a phone number taken.
Price
40/49
Stated the visit fee, or free. Nine could not price a visit when asked.
Paid software ends here · workflow begins
Hold
8/49
Tried to keep the caller at the moment they deferred. Six of the eight were scripted AI agents.
Capture
13/49
Asked for an email. One address announced out loud as fake was accepted.
Recap
1/55
A written recap received before hang-up. The best call in the sample, and the only one.
Follow-up
2 obs.
Two promised callbacks observably arrived. The rest fell outside the study window and are unscored.
Route
0/55
Live referrals to a sibling brand. Not one, in any phrasing.

Observed incidence, 55 mystery-shop calls to 48 PE-owned brands, July 2026. Denominators vary because six calls never reached a voice and some stages only trigger in certain scenarios. The first three stages run on software these platforms already bought. The last five are habits, scripts, and configuration.

Method · limitations first

You diligence companies for a living, so the weaknesses go first.

It cannot tell you close rates.: The shopper never books, so nothing in this study is a conversion claim. I measured whether each brand did the things that close callers: the ask, the hold, the capture, the recap, the follow-up.
Per-brand reads are directional.: Most brands got one call; a handful got two or three. One call proves nothing about one brand. Fifty-five calls with the same misses repeating across 48 brands and roughly 20 sponsors is a pattern.
July is peak season, and that cuts both ways.: A brand with a full board has a rational reason to let a deferring caller go, so some of what I scored is capacity rationing rather than broken intake. Slow soonest-visit dates belong in that column. The stages this study leans on do not: asking for an email, putting the visit in writing, naming the sibling brand that does the work. None of those consume a truck slot. Replacement intent sits outside the argument too, because an $8,500 install runs on install crews, and install revenue is the thesis the platform was bought on.
The economics are modeled, never measured.: No bookings were completed, so the economic section is a scenario model with disclosed assumptions. Confidence intervals are published on every aggregate count, including the two that read zero.
How the calls were run.: Calls went to each brand’s published consumer line between July 16 and 25, 2026, across dayparts including evenings and a weekend. A small cast of homeowner personas ran four standard scenarios, and every call was scored on the same eight stages regardless of scenario. Personas used shopper-controlled phone numbers and inboxes. No bookings were confirmed, no trucks rolled, and no free work was extracted. Brands and sponsors are anonymized throughout, and quotes are verbatim, cleaned only where the transcription engine mangled a word.

The full method detail and ethics notes are in the appendix. This is an annual series: same rubric, every July, so the zero in finding one has somewhere to move. Ask for the July 2027 edition, one email, no list.

What to do Monday

Ten questions. Seven cost an hour of listening to your own calls.

Listen one hour · five recorded calls per brand

Does anyone ask for the caller's email address?This sample: 13 of 49
When the caller defers, is the next sentence a save attempt, or “no worries”?This sample: 8 of 49 attempted
Can the person answering state the price of a visit, and whether it credits against the work?This sample: 9 of 49 could not
When a second problem comes up mid-call, does it get one trip, a second fee, or “call someone else”?This sample: 3 of 13 got one trip
Does anything arrive in writing after the call?This sample: 1 of 55
When the caller holds a competitor's quote, does anyone ask to see it?This sample: 0 of 5
Call your own after-hours line tonight. Can the voice that answers quote the fee, book a job, or commit to a callback with a time on it?This sample: mostly no

Pull three reports · formulas published in full

Call-to-booked-job, by brand, on one definition across every brand. If the definitions differ by brand, that finding outranks the number.
Speed to answer and true abandonment, by brand and daypart, with short abandons excluded before the rate means anything.
Membership attach, where eligibility is written down per brand and existing members are out of the denominator.

If the seven listening checks come back clean at your platform, close the report and keep the hour. In this sample of 48 brands, none would have come back clean.

ZTS AdvisoryField study 01July 2026

The front door is the integration test. Most platforms in this sample have not yet taken it, and the caller can tell.

Download the report (PDF)Book an operating review · 30 min

ZTS runs revenue-capture work inside PE-backed residential service platforms, hands-on in ServiceTitan, HubSpot, and the CCaaS layer on top of them. This study points the same method outward.ztsadvisory.com

The Front Door Integration Test

From forty-nine answered calls to zero referrals, in four steps.

answered by a live voice

tried to hold the lead

put anything in writing

referred a sibling brand

Each one reads standalone, with its denominator in view.

The cross-sell thesis lacks front-door machinery

Same platform, four different front doors

Scripted agents attempt the save more consistently

Replacement intent is not recognized at intake

AI scales configuration, including configuration errors

Eight stages, one call. The selling stops at stage four.

Answer

Qualify

Price

Hold

Capture

Recap

Follow-up

Route

Ten questions. Seven cost an hour of listening to your own calls.