Ren Álvarez has an unfinished email to her agency on one screen and nine idle install-crew days on the other. "The leads are bad. I need a better agency." That is a reasonable suspicion after your best lead month leaves crews waiting. Changing agencies does not diagnose the failure.
Ren is an illustrative owner, not a client case study. Her Tampa business has nine technicians, two dedicated install crews, and roughly $18,000 in monthly marketing spend. June's report showed 94 inquiries and a 14% drop in cost per lead.
On Sunday, she matched that report to dispatch. Sixty-one inquiries were service intent. Thirty-three were install intent. Twenty-two install prospects had not received a callback within an hour.
Her draft email asks for better targeting. The two screens suggest nine better questions first.
What would make Ren's best month a win?
An agency audit should tell you which part of the system deserves confidence, which part needs repair, and who owns that repair. It should not begin with the assumption that every unbooked inquiry proves bad marketing.
The agency owns the work in its agreement: targeting, campaign management, destinations, measurement, reporting, and the other deliverables you actually commissioned. Your operation owns truthful availability, intake coverage, sales follow-up, dispatch, and delivery. Some handoffs require both.
An accurate ad can still send a prospect to a broken form. A working form can still feed an unattended inbox. A well-handled inquiry can still fail to book because the next available installation is too far away.
You need evidence that separates those cases. That is the operating chain in how HVAC marketing actually works.
Bring one recent intake cohort to the review: inquiries first received in a defined period, followed through a stated observation date. Have the agency bring acquisition records and your office bring outcomes. Keep older open proposals visible separately.
Ren had 94 inquiries. Reading that as 94 useful install opportunities was the error. The audit begins by asking what those arrivals were supposed to become.
Which three questions test the goal and the definitions?
1. Which available work are we trying to fill?
Ask for the service type, territory, available dates, and target booked work behind the campaign. Supply the agency with your actual openings. A goal of ten additional repair appointments differs from two replacement consultations for a future install window.
Use your team's true weekly capacity to establish the boundary. The agency should explain how its campaign reflects it. It cannot infer a technician's availability from an ad dashboard.
2. What costs are included in each reported number?
Request media spend, management fees, tracking costs, and other included costs on separate lines. Identify which are allocated to service, install, or the whole program. Ask how shared costs are allocated and keep that rule consistent.
Ad-spend-only cost per lead can help assess a campaign. Total program cost per completed job answers another question. A report should label both rather than quietly change the numerator.
3. What exactly counts as a lead or conversion?
Ask for the configured action, exclusions, and duplicate rule. Does the total count call-button taps, connected calls, form submissions, distinct qualified inquiries, or appointments?
Google Ads describes conversions as actions the advertiser chooses to value, including phone calls and website activity. The word alone does not certify a booking. Ask to see the actual configuration and a few matched records.
Ren keeps the green slide. It may be measuring its chosen action correctly. She needs to know whether that action matches the outcome she has been reading into it.
Which three questions expose what happened after the inquiry?
4. Where did the distinct inquiries come from, and what did they want?
Request source, campaign where known, initial service or install intent, and new-versus-existing customer status. Separate spam, suppliers, duplicates, and out-of-area contacts from eligible opportunities.
Ask how the source was established. A tracked number, tagged form, and customer's recollection are different evidence. Keep unknowns visible instead of assigning them to the channel that makes the report look best.
5. What happened between first contact and the appointment?
Match a sample across call logs, forms, office notes, and the schedule. Ask when the inquiry arrived, who responded, whether contact was made, what appointment was offered, and why it did not book.
The agency should expose the handoff it can see. Your office must supply the dispositions it alone knows. Use the call handling checklist to turn a missing step into an operating task.
6. What did the same cohort cost per booked and completed job?
Ask for one intake period followed to one stated cutoff, separated by work type. Count cancellations, still-open opportunities, completed jobs, and collected revenue distinctly. Do not compare this month's spend with whichever old proposals happened to close this month.
The recurring model in leads don't matter if they don't book uses 80 leads at $250 each: $20,000 divided by 23 bookings is about $870 per booked job. That is a separate modeled $20,000 scenario, not Ren's roughly $18,000 profile budget or her 94-inquiry June.
Ren's own illustrative June cost per booked install exceeded $2,100. The 22 slow callbacks do not prove every one was lost. They identify records the agency and office need to investigate together.
Which three questions test the agency's next decision?
7. What job does each channel perform, and what evidence supports that role?
Ask which campaigns capture immediate demand, which pages support discovery, and which proof helps a homeowner decide. Request work completed, leading indicators, and outcomes appropriate to the channel's time horizon.
Do not demand that a new service page immediately produce the same result as paid search. Do not accept "organic takes time" as an indefinite exemption from showing work. Ads and organic visibility can support the same customer without each owning the entire sale.
8. Can we inspect the accounts, assets, work and records behind the report?
Request the account map, business access, current tracking plan, deliverables, change log, and an export you can reconcile. Establish which assets your business owns, which services are licensed, and how continuity works under your actual agreement.
A glossy presentation should have underlying evidence. An agency can explain a platform limitation or contractual restriction specifically. "The dashboard is all you need" does not explain why a claimed result cannot be checked.
9. What changes next, who owns it, and when will we evaluate it?
Ask for the diagnosis, proposed change, expected signal, responsible person, and review date. Decide in advance what would support continuing, adjusting, or stopping the test. Use a window suited to your volume and buying cycle, rather than inventing a universal seven-day deadline.
An agency may need your team to confirm appointments or update lost reasons first. Put that dependency in the plan. Accountability works in both directions.
Ren's next email can ask for an explanation without pretending her office is outside the system. The 33 install inquiries need campaign evidence and an operational owner.
What can you send before the next review?
You can copy this email and replace the review period and dates. Attach usable capacity and office outcomes rather than asking the agency to discover them from scratch.
Subject: Evidence for our next marketing review
Hi team,
For our next review, please use inquiries first received during [period], followed through [cutoff date]. Separate service and install, and identify anything still unknown. Please send a concise answer and supporting export for these nine questions:
- Which work, service area, available dates, and booking goals are our campaigns intended to support?
- Which costs are included in the reported acquisition figures, and how are shared costs allocated?
- What counts as a lead or conversion, and how are duplicates and ineligible contacts excluded?
- What source evidence and initial intent are attached to each distinct inquiry?
- What happened from first contact to appointment, including response times and lost reasons?
- What did this same cohort cost per booked and completed job, with cancellations and open work shown separately?
- What role does each channel perform, what work was delivered, and what evidence will show progress?
- How can we inspect the relevant accounts, tracking setup, assets, changes, and underlying records?
- What should change next, who owns each step, and when will we review the result?
We will bring current capacity and the office outcomes you cannot observe directly. Please flag missing information and agree a date for filling the gaps. Let's leave the review with one shared plan and named owners.
Thanks, [Name]
Ren does not have to send the email in anger. It asks for a report both sides can work from, with enough detail to find the next useful decision.
How should you judge the answer?
A useful answer makes its assumptions visible. You can follow a sample inquiry from its source to the office outcome, see the cost basis, and understand what remains unresolved. The agency accepts its mistakes and identifies your team's missing handoffs specifically.
An incomplete answer may still be recoverable. A broken integration needs a repair owner and a date. A replacement cohort that has not matured needs a later cutoff. Neither is a reason to manufacture certainty today.
Persistent refusal to define conversions, show agreed work, or reconcile a sample is a different problem. Document the missing evidence, request a bounded correction, and judge the response against your agreement and business needs. Do not confuse a disappointing result with proof of misconduct.
The useful agency report ends with an accountable decision, not a more convincing shade of green.
In Ren's illustrative story, the email replaces the one she started on Sunday. The agency still has to answer for targeting, spending, and reporting. Her office still has to explain the callbacks. Both now have the same 33 install inquiries in front of them.
She has not declared the agency innocent or guilty. She has made it possible to find out what needs to change.