How to measure whether HVAC follow-up actually adds sales

Measure the added process against your normal process

Imagine hiring someone to handle unsold replacement estimates. At month-end they report calls, texts, replies and sold jobs. Those figures show activity and outcomes. They do not isolate what the added follow-up caused. Some households already planned to buy; others may have returned because your advisor called.

The practical question is whether the new process adds enough profitable work beyond business as usual to justify its cost. A control group is simply a comparable set of households that keeps your normal process while another set receives the additional follow-up.

Do not use last January as the only comparison with this July. Weather, lead sources, staffing, prices and demand can change. Running the groups at the same time reduces that problem, though small samples still leave uncertainty.

Write the rules before the first assignment

Choose eligible opportunities consistently. For example: delivered residential replacement estimates that remain open, with the required contact permissions and enough proposal validity for the planned process. Exclude sold work, opt-outs and records that fail your agreed checks before assignment. Record exclusions and reasons.

Define a sale in advance: perhaps a signed replacement agreement with a deposit received. Decide how cancellations, refunds and duplicate proposals will be handled. A booked callback is not a sold installation. Report callbacks separately so a useful conversation is not mistaken for revenue.

Choose one primary measure, usually the proportion of eligible households that sell within a defined window. Also track job value, estimated contribution, opt-outs, complaints, missing outcomes and staff time. These show whether apparent sales gains came with unacceptable costs.

Split households at random, not by advisor preference

Put each eligible household into the added-follow-up group or control by a fair random assignment, roughly half in each. Use a stored random value or a reproducible CRM assignment and freeze it. Do not let an advisor choose “promising” estimates for treatment or move a household after seeing a reply.

Keep all quotes for the same household in the same group, including good/better/best options and revisions. Otherwise one person can receive both processes and be counted more than once. A household identifier can be a consistent customer record; a pile of separate estimate IDs is not enough.

For a small operation, a spreadsheet can work. Assign once, save the assignment record and restrict edits. If you deliberately balance groups by advisor or estimate age, decide that before starting and randomize within those categories. Avoid so many categories that each contains almost no households.

The control still gets ordinary care

Document what your CSR and comfort advisors normally do. Keep that baseline in both groups. Add the tested sequence only to the treatment group. You are measuring an addition, not creating a neglected group that makes the new service look better.

Customer-initiated questions, safety concerns and service requests should be handled in either group. Log departures from the plan, including accidental extra messages to control households. Do not delete inconvenient records. Keep assigned households in the main comparison and explain significant deviations.

If your normal process changes during the test, record the date and what changed. A staffing gap or new discount can affect interpretation. A clean explanation of a messy month is more valuable than a tidy report that conceals it.

Give every household the same time to decide

A household quoted on day 29 cannot have a full 30-day outcome by the next day. Separate enrollment dates from observation dates. If you choose a 30-day outcome window after delivery, wait until each enrolled household has that full window before comparing mature results.

Report weekly progress as provisional. Show how many households have matured and how many are still waiting. For older unsold estimates enrolled at pilot start, define a consistent window from assignment and compare similar ages separately from brand-new quotes where practical.

Do not count a late sale in one group while cutting off the other earlier. If you also report a longer window, define it and apply it equally. The 30-day checklist helps preserve the dates and status changes needed.

A worked example with invented numbers

Suppose 80 eligible households receive added follow-up and 80 stay with the usual process. After equal observation windows, 24 treatment households and 20 control households meet the sale definition.

• Treatment close rate: 24 ÷ 80 = 30%.
• Control close rate: 20 ÷ 80 = 25%.
• Difference: 30% − 25% = 5 percentage points.
• Estimated incremental jobs in the treated group: 80 × 0.05 = 4.

That is an estimate of four added jobs, not a claim that all 24 sales were recovered by follow-up. The relative increase is 20%, because 5 ÷ 25 = 0.20. Label percentage points and relative percentages clearly; they describe different quantities.

Uncertainty matters. A rough independent-proportions calculation puts a 95% interval around this rate difference at about −9 to +19 percentage points. That broad range includes no effect and possible harm. This illustrative small test would not justify a confident sales-lift claim; exact methods and other designs can produce different intervals.

Compare contribution with the full cost

Use your actual job economics. Revenue is not the money left to pay for follow-up. Contribution should account for the variable job costs you define, such as equipment, direct labor, subcontract work and financing fees. Check those figures with whoever maintains your accounts.

For illustration only, four estimated extra jobs at $2,000 contribution each suggest $8,000 before follow-up costs. If service fees, staff time and messaging cost $1,500, the point estimate is $6,500 net. The uncertain job estimate makes that net figure uncertain too.

For planning, $1,500 ÷ $2,000 = 0.75 incremental job-equivalents to cover the cost. In whole-job terms, one extra job at that contribution would cover it. Do not present hypothetical break-even arithmetic as an observed return.

Use a report the owner can audit

Show assigned and mature households per group, sold counts, close rates, rate difference, estimated incremental jobs, contribution assumptions and total cost. Include cancellations, missing outcomes, channel permissions and exceptions. Keep a restricted household-level audit trail behind the summary.

A negative result may mean the extra process does not help this business. An inconclusive result means you do not yet know; it is not a win. Agree whether more comparable observations are worth the time and cost before extending anything.

Explore scenarios in the ROI calculator. If you want help running the messages and comparison, review GablePath Follow-Up. The five-touch guide describes one process to test against the work your team already does.

Frequently asked questions

The test design is practical guidance. All counts, sales rates, margins and costs in the worked examples are invented for arithmetic, not client results or industry benchmarks. No sample size guarantees a conclusive result.