Forty more jobs. Ten left after the comparison.

TL;DR

• HVAC marketing growth from 100 to 140 sold jobs is an observed increase of 40 jobs. If a defensible comparison goes from 100 to 130, a same-change calculation leaves 10 jobs, conditional on that comparison. It does not prove the agency caused ten jobs.
• The comparison needs a reason to reflect what would have happened without the change. A nearby market or another trade is not automatically a control.
• Assuming the same number of additional jobs and the same percentage growth are different models. With different starting volumes, they can give opposite signs.
• Use the free growth calculator and review worksheet to expose the assumption, then inspect earlier records and other changes. Four totals alone cannot measure uncertainty.

What does a 40% HVAC marketing growth claim establish?

The account report has a simple table: one period produced 100 sold jobs; the next produced 140. The agency changed between them. The headline says growth is 40%.

That percentage is correct if both cells use the same job definition and comparable periods. It describes the change in the shop's records. It does not separate campaign work from demand, phone coverage, crew capacity or anything else that changed at the same time.

The first question is therefore about the comparison, before arguing over the percentage. What result would the shop plausibly have seen without this change? You cannot observe that missing outcome directly. You can propose a reference and explain why it belongs beside the shop.

The agencies hub covers the information a report should disclose. This guide adds a calculation owners can inspect when a before-and-after number is presented as credit for a particular change.

What happens when another group also gets busier?

Consider two invented groups over the same before and after windows. The changed group receives the change being reviewed. The comparison group does not. Every job has the same definition and follow-up cutoff. These are teaching numbers, not client results or an industry benchmark.

Suppose the changed group would have gained the same 30 jobs without the change. Its expected count would then be 100 + 30 = 130. The difference between its observed 140 and that expected 130 is ten jobs.

The World Bank's explanation of difference-in-differences, checked October 10, 2026, describes comparing changes across groups. The key requirement is that the comparison represents the change the treated group would have experienced without the intervention. Here, that requirement is a hypothesis to defend, not a fact supplied by the table.

The calculator therefore calls ten jobs a conditional comparison difference. Another change affecting only the first group could explain some or all of it. An unrepresentative comparison could distort it.

Keep the denominator beside any percentage. Ten jobs is 10% of the original 100, but about 7.69% of the expected 130. Neither should be labeled a proven marketing lift from these four counts.

Why can the same records give a positive and a negative answer?

Now keep the changed group at 100 → 140, but use a larger comparison at 400 → 520. The comparison gains 120 jobs and grows 30%.

The arithmetic is correct in both rows. Their assumptions are different. One carries an absolute change across groups; the other carries a growth ratio. Group scale makes the difference visible.

The World Bank's discussion of functional form and pre-trends, checked October 10, 2026, explains why the choice between absolute and proportional trends needs justification. Earlier trends can help examine that choice, but passing a pre-trend test does not establish that the future counterfactual assumption is valid.

Do not choose +10 because it makes the agency look good, or −80 because it supports leaving. Start with why the groups differ in size and what moved them before the change. If neither assumption fits the records, neither result deserves a performance claim.

The calculator's unequal-group example reproduces both rows. It makes the disagreement visible rather than silently normalizing away the larger group.

What makes a comparison worth considering?

Write the proposed groups down before examining the result. Keep their membership rules fixed: the service areas, job types or locations should not move between cells to improve the story.

A group may share weather and customer demand with the changed group but have a different crew, answering schedule or job mix. A neighboring service area may contain customers reached by both campaigns. Technicians may move between branches. A control that receives some of the change is not untouched.

Record these differences beside the totals. In an account review, the practical questions are concrete: did both groups add technicians, did one extend phone hours, did prices change, did budgets move, and did the CRM start counting a different outcome?

Also inspect several earlier periods. If one group was already gaining jobs faster, a simple same-change story needs more work. A chart of earlier levels can show whether the two groups are similar enough to discuss at all. It cannot guarantee that they remain comparable after the change.

For HVAC, put actual period weather beside the records. The weather and capacity review gives that check; the seasonality calendar supplies a historical planning baseline, not the weather that actually occurred during your test.

Is this the same thing as a Google Conversion Lift study?

No. Google's lift-study guide, checked October 10, 2026, describes a controlled study comparing groups shown and not shown ads. It distinguishes conversion, brand and search lift. The free calculator here does not assign groups, withhold advertising or run that study.

It also does not read a Google Ads conversion column. Use the outcome relevant to the claim: inquiries, booked visits and sold jobs are different records. A marketing claim about sold work should not switch to phone calls in the after period because that number is easier to obtain.

If a planned experiment is warranted, define its assignment, outcome, timing, overlap risks and analysis before starting. This article provides review arithmetic, not instructions to cut a shop's advertising to create a holdout.

What should leave the next account review?

Bring three things: the four-count table, the rule behind the comparison, and a dated list of other changes. Download the scenario CSV from the calculator so the arithmetic can be repeated. Keep the worksheet with it; the CSV cannot explain why a comparison was chosen.

The result may be a narrower statement than the opening headline. In the equal-group example, the changed group gained 40 jobs while the proposed comparison gained 30; a same-change scenario leaves ten jobs, with cause and uncertainty unresolved.

That statement gives the next review a useful question: what evidence could support or weaken the comparison? A bare growth percentage gives it only an argument over credit. Use the reporting red flags guide to check the rest of the report, and the HVAC marketing page for the service context.

What this doesn't cover

• Uncertainty or significance. Four aggregate counts are insufficient for the uncertainty calculation this tool would need. It prints no confidence interval or significance label.
• A complete causal design. Selection, spillovers, changing composition and simultaneous changes can invalidate a comparison. The calculator cannot detect them.
• Profit or budget allocation. More jobs may have a different service mix and contribution. Use job economics after dispatch for that separate question.

Check one growth claim in your own records

1. Name the specific change and locate its implementation date.
2. Fix the outcome definition and give all records the same completion cutoff.
3. Select a proposed comparison for a stated reason, before choosing a flattering result.
4. Inspect several earlier periods, group scale, spend, capacity, weather and overlap.
5. Enter the four counts, compare both assumptions and retain undefined results as undefined.
6. Save the CSV and worksheet with the records. State what is observed, what is conditional and what remains unknown.

FAQ

Can I use last year as the control group?

Last year can add context, but it is not automatically a control. Prices, weather, auctions, staff and measurement can change across years. Document why that comparison fits the claim and which differences remain.

Should I always use percentage growth for different-sized branches?

No. The percentage scenario assumes a common growth ratio; different starting volumes do not establish it. Examine earlier patterns and the reasons for the volume difference before selecting either model.

What if the before count is zero?

The observed percentage is undefined, and the calculator leaves the proportional scenario undefined when either before count is zero. An absolute-change scenario may still be arithmetically possible, but a zero baseline needs explanation before treating it as a useful comparison.

Does a positive comparison difference justify more spend?

It does not establish profitable incremental work. Review the comparison's credibility, uncertainty, job mix and costs before making a budget decision. A conditional job difference alone cannot supply those answers.

Sources

• The often unspoken assumptions behind the difference-in-difference estimator in practice — World Bank Development Impact, read October 10, 2026.
• Revisiting the Difference-in-Differences Parallel Trends Assumption, Part I Pre-Trend Testing — World Bank Development Impact, read October 10, 2026.
• About lift studies — Google Ads Help, read October 10, 2026.

Lonta AI manages home service marketing against the shop's operating records. The free written strategy identifies the account and measurement questions to resolve first.