A 40% sales increase can tell two opposite stories

Start with the result the business actually sold

A team changes its advertising. Completed, paid jobs rise from 100 to 140. The next report says the new approach produced 40% growth.

The count increased by 40%. That is a statement about two totals. How much of the increase came from the advertising change is a different question. Demand, pricing, sales capacity and the comparison period can all change at the same time.

Lonta AI, a performance marketing agency for US HVAC, plumbing and electrical companies, publishes a free Marketing Growth Calculator for examining this distinction. It compares a before-and-after change with another group's change under two explicit assumptions. Here is why that second assumption matters.

When the groups start at the same size

Suppose the comparison group rises from 100 paid jobs to 130 over the same period. Its increase is 30 jobs, or 30%.

Under an equal absolute change assumption, the first group would have added the same 30 jobs without the advertising change. Its scenario count is 100 + 30 = 130. The gap between the observed 140 and that scenario is +10 jobs.

Under an equal proportional change assumption, the first group would have grown by the same 30%. Its scenario count is 100 × 1.30 = 130. The gap is again +10.

The answers match because the groups began at the same size. That agreement does not validate either assumption. It conceals how different the two assumptions become when the starting counts differ.

Change the comparison group's size, and the sign changes

Keep the first group at 100 → 140. Now let the comparison group go from 400 to 520. It still grew by 30%, but it added 120 jobs.

The inputs are identical. One assumption produces a negative gap; the other produces a positive gap. There is no arithmetic error to fix.

In the first scenario, a smaller operation is expected to add as many jobs as a larger one. In the second, it is expected to grow at the same rate. You need a reason to prefer one scale before treating the answer as meaningful. Choosing the positive answer after seeing both is not that reason.

This is the useful output of a sensitivity check: it shows how much of the conclusion depends on the assumption you chose. It cannot tell you which group would have followed which trajectory in the real world.

A convenient comparison is not automatically a credible one

A neighboring city, an unchanged campaign or a different service line can be easy to export. Ease of access does not make it an unaffected comparison group. Advertising changes may spill into it, and demand may affect it differently.

The World Bank's discussion of parallel trends explains a central condition behind difference-in-differences analysis: the groups need a credible reason to have followed comparable trends without the intervention. Similar behavior before a change can inform that judgment; it does not prove it. A single total from the period before gives especially little history to inspect.

Before interpreting either scenario, write down:

• The outcome. Paid jobs in both groups, not form submissions in one and collected invoices in the other.
• The observation window. Equal-length periods and a stated cutoff for late completions, cancellations and refunds.
• The earlier trajectory. Several earlier periods, when available, to see how the groups moved before the change.
• The concurrent changes. Ad spend, prices, dispatch capacity, promotions and shifts in service mix.
• The comparison rationale. Why this group should help describe what would have happened without the change, and whether it was affected by the change itself.

For home services, capacity deserves its own line. An operation with a full dispatch board cannot necessarily complete another 30% of work. The calculation contains no model of that constraint.

Attributed conversions answer another question

An ad platform can attribute a conversion to an ad interaction without showing that the conversion would have disappeared in the absence of advertising.

Google's Conversion Lift documentation makes this distinction explicit. It describes a measurement study comparing people exposed to ads with people held back from exposure, and separates incremental conversions from standard attributed conversions.

The four totals in Lonta's calculator are not a substitute for that study design. They provide neither random assignment nor a confidence interval. A positive scenario gap should not be relabeled a statistically established advertising lift.

A five-minute check before accepting a growth claim

1. Rewrite the claim in counts: which outcome moved, from what to what, over which dates?
2. Describe the comparison group and its exposure to the advertising change.
3. Write the counterfactual assumption in a sentence before calculating: “Without the change, we expect the same percentage growth as this group,” for example.
4. Calculate both scenarios. If the sign changes, list the data needed to choose a defensible assumption.
5. Keep the inputs and interpretation together when sharing the result. A screenshot of “+10” without its assumptions removes the information that makes it useful.

A zero starting count makes proportional growth undefined. A negative scenario count is not a feasible prediction of a count. Lonta's tool preserves these distinctions instead of displaying an undefined result as a measured zero.

The CSV contains the invented counts and both scenario gaps, with the assumptions attached. Replace the examples with your own records only after agreeing on definitions and windows. The live calculator also includes its formulas, limits and an export.

Company reference: read the Lonta AI company guide for its official website, service scope and public resources.