Before and After: A Practical Measurement Plan for Local Marketing

Start with one decision question

“Did marketing work?” is not a measurement question. A useful version names the audience, action, time, and decision: “After the connected visual route was published, did the share of location-page visitors who submitted a qualified venue inquiry change enough to justify maintaining and refreshing the route?” This does not assume the route caused the change. It identifies the behavior to observe and the economic decision the observation will inform.

Write the intervention precisely. Record when the media became publicly reachable, which profile or pages changed, what customer questions the new experience was intended to answer, and what remained untouched. If photos, categories, hours, website copy, paid ads, pricing, or call handling also change, record each date. A before/after study with a bundle of simultaneous interventions can evaluate the bundle’s timing, but it cannot isolate one component.

• Decision: what will the business do differently after seeing the result?
• Population: which profile, location, audience, and service are in scope?
• Intervention: what changed, where, and at what time?
• Outcome: which observable behavior is closest to the intended effect?

Preserve a comparable baseline

Choose enough pre-intervention data to reveal the business’s weekly cycle, trend, and obvious seasonal variation, then compare it with a post period defined by the same calendar logic. A restaurant should not compare a holiday weekend with ordinary weekdays and call the difference lift. A venue should not compare peak inquiry season with a quiet month. Use the longest reliable history that still reflects the current offer and operations, and show daily or weekly observations rather than only two totals.

Do not tune the start and end dates after seeing a favorable spike. Decide in advance whether launch day belongs to a transition window, when the first formal read will occur, and which outages or closures justify exclusion. If the business has low event counts, aggregation may make the series more readable, but it also reduces the number of observations. Report counts alongside rates and avoid drawing a strong conclusion from a handful of events merely because the percentage change looks large.

• Align weekdays, open days, holidays, and known seasonal periods.
• Keep raw exports so later platform displays do not rewrite the baseline.
• Document missing data, tracking changes, and exclusions before analysis.

Measure the chain, not one dashboard number

Build a simple measurement ladder: eligible profile or page opportunities; qualified actions such as calls, forms, bookings, or direction requests; accepted leads; customers served; and contribution margin. Google states that Business Profile performance can include views, searches, and supported interactions, with metric availability varying by business. Export the available profile data, but join it to the business’s own call, form, booking, point-of-sale, or customer records where possible. A click is an action, not proof of a sale.¹

Use both counts and conversion rates. If qualified inquiries rise only because profile exposure doubled, the qualified-action rate may be unchanged. If the action rate rises while accepted leads fall, the definition of qualification or the audience mix may have shifted. If customers rise but contribution falls, the new mix may be buying lower-margin services. Following the chain shows where momentum appears and where it gets bogged down.

• Exposure: who had a plausible chance to encounter the change?
• Action: what did those people do next?
• Quality: which actions met a predefined qualification rule?
• Economics: what contribution followed after variable service cost?

Report the before/after pattern honestly

For each metric, show the pre-period level and trend, the post-period level and trend, the absolute difference, the rate difference where a denominator exists, and the underlying event counts. Add an annotated time plot marking the launch, closures, promotions, review bursts, staffing changes, tracking repairs, and local events. This turns “up 30%” into a trace a reader can interrogate. If variability swamps the apparent change, say the result is inconclusive.

Interrupted time-series analysis can estimate changes in level and trend around a defined intervention, while accounting for features such as underlying trend and seasonality when the data support them. Published tutorials emphasize design, model specification, and threats to validity—not merely fitting a line before and after a date. For a local business, the value of the framework is discipline: inspect the pre-trend, model recurring structure, test residual behavior, and avoid presenting a discontinuity as causal when another event occurred at the same time.³

Strengthen the counterfactual when the decision warrants it

The missing object in every before/after comparison is the counterfactual: the outcome that would have occurred during the post period without the change. A staged randomized rollout across comparable locations can estimate it most directly when operationally feasible. Otherwise, a similar untreated location, unaffected service line, or stable upstream metric may provide context. The comparison must share the forces that drive the outcome yet remain untouched by the intervention; finding one after the fact because it produces the desired result defeats the design.

Bayesian structural time-series methods can construct a synthetic control from contemporaneous series and quantify uncertainty around the estimated impact. Their usefulness depends on a stable pre-intervention relationship and on control series that were not themselves affected. The method does not manufacture identification from weak inputs. For a small business without adequate volume or credible controls, a transparent descriptive report is stronger than an elaborate causal label attached to fragile data.²

• Best feasible: randomized timing across comparable units.
• Stronger observational: preplanned, unaffected comparison series.
• Descriptive: calendar-aligned before/after with complete annotations.

Close with a decision log, not a victory slide

Write the conclusion in four parts: what changed in the observed series; how uncertain the estimate is; which alternative explanations remain; and what the business will do next. “Qualified inquiry rate rose after launch, but a new promotion began in the same week, so attribution is unresolved; keep the route live and repeat the comparison after the promotion ends” is more useful than “the tour drove leads.” It preserves what is known and defines the next learning step.

Keep the original hypothesis, raw data, exclusions, and code or spreadsheet formulas with the report. If the outcome is flat, inspect whether the intervention reached enough eligible people and whether the targeted customer uncertainty was real before concluding the medium never works. If the outcome is favorable, resist expanding the claim beyond the measured location, offer, and period. Comparable cycles with adequate data and improved controls can narrow uncertainty; repeated descriptive comparisons do not by themselves strengthen causal attribution.