A baseline is a record of how a defined outcome was measured before a change. Without the same definitions and sources later, the comparison can describe two different systems rather than improvement.
Freeze the measurement contract
Before changing a campaign, page, or workflow, write down the outcome, event definition, source of truth, time window, market, segment, and known limitations. Save the report or export with a timestamp and note what else was changing: price, stock, season, staffing, tracking, or offer.
For a lead journey, a useful baseline may separate: landing sessions, contact clicks, messages received, qualified inquiries, appointments, and completed services. If the team can verify only the first two, say so. Do not label the click count “leads” to make the report look complete.
Choose a comparison that fits the decision
Use equivalent periods where possible and account for seasonality or operating-day differences. Keep the same event logic and filters. If tracking changed midway, mark the break and avoid calculating a seamless before/after percentage. A baseline is not a causal control group; it tells you what was observed before, not what would have happened without the change.
Make it reproducible
Record report link, filter set, event version, owner, and extraction date. After the intervention, rerun the same view and add a separate note for any changed definition. If data is incomplete, preserve the gap instead of backfilling estimates as observed fact.
Optional AI-assisted work brief — Baseline specification
Define the metric, population, source, timezone, attribution window, known gaps, and date range before a campaign or site change. Check seasonality and operational interruptions. Preserve a snapshot and query definition so the comparison can be recreated. Separate observed historical performance from the target the team hopes to reach.
Worked example — Freeze the comparison before launch
Record four complete weeks if that matches the business cycle, define the conversion event and timezone, and annotate holidays, outage days, and campaign changes. Save the source query and exported totals. A baseline does not tell you what will happen next; it prevents a later report from quietly changing the definition of “lead” or the dates compared.
AI disclosure: AI tools assisted with drafting. The measurement path is illustrative.