Choose a baseline period that resembles the pilot’s workload and has usable outcome evidence. A convenient calendar month can be misleading when staffing, demand, marketing, or opening hours differ materially.
A baseline is a comparison design
A baseline period provides the reference against which pilot outcomes are described. It should use the same eligibility, completion, deduplication, cost, and follow-up definitions as the pilot wherever the underlying data permits.
For a proposed selection process, list candidate windows and assess weekday mix, staffed hours, service demand, locations, marketing activity, holidays, and data completeness. Record why the chosen window is appropriate and which differences remain.
Do not select the weakest historical period after seeing the pilot result. That makes the apparent improvement depend on a favourable comparison rather than a predeclared evaluation method.
Compare fictional baseline candidates
A repair office pilots in a hot-weather week. One historical week has ordinary demand, another has similar weather-driven call volume, and a third coincides with an office closure. The closure week may exaggerate recovered-call claims and should not be chosen merely because it produces the largest contrast.
If no single window is comparable, report several reference periods or limit the conclusion. The data may support a description of current performance without a strong causal before-and-after claim.
| Candidate property | Review question |
|---|---|
| Call mix | Are supported intents similarly represented? |
| Staffing | Were answer and callback resources comparable? |
| Demand conditions | Did unusual events affect volume or urgency? |
| Outcome maturity | Did enquiries have equal time to resolve? |
| Data quality | Can eligibility and completion be reconstructed? |
A transparent selection note is a useful artifact. Include window dates, available counts, known exclusions, and the definitions that could not be aligned. Do not fill missing baseline records with optimistic assumptions.
Balance length and relevance
A longer baseline can stabilise counts while mixing obsolete policies or staffing patterns. A short recent baseline can be more comparable but highly variable. Show that tradeoff and retain raw counts beside rates.
Where feasible, a concurrent comparison or controlled rollout can strengthen interpretation, but it introduces its own allocation and contamination concerns. A simple historical baseline remains useful when its limits are stated honestly.
This baseline method is proposed. The HVAC case study demonstrates a request workflow without measured before-and-after results. Pair it with same-definition outcome comparisons.
Baseline period FAQ
Is the previous month always the best baseline?
No. Review comparability and data quality rather than assuming proximity is enough.
Can several periods be shown?
Yes. Multiple transparent references can reveal how sensitive the interpretation is to the selected window.
What if historical outcome data is missing?
State the limitation and focus on observed pilot outcomes or collect a prospective baseline before making stronger claims.