Take the number before anyone builds
How to baseline one service-business workflow in two weeks, which measures matter, and why saved time alone is not a result.
In short: Before anyone builds, baseline one workflow for about two weeks. Pick measures that show revenue captured, speed through the path, or capacity returned. "We saved time" without a number is not a result you can manage.
Part of The Owner's Guide to Buying AI Implementation.
The operating situation
A team is excited. A vendor is ready. Someone says the automation will "save hours" and "boost conversion." Six weeks later nobody can say what changed. The pilot is declared a success in a meeting and quietly abandoned in the operation.
The failure started before the first line of configuration. There was no baseline, no single owner of the metric, and no agreement on what would count as proof.
What to measure
Choose measures that match the leak:
| If the leak is... | Baseline candidates |
|---|---|
| Missed or slow first response | Speed to first human or system response; contacts recovered after hours; bookings from recovered contacts |
| Weak follow-up on estimates or proposals | Open estimate aging; follow-up completion rate; close rate on quoted work; days quote-to-yes |
| Intake that stalls | Inquiry-to-consult time; consult show rate; intake-to-signed engagement rate; days to complete documents |
| Closeout and cash | Days job-complete-to-invoice; percent of jobs with complete packets; days invoice-to-paid |
| Owner as bottleneck | Hours the owner spends on the workflow per week; decisions waiting on the owner; cycle time of those decisions |
Pick one primary measure and one or two supporting measures. More than three and the project loses focus.
How to baseline in two weeks
- Freeze the workflow definition. Same start event, same end event, same exclusions.
- Pull what you already have. Phone reports, CRM stages, calendar data, invoice dates, inbox tags. Imperfect history beats no history.
- Fill gaps with a simple log. If the system cannot count it, have the operating owner tally it daily for ten business days.
- Separate volume from rate. Know how many opportunities entered and what share moved forward.
- Write the assumptions. Seasonality, marketing spikes, staffing gaps. You will need them when you read the after number.
- Agree the review date. When the build is live, you re-measure the same way, not a new way that flatters the project.
Two weeks is enough for most recurring service workflows to show a pattern. If volume is very low, extend the window rather than inventing precision.
Example: trades (estimates and recovered demand)
Example, not a client result. A roofing or HVAC company wants fewer lost leads and faster movement on quotes. Before build, they baseline for two weeks: inbound leads by source, median minutes to first response, percent of leads that reach inspection or estimate, open estimates older than seven and fourteen days, and close rate on estimates issued in the period.
After a missed-call recovery and estimate follow-up system ships, they re-run the same report. The question is not whether the team "liked the AI." The question is whether response got faster, more estimates were worked, and more quoted work closed without adding office headcount. Related pages: Roofing, HVAC, Service Companies.
Example: professional services (inquiry to engagement)
Example, not a client result. A professional firm wants inquiry response and document collection under control. Before build, they baseline: median hours inquiry-to-first-response, percent of inquiries that book a consult, show rate, days consult-to-engagement-sent, percent of engagements signed, and median days to complete the document checklist on new matters.
After intake and document-chasing workflows ship, the same definitions get re-measured. Staff may feel less harried. That feeling is useful. It is still not the result until the rates and cycle times move. Related pages: Professional Services, Attorneys.
What "saved time" gets wrong
Time saved matters when it returns to billable work, field capacity, or owner focus you can point to. If the hours simply refill with more unstructured admin, the business did not gain capacity. Tie time claims to a capacity measure: more jobs per tech without more overtime, more matters per attorney without nights in the inbox, or owner hours removed from a named workflow.
What to avoid
- Building first and promising to "add analytics later."
- Changing the definition of the metric after the build to protect the project.
- Vanity counts (messages sent, AI replies) that never touch revenue or cycle time.
- Blaming the tool when the baseline was never taken.
- Asking for niche proof you cannot support. Method proof is fine when labeled; invented precision is not.
Method note
Our delivery path baselines in week one of a six-week transformation, then stabilizes and reports against that baseline near the end. CG Service Pros and Avivo Homes are method references for how measurement stays attached to the work. They are not stand-ins for your numbers.
Leave with this
If nobody can say what the number is today, that is the first work. The cheapest way to waste an implementation budget is to automate a workflow you never measured.
Next chapter and next step
Previous: What AI can and cannot do in a service business today. Next chapter (coming): Inside your tools or a new platform?
If you already have a workflow in mind and want help choosing the baseline, book an AI Strategy Call.
Back to the buyer's guide pillar.
Have a workflow you are considering for AI?
Start by mapping the problem, the people, the systems, and the measure that matters. A strategy call can help you decide whether there is a useful next step.
