The baseline expires, and this is unrecoverable
It is the most common and most expensive mistake in this work. Once the new process is live, the old behaviour is gone. You cannot reconstruct what your response latency was, or how conversion varied by latency, from a CRM that has been reorganised around the new workflow. The comparison you needed is no longer available at any price.
Capture, before anything is built: enquiry volume by channel; time from enquiry to first genuine human response; conversion split by that response speed; average first-year value; and the share of enquiries receiving no second contact. Take at least a full sales cycle, and longer if the business is seasonal.
Store it outside the systems being changed, dated, with the extraction method written down so somebody can reproduce it in a year — and that somebody will not be you, it will be a finance business partner who was not in the room. A baseline nobody can re-derive is an assertion.
Found, sealed and payback
Three words with three definitions, and keeping them distinct is what makes the reporting credible. Found is annualised leakage identified during the diagnostic. Sealed is annualised recovery measured after implementation, over the period stated. Payback is the weeks required for measured recovery to equal total fees paid, diagnostic plus implementation.
The relationship between the first two is the important part. Sealed is always smaller than found, and a report claiming otherwise should be discarded. Some of what was found was never recoverable; some of it required a decision the business chose not to make; some leaked out through a different gap once the first was closed. A complete seal is not credible and reads as fabrication to anyone who has done this work.
A worked shape makes the relationship concrete. A diagnostic finds $3.0M of annualised leakage. Implementation closes the top three leaks across two phases at $60,000. Measured recovery over the following two quarters annualises to $1.1M — smaller than what was found, as it always is. Total fees paid are $65,000 including the $5,000 diagnostic, so payback is roughly three weeks of recovered revenue. Those are the four numbers a board wants: found, spent, sealed, and weeks to payback.
The metrics that hold up in front of a board
- Time to first genuine contact. The most direct measure of the thing that changed, visible within a week, and difficult to argue with because it is a timestamp rather than an inference.
- Conversion by response speed. The causal link. If fast-response conversion was already higher before the build, and volume has moved from the slow group to the fast group, attribution is an argument rather than an assertion.
- Revenue per enquiry. Volume-independent, which matters because it survives a good quarter and an ordinary one without needing to be explained.
- Dollars recovered, annualised, with the system and the window named. This is the number a CFO will engage with and the only one that belongs in a board pack headline.
- Payback in weeks. Converts the whole exercise into the unit an investment committee already uses.
And the ones that do not
Hours saved. The saving is real and it never lands on a financial statement, because the freed hour is absorbed into the working day rather than redeployed to something measurable. It cannot defend a renewal and it should not be the headline.
Activity counts — messages sent, workflows built, tasks automated. These describe the system rather than its effect, and their improvement is guaranteed by the project having happened at all.
Satisfaction scores in isolation. Useful as a guard against having broken something, useless as evidence of value, and easily moved by things unrelated to the build.
Attribution without a research budget
The fair objection to any before-and-after is that other things changed too. Three defences, in ascending order of rigour, and all three are available to a mid-market company.
- 01Compare against the same period last year rather than the previous quarter. Crude, and it handles seasonality, which is the most common confound.
- 02Use a control. If you have multiple locations, territories or channels, roll out to some and not others. This is the strongest option most companies have and it is far easier than it sounds — it mostly requires resisting the urge to launch everywhere at once.
- 03Segment within the change. Compare conversion among enquiries that were already fast before the build against those moved from slow to fast. If the gain is concentrated in the second group, the mechanism is doing the work rather than the calendar.
When to measure what
- Week one: operational metrics only. Latency, error rate, escalation rate. These move immediately and tell you whether the thing works at all.
- Weeks two to four: leading indicators. Contact rate, booking rate, drop-off. Do not report conversion yet.
- One full sales cycle: conversion and revenue. For some businesses that is three weeks and for others two quarters. Reporting a conversion lift before a cycle has closed is reporting noise, and doing it once costs you credibility for every real number afterwards.
- Quarterly thereafter: same metrics, same method, same definitions. Consistency is what makes the series worth anything, and changing a definition mid-series destroys it.
The one-page report that survives
The baseline, dated, with its extraction method. The change, in the same units. The attribution argument, including what else changed in the period and why you believe it does not account for the result. The dollar figure with its arithmetic visible. And the assumptions listed, so a sceptical reader can adjust one and watch what happens.
That last item is the one people leave out and it is the one that does the work. A conclusion that survives a hostile adjustment of its own inputs is worth more than a larger number that does not, and inviting the adjustment is what signals you already tried it.