The baseline you did not take
Because they measured after deploying. Once a process has changed, its prior state cannot be reconstructed with any rigour, so there is nothing to compare against. The cause is sequencing, not dishonesty — and the fix is five days of measurement before any code ships.
The pattern
A system goes live. It clearly helps — the team says so, the inbox feels calmer, replies go out faster. Six weeks later someone asks by how much, and the honest answer is that nobody knows.
The data to answer it existed. It existed right up until the moment the process changed.
Why reconstruction rarely works
You can sometimes recover a partial baseline from source-system timestamps — enquiry creation times, first-reply times, booking records. It is worth attempting, and it is better than nothing.
What you cannot recover is anything that was never recorded: how many enquiries were handled by phone and never logged, how often someone was chased twice, how many quietly went unanswered. Those are usually the numbers that matter most.
What a Baseline Week contains
- Volumes, by channel, over a defined window.
- Median and p90 first-response time from source timestamps.
- Leak rate — enquiries that received no reply at all.
- Error and rework rate in the current manual process.
- A countersignature from the client confirming the figures are accurate.
Why the signature matters
An unsigned baseline is a number you produced about work you were about to be paid to change. The countersignature is what makes every later claim checkable by someone other than you — and it is the reason most firms never ask for one, because it constrains them afterwards.