Measure the workbefore you change it.
Every tool you own reports what AI costs. None of them can report what it returned, because the return lives in your business process and none of it is an AI number.
And it can only be captured against a number somebody took before anything changed. That makes it a decision you make at build time, not a report you run afterwards.
Getting what your people knowinto a form that runs.
Every service below is the same operation on a different subject. You cannot hand over what nobody has described, and you cannot measure it either, which is why the same discipline produces both the automation and the number.
TodayField meanings known by the two people who have been there longest
AfterA schema, with each source scored for what it can carry today
TodayHabit, and an exception path nobody ever wrote down
AfterA spec with steps, approval gates, and the exceptions stated in your words
TodayHow we do things here, visible only in the output
AfterAn installed skill, plus the checks that run before anything leaves
TodayOne person's recollection of why something was decided
AfterDecision records you can search, with the reasons attached
TodayA self-rating nobody believes
AfterA researched ladder with observable rungs, and where each person sits on it
The reason this works where a knowledge-management programme does not: writing it down is a by-product of building the thing your team asked for. Nobody documents a process for the wiki. They will describe it in detail to get the job they hate off their desk.
Automating it is oneof five answers.
Cheapest first, and we try them in that order. All five are measured against the same baseline, which is why the number is worth taking whichever way a candidate lands. A report where everything comes back as "automate it" is a sales document.
Delete it
Nobody reads the output. The approval has never once rejected anything.
All of the hours, at no build cost
Fix it upstream
The work only exists because two systems do not talk, or a setting is wrong in software you already own.
Most of the hours, for a configuration change
Redesign it
The steps are in the wrong order, the approval is in the wrong place, or it should be batched.
Large, and it needs no technology at all
Automate it
It happens daily, the data is already reachable, one person can approve it, and somebody hates doing it.
The build, measured against the baseline
Leave it alone
The volume can never justify the build, or so many cases are exceptions that a person ends up doing it anyway.
Nothing, and saying so is the finding
Every figure carrieshow much to trust it.
Assumed
Our own sourced defaults, before we have seen anything of yours. Every figure in a proposal starts here and is labelled.
Client-input
Your volumes, your rates, your sample. This is what a signed baseline gets you, and it is where honest scoping happens.
Measured
Out of the running system's own log, against the baseline. This is the only tier that appears on a monthly report.
Time reclaimed is capped at 60 to 80%, because no job ever gives back every hour it touches. Anything about revenue is labelled modelled. And a tier never gets upgraded in a meeting.
The numbers are also allowed to go down, and the month they do, the report says so with the reason. Any supplier whose reported savings only ever climb is doing marketing.
Four questions,including for us.
If somebody has quoted you a savings figure and you cannot tell whether it is real, these are the four questions that settle it. We will answer all four in writing, and we are happy for you to put them to everyone else on your list.
- 01What is the baseline, and who measured it?
- 02When was it taken, before or after the change?
- 03Does the running system report its own numbers, or does someone maintain a dashboard?
- 04Can the number go down, and has it ever?
Start with one job.We will measure it before we quote it.
Start with your people or with the work. One job at a time, on probation, measured in hours and money. Everything we build stays yours.