A baseline nobody agreed is worth nothing
Two things go wrong with measurement on automation work, and the second is worse than the first. The first is that nobody takes a reading at all, so at the end of the engagement the result is a feeling. The team says things seem calmer. That is worth something, and it doesn't survive the first conversation with a finance director.
The second is that a firm takes the readings on its own, reports them at the end, and the client is handed a number they had no part in producing. The conversation that follows is about the query and not about the work: which contacts were counted, whether that month was typical, whether the reply time included the automated acknowledgement. The work might have been good. It cannot be shown.
There is a third outcome that looks like success and isn't. The client accepts a figure without reading it, because querying it would cost an afternoon nobody has. The number goes into a deck, the engagement ends politely, and neither side learns anything they could use on the next process.
So the order matters more than the metric. The numbers are taken before the proposal, defined in writing, confirmed by the client, and taken again the same way at day thirty.
The five numbers we take
Five is not a target we worked towards. It is what was left after taking out every number the client could not check for themselves inside their own systems.
The first is the count of dormant contacts in the database, using whatever definition of dormant the client already works to. The second is the count of contacts with no logged activity in twelve months, which is the harder and less flattering figure, because it does not rest on anyone's definition.
The third is first response time, measured from three test enquiries sent through the client's own website form at three different hours. The fourth is the number of meetings booked from the existing database in the last ninety days, which separates the demand the company already paid for from the demand it is about to buy again.
The fifth is the team's own estimate of manual hours a week. It is the softest number on the list and the one the room cares about most, because it belongs to the people doing the work and not to us. We record it as an estimate, in their words, against the tasks they named.
All five are taken inside the client's systems, in front of the client wherever that can be arranged. A number somebody watched being produced is much harder to dispute four weeks later, and the query gets saved with its filter in it, so the same person can run it again without us in the room.
Why the reply time comes from outside
The average response time in a CRM report is usually not the number that matters. It is an average over the enquiries that got a reply at all, taken from a system that starts its clock when somebody opens the record.
So we send three test enquiries through the form the company advertises, at three different hours, and we time what comes back. One goes in during the working day, one in the evening, one at a weekend. The evening and the weekend enquiries decide the figure, because a team's follow-up is as good as its hours until something automated covers the rest.
We do this before the first call, not after the engagement starts. It is the one number on the list we can take without access to anything, and it often makes the case on its own, because the client is reading their own reply time to a real enquiry that nobody in their building noticed.
One rule goes with it. The enquiries are genuine, sent by a person from a real address, and anybody who replies is told what it was. The purpose is to time the process, and a test designed to catch a junior out has cost more than it measured.
Definitions decide the argument at day thirty
Every number on the list hides a definition, and every definition is where the disagreement will be. Dormant means whatever the client says it means. What they say goes into the baseline document before the build: the field, the filter, the date range and the tool the query ran in.
Day thirty repeats that query. Same filter, same field, same tool, same person where the person is still there. A number produced a different way is a different number, and the comparison is worthless even when it flatters us.
There is a discipline in keeping the list short. Five numbers can be read aloud in a meeting and argued with on the spot. A dashboard of forty charts gets approved by everyone in the room and opened by nobody afterwards, which leaves a client with no baseline and the impression that measurement was handled.
The baseline document
The five numbers live in a short written document, one page, agreed before the proposal goes anywhere. Each number carries the date it was taken, the method behind it, the tool it came from and the name of the person who ran it. The document belongs to the client and it is theirs to keep.
It has a second use we did not design it for. When somebody joins the client's team six months later and asks why a process runs the way it does, that page is the only record of what the week looked like before. Teams forget their own manual work quickly once it stops, and a returned hour is invisible in a way a lost hour never is.
