Track: The Baseline Is the Only Thing That Makes After Mean Anything

The question every automation result should have to answer

Compared to what?

It is a boring question and almost nothing survives it. A workflow now runs in four hours. It used to run in - and here the sentence usually slows down, because nobody wrote the old number down, and the figure that arrives next was produced from memory by people who would prefer the project to have worked.

That is not dishonesty. It is what memory does. Ask anyone how long a task used to take and you get the version they remember most vividly, which is the worst one.

What Track measures

Four things per workflow, and no more, because a baseline with twenty metrics in it is one nobody will maintain or defend.

• Cycle time from the moment work enters the process to the moment it leaves.
• Throughput per week, including the runs that were abandoned.
• Error rate, defined as work that had to be redone rather than work someone disliked.
• Cost per run, at loaded hourly cost rather than headline salary.

Then the window: long enough to include a bad week, chosen before the data is seen rather than after.

The signature

The operator signs it. Not the vendor.

This is the only stage of the engagement with a signature on it, and the reason is narrow. A before number produced by the party who will be judged against it is a negotiating position. Signed by the client, it becomes the fixed point that makes the after number mean something, in both directions: the vendor cannot revise it down when the result disappoints, and the client cannot revise it up when a better story is wanted.

It also forces the definition of success into the open at the one moment when neither side knows whether they will like it.

What breaks without this stage

The engagement still works. That is the problem.

The build ships, the new cycle time is real, and the improvement is reported against a before figure assembled after the fact. Everyone is pleased for a quarter. Then somebody in finance asks how the payback was calculated, and the answer traces back to a number that was never measured, and at that point the entire result becomes unusable - not wrong, necessarily, but unusable, which for a capital decision is the same thing.

The arithmetic that justifies the build is only as good as its first term. Track is the first term.

Where it sits

Track is the third of six stages, and it only works if the process was stabilised first - the reason for that is in why the scope gets frozen. What happens to the measured process next is elimination before encoding, and the argument for putting an audit ahead of a build at all is in the audit before automation.