Write down where you started
Before anything is switched on, record the baseline. How long does the job take each week? How often does it go wrong, and what does a mistake cost to fix? How long does a customer wait? Without these numbers, any improvement is a feeling, and feelings are easy to argue with.
Keep it simple. Two or three numbers that the person who does the job recognises are worth more than a dashboard nobody trusts.
1. Time and errors
Compare the same measures after a few weeks of real use. Count the hours that came back and the mistakes that didn’t happen. Be honest about the ones that moved the wrong way.
2. How often people correct it
If every action starts in an approval queue, you get a useful record for free: how often people approve, edit or reject what the system proposes. A falling edit rate shows the approved corrections, written into its rules, are doing their job. A rejection is evidence too, and one bad call should cost it its freedom.
3. Whether it’s actually used
Look at who opens it, and how often. A system that worked in week one and is ignored by month three hasn’t worked. This is usually an adoption problem, not a technical one, and it’s worth catching early.
4. What it costs to run
Track the running costs against the limits you set. If the cost of running it grows faster than the time it saves, that’s a decision to make on purpose, not a surprise on an invoice.
5. What’s still done by hand
Every workflow keeps some manual steps, and some should stay that way. Write them down. They show where the next improvement might be, and where the line deliberately sits.
Review these numbers on a regular rhythm with the person who owns the workflow. That review, more than any single figure, is what keeps a system working after the people who built it have gone. It’s the core of what we do in Embed.
Not sure where your business stands with AI? Find your AI profile: ten questions, about four minutes, no sign-up.