A week with a stopwatch prices six manual workflows by the seat that runs them. Three get built, one of them wired to a score, and the rest are deferred or declined with the reason written down.
The operations manager has to decide which manual workflows to automate first. The rule is the team's own sense of where the time goes. It feels fine because everyone watches this work every day. People are reliably bad at estimating work they have done a thousand times, and the error runs in a consistent direction.
We name the baseline at kickoff. The ops manager estimates each of the six workflows before any timing begins, and those estimates are written down.
We time the work as it runs. Forty-one timed runs across five seats, median and 90th percentile reported separately, interruptions counted, and monthly volume taken from the ERP over 90 days.
We price each workflow at the loaded rate of the seat that does it, then rank all six by recoverable spend against build difficulty.
We try the simpler way first. The ERP's own order acknowledgment emails are licensed and switched off, so turning them on comes before any build. A customer ordering portal is declined, and invoice matching is deferred because two vendor portals publish no API.
We build three workflows beside the desk, never in place of it. Order intake runs in shadow against live mail before cutover. The reorder forecast is fitted on 24 months and checked against the most recent quarter it never saw.
We hand back each build with a named owner, documentation for the people who run it, and a retime at 90 days.
Order intake is estimated at 5.0 minutes and timed at 11.4. Credit hold review comes in 23% faster than estimated.
So what A method that only ever finds work costing more than you thought is not measuring.
Priced at the estimates, reorder point maintenance ranks fifth and gets left alone. On the stopwatch it is second.
So what A workflow costing $73,900 a year would have been left alone on a guess.
Only two of the six workflows want a score instead of a rule, and only one of those has the history to fit one today.
So what Anyone selling a model across the whole department has not timed the department.
Inbound orders arrive in the ERP as drafts for one-click confirmation, with anything uncertain flagged rather than guessed. The purchasing manager opens only the Review tier of a weekly suggested-order list. Backorder replies arrive drafted for a rep to read and send. The credit-hold release gets a reason code, so that six months on there is history to fit a score against. The proof is designed in twice. Each build runs in shadow against live volume before cutover, and the same stopwatch retimes the same workflows at 90 days.
The 335 hours become value only if they are pointed at something. If the desk absorbs the slack, the return is zero.
Forty-one timed runs describe the week they were timed in. A seasonal peak or an absent colleague changes the medians.
Extraction accuracy is measured on held-out historical orders before it is measured on live mail, and live mail can carry formats the history did not.
We'll tell you what your own history can say about it, and which offering fits. If none does, we'll say that too.
Start a conversation