seyhun akyürek
book a call
all free tools

agent workflow fit scorer

Rate one process on six things that decide whether agents will work: how clear the rules are, how clean the inputs are, how often exceptions happen, what an error costs, whether systems have APIs, and whether past cases exist. You get a score, the workflow shape that fits, and the risks to fix first.

free, no sign-up, runs in your browseropen source on GitHub
the work today
rate the process
57fit score out of 100
500hours of work per month today
325hours saved per month (modelled)
$175,500value per year

agent workflow with a human approval step

suggested architecture

  1. Intake: document extraction (OCR plus a small model) into a typed schema
  2. Router: a small, cheap model classifies cases and sends exceptions to a person
  3. Worker agents: one per decision type, each citing the rule it applied
  4. Approval: a person approves before anything is sent, paid or filed
  5. Evaluation: a golden set re-run on every prompt or model change

in its favour

  • Written rules cover most decisions, so agents can follow and cite them.
  • Plenty of past cases to build a golden test set from.

fix or plan for

  • Messy inputs: budget for document extraction and test it on your worst scans.

Hours saved assume the pattern removes 65% of handling time. Test that in a pilot before you bank it.

questions

how is the score calculated?

A weighted average of the six ratings (rule clarity carries the most weight, and a high error cost counts against the fit), adjusted up or down by the monthly volume of work. The weights are in the open-source code.

what do the patterns mean?

Full automation runs straight through with sampled quality checks. Agent with review prepares everything and a person approves. Copilot drafts and a person decides. Not yet means fix the inputs or write the rules down first.

where do the hours saved come from?

Today's volume times handling time, multiplied by a typical share of that time each pattern removes: 85%, 65%, 35% and 10%. Treat it as a starting estimate to test in a pilot.

more free tools

Want the numbers checked against your own process? Book a 20-minute call.