Before you migrate 2,000 workflows, build the inventory
A platform migration usually starts as a copy job in everyone's head: workflows go in one side and come out the other, on a newer tool. It's an appealing plan because it's easy to schedule and easy to describe to a sponsor.
It also assumes the old estate is a tidy set of things worth keeping. It almost never is. Years of small requests leave behind duplicates, one-off fixes that became permanent, and workflows whose owner left in 2019. Copying all of it faithfully means paying to carry that mess into the new platform, and then maintaining it there.
So the first deliverable isn't a plan. It's an inventory.
One row per workflow, and nothing you can't fill in
The temptation is a 60-column template. Nobody fills in 60 columns, and the blank ones get ignored at the exact moment you need them. Six fields, honestly completed, will carry a migration:
- Owner. A named person, not a team. "The data team" means nobody.
- Schedule. How often it runs, and whether it's actually scheduled or run by hand.
- Inputs and outputs. Source systems, tables, files — and where the result lands.
- Downstream consumers. Which dashboards, reports or teams read the output. This is the field that decides risk.
- Complexity signals. Tool count, macros, embedded code, custom connectors.
- Business criticality. What breaks, for whom, if it doesn't run tomorrow.
A spreadsheet is the right tool here. The inventory is a working document that three people will edit in a meeting, not a database.
What "duplicate" actually means
This is the judgement call that decides how much of the migration you avoid, and it's where most inventories go soft. Two workflows that read the same source and write the same target are not automatically the same workflow. They may run at different frequencies, apply different filters, or serve different consumers who will notice immediately if the definitions merge.
The workable test is: would one output, with the union of the logic, satisfy every current consumer? If yes, retire both and build one. If the answer is "probably", it's a no until someone has spoken to the consumers.
Score complexity on what drives rebuild cost
Complexity scoring usually goes wrong by trying to be precise. A score of 73 out of 100 means nothing to the person funding the project, and nothing to the developer who has to hit a date.
What works is a small set of bands, each tied to a real cause of effort:
- Inputs and outputs — five sources and three targets is materially more work than one and one.
- Embedded logic — macros, iterative macros, R or Python tools. These are the ones that have to be rethought rather than re-pointed.
- Non-standard connectors — anything bespoke or undocumented adds discovery time you can't estimate from the tool count.
- Business logic density — a workflow with 40 formula tools encodes rules that exist nowhere else. That's the real risk, not the number of tools.
Three bands — small, medium, large — with a one-line reason attached to each, survives a steering committee. A decimal score does not.
Check usage against the schedule, not the other way round
Scheduler logs tell you a workflow ran. They don't tell you anyone read the output. A daily schedule writing to a table nobody queries looks identical to a critical pipeline.
Pair every workflow's runtime with its downstream consumers, and ask the owners directly. "Does anyone use this?" gets a more honest answer in a ten-minute conversation than in a form, and it costs less than migrating something nobody wanted.
The four decisions
Every row ends in one of four outcomes, and the split between them is the actual value of the exercise:
- Retire. No consumers, or superseded by something else. The cheapest migration is the one that doesn't happen.
- Rebuild as-is. Critical, well understood, no reason to change the logic mid-migration.
- Rebuild simplified. The workflow carries years of accumulated fixes; the rebuild is the chance to drop the ones nobody can explain.
- Leave for now. High complexity, low urgency, unclear ownership. Naming these explicitly is better than pretending the migration covers everything.
Every "retire" needs a named person who agreed to it, and every "rebuild" needs a stated reason it can't be retired. Six months later, when someone asks why their report stopped, that record is the difference between a decision and an accident.
The uncomfortable part
The inventory will show that a meaningful share of the estate is not worth carrying. That's not a failure of the previous team — it's what any system looks like after a few years of people solving problems under time pressure.
Saying so, with evidence, is the job. The migration plan that follows is smaller, cheaper and more honest than the one everyone imagined at the kickoff, and the people who have to maintain it afterwards will thank you for it.
Reviewed 19 November 2026. The inventory fields and the retire / rebuild / leave decisions are method, not tool version, so they age slowly — but if your estate works differently, that's worth knowing. Email me.