Agencies, discovery week · September 2026 · 3 min
Get the real process out of the discovery calls
The documented process is the ideal. The real one lives in Excel, Outlook and one person's head. A week of discovery calls holds all of it, if someone has time to pull it out before the report is due.
/ How it runs
- Call recordings and transcripts
- Discovery notes
- The agency's scoring rubric
- Extract steps, handoffs, systems and exceptions
- Draw the swimlane: a lane per role, a box per task, a diamond per decision
- Mark what was said against what an operator confirmed
- Score each finding with the rubric line and the segment
- Swimlanes an executive reads in one look
- Exception and workaround list
- Scored findings report, cited to the calls
What people say about it
Someone running Celonis at a mid-sized company said the first run is always eye-opening, because people see how far the real process is from the flowchart they call the process. Getting the event log together is most of the project, though. One consultant put it at 80 percent of the time. Mid-market companies do not have a data lake for this.
The business analysts skip the logs. One of them said he never uses BPMN because nobody reads it. A lane per person or system, a box per task, a diamond per decision. Then you sit with the people who do the work and walk through it, and that is when the side spreadsheet and the email to a friend in another department come out.
Why calls and not logs
In a one-week engagement the calls are the only evidence you get. They also hold what a log never does: the exceptions people mention as asides. What happens when the portal is down. Who gets a call when an order does not fit the form. That is the real process.
What we would build
Transcripts go in. For each workflow the client described, the miner pulls out the steps, the handoffs, the system at each step and every exception anyone mentioned, and draws the swimlane. Each step is marked as said on a call or confirmed by an operator. Each finding is scored against the agency's rubric, with the rubric line and the call segment next to the score. The pod lead approves before anything ships.
The last call of the week is the check: the swimlane goes in front of the people who do the work and they correct it. If it works, the report is drafted on Thursday instead of Sunday night, scoring is the same from pod to pod, and the rubric becomes something the agency keeps improving. Event logs, task mining and automation proposals come later, if at all.
/ More notes
Do the estate audit before the migration plan
The tooling is not the hard part of a Data Center to Cloud move. The hard part is that nobody knows what is in the instance, and the assessment that would tell them is done by hand by the most senior people.
- Jira and Confluence exports, read only
- Marketplace app list
- Script registry
- Usage: last update, issue counts
- Rate each item: duplicate, unused, migratable, blocker
- Keep the rule that fired and a confidence score
- Low confidence goes to a reviewer
- Consultant accepts, edits or rejects
- Traffic-light inventory of the whole estate
- Phased backlog in Jira Cloud
- Assessment report in Confluence
Draft the synthesis, with the quotes attached
An assessment ends with someone turning a pile of transcripts into findings and a roadmap. It is the valuable part, and it gets done at night by one person because there is never time for it.
- Transcripts from Otter, Fireflies or Whisper
- Interview notes and questionnaires
- The firm's codebook
- Pass one: verbatim quotes with speaker and time
- Pass two: themes that cite quote IDs only
- Count participants per theme
- Reviewer accepts, edits or rejects
- Findings with a confidence line
- Every claim traceable to who said it
- A draft roadmap
Have a workflow like this one?
Thirty minutes. We map where it is stuck and what a first build looks like.
Book a free AI audit