Alternatives to consultant interviews for process discovery in a 500-person company

The alternatives to consultant interviews for process discovery are document review, observation, workshops, surveys, process and task mining, and AI interviews that talk with everyone. None of them replaces the others. Each looks at a different object: the designed process, the observed task, the argued handoff, the recorded event, or what people actually meant. In a 500-person company the honest question is not which method is best but which combination hears the whole company before someone starts building.

Why the interview programme runs out at forty people

The standard textbook on the subject sorts discovery into three families. Evidence-based discovery works from documents, on-site observation and event logs. Interview-based discovery depends on process participants and domain experts working through an interview, model, validate cycle. Workshop-based discovery puts the participants in one room (Dumas, La Rosa, Mendling and Reijers, 2018). Consultancies lean on the second family because it produces defensible diagrams quickly.

The problem is arithmetic. In our experience a consultant-led programme at a company this size interviews the leadership team, the function heads and a handful of people nominated as process owners: somewhere between twenty and forty conversations over four to six weeks. That is a sample, and a biased one. It is the view of the people who designed the process and are accountable for it. The planner who keeps her own spreadsheet because the ERP field is wrong, the dispatcher who phones the warehouse every morning because the pick list arrives late, the credit controller who releases orders on a WhatsApp message: none of them were on the list.

Then the interviews are reconciled. Where two people described a handoff differently, the consultant picks a version or draws a diamond. The contradiction, which is usually the most valuable finding in the whole programme, is averaged away. We wrote about why those disagreements are worth keeping in When three teams describe the same process three ways.

Document review: the designed process

Start here, because it is nearly free. Standard operating procedures, the org chart, system manuals, the approval matrix, last year's audit findings. Two days of reading tells you what the company believes its processes are and gives you a vocabulary for everything that follows.

What it misses is everything that changed after the document was written, and everything that was never written down. Treat the result as a hypothesis to be tested, not as a map.

Observation and job shadowing: the task as performed

Sitting beside someone for a shift is the only method that shows the real sequence of a task, including the seventeen clicks, the second monitor with the private spreadsheet, and the walk to the printer. On a shop floor or in a warehouse it is often the only way to see the work at all.

It does not scale. One observer sees one seat at a time, and a 500-person company has hundreds of distinct tasks. It also changes what it watches. A systematic review of the Hawthorne effect concluded that consequences of research participation for the behaviours being studied "do exist, although little can be securely known about the conditions under which they operate, their mechanisms of effects, or their magnitudes" (McCambridge, Witton and Elbourne, 2014). People being watched tend to follow the procedure. Use observation for the three or four highest-volume tasks once you know which they are.

Workshops: the handoff argued out in one room

A half-day workshop with the people on both sides of a handoff produces a shared map faster than any other method, because disagreements surface immediately and get settled with everyone present.

The room has a hierarchy, though. The senior person's version tends to win, the quiet people do not contradict their manager in public, and anyone not invited does not exist. Workshops are excellent for resolving specific contradictions you have already found. They are a poor way to find them, because the official story is what gets performed in front of colleagues.

Surveys: breadth without depth

A questionnaire reaches all 500 people in a day. It is the cheapest way to learn which functions think their work is broken and where people say they spend their time.

A fixed questionnaire cannot follow up. It asks the questions you thought of, in the words you chose, and people answer what was asked rather than what matters. Long forms go unfinished. Free-text boxes fill with one-line answers. Use a survey to decide where to dig, never as the record of how the work happens.

Process mining and task mining: what the systems recorded

Process mining reconstructs a flow from event logs, discovering "real processes (i.e., not assumed processes) by extracting knowledge from event logs" (IEEE Task Force on Process Mining, 2012). Where a process lives inside one system, it is unbeatable: every variant of order-to-cash across a year, with cycle times, from data you already hold. It needs a case identifier present on every activity, an activity name and a timestamp (Microsoft Learn, 2026).

Anything that never touched the system is invisible: the phone call that unblocked the shipment, the approval that happened in an email thread before anyone clicked approve. Task mining, which records desktops, extends the view to the screen and stops at its edge. It is also worker monitoring, which the UK Information Commissioner's Office says "is likely to capture excessive amounts of workers' personal information" and requires an impact assessment where the risk is high (ICO, 2023). The longer comparison is in Process mining vs task mining vs AI interviews.

AI interviews: talking with everyone, privately

A private AI conversation with each of the 500 people, at the same time, in the first week. It follows up the way a good consultant does, asks the dispatcher why the pick list is late, and keeps every answer. Because it talks to everyone rather than a sample, the contradictions between teams are found rather than assumed, and because nobody is in a room with their manager, the workaround gets described.

Its limits are the same as any interview. Memory rounds off frequencies, so it will not give you cycle times across 40,000 cases. And candour has to be engineered, not requested. If people believe their manager will read what they said, they recite the procedure. In Transcript, individual answers are never visible to management: leaders see the world model of the organisation, never a quote or a summary of one person, and each person approves what was learned from their words before it enters the record.

The six compared

The table is our framing, illustrative rather than measured.

Our view: a sequence, not a choice

For a company of about 500 people, this is the order we use.

1. Read the documents for two days and write down the designed process as a hypothesis.
2. Hear everyone in week one. We use AI interviews for this. If you do not have one, a short survey followed by forty human interviews chosen by the survey, not by the org chart, is the next best thing.
3. Hold workshops only on the contradictions the interviews surfaced. Each one is a specific question with a specific set of people.
4. Mine the flows that live inside systems to measure what the interviews found. The interviews tell you which logs are worth extracting.
5. Observe the three or four highest-volume tasks before anyone redesigns them.

Consultant interviews keep one job in this sequence: the ten conversations with leadership about what the company is for and what it will not do. Those are not process discovery, and they should not be mistaken for it.

Questions people ask next

How many people does a consultant interview programme usually cover?

In our experience, twenty to forty in a company of 500, over four to six weeks, weighted towards managers and named process owners. The people doing the work are underrepresented by design, because the list is drawn from the org chart.

Can we do process discovery without consultants at all?

Yes. Document review, a survey and workshops can all be run internally. The hard part internally is candour: colleagues will not describe workarounds to someone who reports to the same director. Whichever method you choose, separate who hears the answers from who manages the people giving them.

Is an employee survey enough?

No. A survey tells you where people think the problems are and gives you the shape of the company's opinion. It cannot follow up, so it never reaches the reason the planner keeps her own spreadsheet. Use it to aim the deeper methods.

How long should discovery take before an AI project starts?

Long enough to hear everyone and short enough that the findings are still true when you build. With AI interviews the whole company can be heard in a week and the contradictions worked through in the second. A programme that takes a quarter usually reflects the method, not the company.

Does interviewing everyone create a privacy or works-council problem?

It can, and it should be designed for. Individual answers must never be visible to management, the guarantee has to be enforced in the system rather than promised in an email, and employee representatives should see the design before the first conversation. In Transcript that separation is built into the database, so leaders can only ever see the pattern.

Lightbloom AI starts by hearing the whole company rather than a sample of its managers, because the process that costs the money is the one nobody wrote down. Transcript does the hearing and builds the world model from it. For the definitions behind this piece, read What is AI process discovery?.

References

Dumas, M., La Rosa, M., Mendling, J. and Reijers, H. A., Process Discovery, in Fundamentals of Business Process Management, second edition, Springer (2018), https://link.springer.com/chapter/10.1007/978-3-662-56509-4_5

McCambridge, J., Witton, J. and Elbourne, D. R., Systematic review of the Hawthorne effect: new concepts are needed to study research participation effects, Journal of Clinical Epidemiology 67(3), 267 to 277 (2014), https://pmc.ncbi.nlm.nih.gov/articles/PMC3969247/

IEEE Task Force on Process Mining, Process Mining Manifesto, in BPM 2011 Workshops, LNBIP 99, Springer (2012), https://www.tf-pm.org/upload/1580737614108.pdf

Microsoft Learn, Prepare processes and data, process mining in Power Automate (2026), https://learn.microsoft.com/en-us/power-automate/process-mining-processes-and-data

Information Commissioner's Office, Employment practices and data protection: monitoring workers, specific data protection considerations for different ways or methods of monitoring workers (2023), https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/employment/monitoring-workers/specific-data-protection-considerations-for-different-ways-or-methods-of-monitoring-workers/