What is AI process discovery?

AI process discovery uses AI to find out how work actually happens in a company, rather than how the procedures manual says it happens. Where process mining reads system logs, the newer approach interviews the people who do the work, hundreds at once, and reconciles what they say. The output is a map of the real processes, the points where teams contradict each other, the people the work depends on, and a list of fixes worth doing, each with a number.

Over two fifths of what a company knows sits in one head

A 2018 survey of over 1,000 US workers by Panopto and YouGov (Panopto, 2018) found that 42 percent of institutional knowledge was unique to the individual holding it, and that workers spent 5.3 hours a week waiting for information or recreating it.

Every company has two process documents. The first is the SOP folder. The second lives in the heads of the people doing the work: the workaround for the supplier whose invoices never match, the spreadsheet that is the real inventory system. Discovery is the work of writing the second document down.

Only a process someone can describe can be automated

Automate the written process and you automate a fiction. PwC (2025) puts about 20 percent of an AI initiative's value in the technology and 80 percent in redesigning the work around it. BCG (2024) found roughly the same from 1,000 executives: the leaders put 70 percent of their resources into people and processes. You cannot redesign work you cannot see.

Eurostat's 2025 survey (Eurostat, 2025) adds a telling detail: among EU enterprises that had considered AI and decided against it, 71 percent cited a lack of relevant expertise. The missing expertise, in our experience, is knowing which process to point a model at.

Process mining reads logs, and logs miss the interesting part

Process mining takes event logs from an ERP or ticketing system and reconstructs the flow: which step followed which, how long each waited. It is precise, and it is blind in a particular way. It sees only what touched a system. The phone call that unblocked the shipment and the email chain that is the real approval are not in the log.

Our view, against the process-mining consensus: in a mid-sized company the log covers the minority of the work. Discovery that starts from logs maps the transactional spine well and misses the muscle around it.

Interviewing at scale is what changed

Consultants have always interviewed people. The limit was arithmetic. Twelve interviews of an hour each, then a workshop, then a slide deck. The twelve were managers, because managers were available. The people who run the process were summarised by the people who manage them.

An AI conversation removes the arithmetic. Everyone in the company can have a private, ongoing conversation, a few minutes a day, typed or spoken, on a phone if they are not at a desk. It asks follow-up questions, tracks what has not been covered, and comes back tomorrow.

Contradictions are the most valuable finding

When two teams describe the same handoff differently, a traditional write-up averages them into one tidy diagram. That destroys the most useful information. If sales says the discount is approved by the sales director and finance says the CFO, the real answer is that nobody is sure, and the company is leaking margin at that point. Good discovery keeps the contradictions, names them, and puts them in front of management to resolve.

What does not come out is a finished automation. Discovery tells you what to fix and what it is worth. Someone still has to build the fix and prove the number, and the same team should do both, or the map becomes a presentation.

Privacy decides whether people tell the truth

People describe workarounds only if their manager will not read the transcript. Discovery tools have to make individual conversations invisible to management by design, and show only organisation-level patterns. If privacy is a policy rather than an enforced control, people will work it out.

Questions people ask next

How is AI process discovery different from process mining?

Process mining reconstructs a flow from system logs. AI process discovery interviews people and includes the work no system records. Most companies benefit from both, with interviewing covering the larger share.

How long does it take?

A few weeks of conversations, with each person spending five to ten minutes a day.

Will employees actually talk to an AI?

Yes, if two conditions hold: the conversation is private from management by design, and each person can see and correct what was learned from their words. Remove either and you get the official process restated.

Does this replace consultants?

It replaces the interviewing and mapping phase. Deciding what to fix and building the fix is still work someone has to do.

Transcript, built by Lightbloom AI, holds private conversations with everyone in a company, typed or spoken, and turns them into a world model of how the work actually happens: processes, contradictions between teams, the people the work depends on, and problems valued in the company's own hours and cost. Managers never see an individual's conversation; see how it works at /transcript.

References

1. Panopto, Inefficient Knowledge Sharing Costs Large Businesses $47 Million Per Year (2018), https://www.prnewswire.com/news-releases/inefficient-knowledge-sharing-costs-large-businesses-47-million-per-year-300681971.html
2. PwC, 2026 AI Business Predictions (2025), https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-predictions.html
3. BCG, AI Adoption in 2024: 74% of Companies Struggle to Achieve and Scale Value (2024), https://www.bcg.com/press/24october2024-ai-adoption-in-2024-74-of-companies-struggle-to-achieve-and-scale-value
4. Eurostat, Use of artificial intelligence in enterprises (2025), https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises