00 ~1 day
Alignment
Which areas are in, who gets invited and which cost per hour puts a value on everything. The interview script is adapted to the company's own vocabulary.
A census, not a sample · two weeks
AI process mapping: the whole team interviewed, not ten people.
Here, AI process mapping does not mean generating a flowchart from a text. It is an AI agent that interviews 100% of the people in the area, in parallel, and returns in two weeks the work as it actually happens: every task with its frequency, duration, system and where it gets stuck, totaled in hours and currency per year. We call it a process census by AI agent. It does not plug into any system.
Drawing is the easy part. The hard part is knowing what to draw.
There is no shortage of tools for drawing flows, and several already generate the diagram with AI. What stalls a mapping project is the fact-finding: learning what each person really does, not what the procedure says they do. A good part of administrative work lives in spreadsheets, email and messages, outside any system.
The traditional method solves this with interviews, and interviews are bounded by the interviewer's calendar. That is why a diagnostic usually hears from eight to twelve people in six weeks. Not because ten is the right number, but because it is what fits. And the ten end up being whoever had a free slot.
An agent has no calendar. The conversations happen in parallel, in the same week, and a hundred people take the same time as ten. That is what turns a sample into a census, and opinion into a number: the same task described by two people becomes a confirmed data point.
// six ways to capture a process
What each method sees, and what it leaves out
None of them replaces the others. Process mining is unbeatable for what runs through the system; a workshop is the best place to design the future. The question is which one covers the work nobody sees today.
method
what it captures
coverage
what it requires
Consultant interviews
The process as the person tells it, with context and exceptions
A sample: 8 to 12 people in about 6 weeks
Everyone's calendar, plus consultant hours for each interview
Workshop with the team
The consensus flow, good for designing the to-be process
Whoever is in the room, and whoever talks loudest
Several people pulled off their work at the same time
Survey or spreadsheet
A self-reported list of activities
Broad but shallow: nobody asks a follow-up question
Someone to consolidate and fix answers that do not add up
Task mining
Clicks and screens, with exact timing
The machines where the recorder was installed
Desktop installation, plus IT and privacy sign-off
Process mining
The real path of every case inside the system
100% of what runs through the system, and nothing of what happens outside it
An event log extracted from the ERP
Census by AI agent
Task, frequency, duration, system and sticking point, with follow-up questions
100% of the people in scope, in parallel
A link and about 40 minutes from each person
// how it works
From invitation to numbers in two weeks
These timelines are for a scope of up to three areas. What changes from one company to the next is the size of the census, not the playbook.
01 ~7 days
The census
Each person talks to the agent through a link, whenever they can. It captures task, frequency, duration, system and where things get stuck, and asks a follow-up when an answer is vague.
02 ~2 days
Consolidation
Accounts of the same task are merged so nothing counts twice. Out come the handoffs between areas and the hours per year of each task.
03 ~3 days
Numbers and priority
Hours are converted into currency per year, with a floor and a ceiling: the floor is what at least two people confirmed. The queue comes out ranked by impact and effort.
The individual conversation is confidential: leadership receives the consolidated view, never one person's interview. And no one is asked to name a colleague. That is what gets people to describe their real routine, not the procedure's version of it.
// what comes out of the map
Four kinds of findings, in hours and currency per year
Every opportunity comes with its absolute value per year, a floor, a ceiling and the cost-per-hour assumption printed. Anyone who wants to redo the math has everything at hand.
The process that goes in circles
Rework, stalled approvals, the same data keyed in by two areas. It gets fixed without technology, as long as someone sees the whole path.
Where an AI agent delivers
A repetitive task, with a clear rule and a system in the middle. It enters the queue with its effort estimated, not as a promise.
The hours of those who only sit in
The ten-person meeting where three people decide. It needs no technology at all and is often the cheapest opportunity on the map.
What already runs by itself
What the area automated on its own. It counts as zero: the company already has that gain.
// when it makes sense
Before buying the tool, switching ERPs or right-sizing the team
- You want to know where AI or automation pays off before choosing the tool.
- You are implementing or migrating an ERP, and the fact-finding depends on one key user per area.
- You need to size an administrative area and there is no time tracking.
- You mapped processes once before, and the PDF went stale without becoming a decision.
And when it is not the right fit: if the whole process runs through a system and the log is at hand, start with the data. The diagnostic on your ERP extract answers faster. The two complement each other: where the team's perception and the data disagree is usually where the money is.
The census is delivered by Nous Scan, a tool from UpFlux, at three levels of coverage: one area, up to three areas plus what flows between them, or the whole company.
// frequently asked
Frequently asked questions
What process managers, PMOs and directors ask before mapping.
What is AI process mapping?
It is the use of artificial intelligence to capture and document how work happens in a company. There are two very different uses: generating the drawing (a flowchart or BPMN) from a text, which speeds up documentation, and running the fact-finding itself. At UpFlux, AI process mapping is the second: an AI agent interviews 100% of the people in the area, in parallel, and returns every task with its frequency, duration, system and sticking point, totaled in hours and currency per year.
What is a process census by AI agent?
It is a process fact-finding method in which an AI agent talks individually with every person in scope, not with a sample, for about forty minutes per person, by link and with no scheduling. Accounts of the same task are merged so nothing counts twice, and the result comes out in hours and currency per year, with a floor (what at least two people confirmed) and a ceiling. UpFlux delivers this method with Nous Scan, in about two weeks for up to three areas, with no integration with any system.
How is this method different from process mining and task mining?
Process mining rebuilds the process from the system's event log: it sees 100% of what runs through the ERP and nothing of what happens in spreadsheets, email or messages. Task mining records the user's screen and requires desktop installation. The census by AI agent asks the people who do the work, so it captures the work outside the system, the reason for the exception and what the person thinks should not be their job. The methods complement each other: where the team's account and the system's data disagree is usually where the biggest opportunity lies.
Does it require integrating with the ERP or installing anything?
No. The input is what people say about their own routine. Each invitee receives a link, confirms their corporate email and talks to the agent whenever they can. There is no connector, no data extraction, no agent installed on any machine and no load on the IT team.
How long does it take to map an area's processes?
With the census by AI agent, about two weeks for a scope of up to three areas: one day of alignment, seven days of interviews in parallel, two of consolidation and three for the numbers and prioritization. The timeline barely grows with the number of people, because the conversations happen at the same time. A traditional diagnostic based on consultant interviews usually hears from eight to twelve people in about six weeks.
Does leadership see what each person answered?
No. The individual conversation is confidential, and leadership receives only the consolidated view by task and by area. The respondent is also never asked to name a colleague. That rule is what gets people to describe their real routine, not the procedure's version of it.
Is the result a flowchart?
The result is a map of how work flows, with the handoffs between areas and systems, plus an improvement queue prioritized by impact and effort. Each item comes with hours and currency per year and the cost-per-hour assumption printed. The queue separates the process fix that needs no technology, the task an AI agent executes, the hours of those who only sit in on meetings and what the area has already automated.