What we hear from customers

“We're making decisions without capturing the information we actually need.”

“
The board wants the new operating model by June, and the answers sit with 100 people across a dozen teams. There's one of me, and the calendars alone would take a quarter.
Program director, engineering services

“
Before redesigning the claims process we surveyed 200 handlers. A fifth replied, in tick boxes. We found out how many people touch a claim. We found out nothing about how they handle one.
Operations lead, insurance

How it works

Tell Workstuff who to talk to and what to find out. It interviews them all and reports back everything it learns.

Workstuff runs as an MCP server, so it works from whatever agent you already use: ChatGPT, Copilot, Cowork or Claude.

workstuff

> Find out how our claims process really works. Ask everyone in claims ops.
● Planning interviews
  · 180 people across 6 teams
  · 12 questions, voice or text
  · portal at acme.workstuff.ai
● Portal live. Invitations sent to 180 people.
● 148 accounts reconciled into 47 steps, 9 decisions, 12 exceptions.
>

The solution

An AI interviewer that maps the process, the exceptions, and how decisions really get made today.

Send a link. 100 people talk through how the work really gets done, and the AI probes for detail.
Your questions guide the interviews. Workstuff reconciles every account into one version, and flags where they clash.
When the build team asks how the process runs, you have the answer, with a name attached.

An AI-led interview running on a phone, by voice or text

100 people interviewed in 6 days. No meetings booked.

100People interviewed

6 daysTo a full library

Every answerTraced to who said it

Build-readyA spec, not a slide deck

See which processes are actually ready.

Turn the interviews into a scored view of every process: documented steps, an owner for every decision, a known list of exceptions. You know which one to automate first, and which one will bite you.

Process health scored across steps, decisions, exceptions and readiness to automate

And how the work actually gets done.

Not one person's version. Every account of the same process, reconciled into one map. Where people agreed, it is settled. Where they did not, you see both versions rather than the loudest one.

A current-state process map with numbered steps, a loop and a revenue leak

01

Process steps

What actually happens, in order, including the step that never made it into the SOP.

02

Decisions

Who decides, on what basis, and who they go and ask when they aren't sure.

03

Exceptions

The cases that don't follow the rule. The ones that break an agent on its first day.

04

Pains

Where the work is slow, done twice, or quietly hated by everyone doing it.

05

Ideas

What the people doing the job would fix first, which nobody has ever asked them.

The difference

The same process, two very different builds.

Build from an SOP somebody wrote two years ago

↓

Build from what the people doing the job said last week

Discovery quality depends on which analyst you can spare

↓

Your best analyst's method in every conversation, whoever runs it

Debrief after every workshop and hope the notes are good

↓

Every conversation comes back structured: the steps, the decisions, the exceptions

Ship the agent and find the edge cases in production

↓

Know the exceptions before anyone writes a line of it

Send a survey, get a 20% response rate, get charts

↓

Send a link, get 80% completion, get the story behind the numbers

Every process starts from scratch

↓

Every process builds on the library you already have

The AI agent that interviews all your stakeholders at once.