The problem AI Scan solves
Every company runs on some work that never made it onto an org chart. A spreadsheet nobody signed off on. A checklist one person keeps in their head. It is not a secret. It is just invisible, because nobody ever asked about it. This gap is why so many companies get their first AI project wrong. Leadership decides AI matters, often right after a board meeting or a competitor's announcement, and asks the obvious question: where do we start? Too often, the answer comes from whoever argues best in a meeting, not from whoever actually knows where the week goes. The budget ends up funding the loudest complaint, not the most expensive one.
What AI Scan actually does
AI Scan replaces that guesswork with evidence. Over five to ten working days, an agent has a short daily conversation with your employees, right inside the Slack or Teams workspace they already use. Each conversation runs to about ten exchanges.
Every answer feeds into one report. That report describes how your company actually runs, and ranks the work AI could help with, from most promising to least. That is the whole engagement. AI Scan tells you where to focus. It does not build anything. In practice, that means a ranked, evidence-based shortlist, instead of a decision made by whoever spoke last in the meeting.
How the conversation works
Employees never see a form, a rating scale, or a dropdown menu. Each message from the agent asks a single, open question about what is actually happening in their work right now. There is no list of options to choose from, because a menu makes people pick from what they were handed instead of describing what they actually do. A few of the real openers:
- "So, what are you working on today?"
- "Is there anything you have to re-do or double-check more often than you'd like?"
- "Is there a part of your work where you often end up chasing someone for an update?"
- "Do you ever find yourself copying the same information into multiple different systems?"
The agent works from 45 opening questions spread across eleven themes, from the tools people rely on, to who they end up waiting on, to the moments a decision comes down to gut feel rather than a rule. These are starting points, not a script, so two people at the same company can end up in very different conversations.
The first question rarely does the real work. The second and third one do. When someone names a process, the agent asks how often it happens. When someone names the system they use, it asks what happens to that information once it leaves the system. That is usually how a manual handover between two tools comes to light. When someone mentions waiting on a colleague, the agent asks if they do anything to speed it up. That is usually where the chasing emails and the private tracking spreadsheets turn up. And when an answer stays vague, the agent asks when it last happened, rather than repeating the same abstract question.
Across the scan, each person typically describes between three and ten pieces of recurring work, something like reconciling supplier invoices every Friday. For each one, the agent tries to pin down five things:
- What the work is, in plain language
- Which steps are done by hand
- How often it happens, and at what volume
- Which parts come back or get redone
- Which decisions need a person, and which a machine could take on
Why the answers have to come from the people doing the work
Nobody at a two-hundred-person company holds the whole picture. A department head can describe what their team is supposed to do. Only the person doing the work every Friday knows that the spreadsheet exists at all. That kind of thing rarely survives a workshop with a manager in the room. Yet it is exactly what a company should know before it commits a budget.
People are also more candid with an agent than with a room full of colleagues, which is the main reason these conversations happen one on one, in writing. Answers are separated from names as soon as they are collected. The analysis looks at a role and a department, never a named person. And the contract our clients sign commits them to leaving employees alone about what they said.
Is this about replacing anyone?
It is the first question every participant asks. The honest answer shows up in what the agent does, and does not, ask about. A question can land on the skilled part of someone's job, the part they trained for and get paid for. When it does, the answer usually comes back as some version of "that's what I'm here for." The agent takes that answer and moves on. It is far more interested in what happens right before that skilled work, and right after it.
The same patterns show up again and again. Information gets moved by hand between systems that do not talk to each other. Someone assembles a weekly report from two or three different sources, by hand. Two more come up almost as often. Triage: an email or ticket someone opens only to decide whose job it is. And chasing: the work of finding out whether someone else has done their part yet. Rounding out the list is drafting: documents built from a house template with a few details swapped each time. These are the parts of the job people are visibly relieved to finally talk about.
A scan is just as willing to say a process should stay exactly as it is. Something can be painful and still be the wrong candidate for AI. Maybe it only happens four times a year. Maybe the judgment involved is the whole point of the job. Or maybe the data it needs lives somewhere nobody has agreed to open up. Those calls go in the report as clearly as the opportunities do.
Inside the report
The report itself is a platform you log into and keep. You can highlight sections, leave comments for colleagues, and export it any time. It has six parts, and each one takes a different shape rather than blending into one long document.
Overview is a short executive summary up front: how many people took part, and the two to four themes that came up again and again across their answers. It also states how much of the picture the interviews actually cover, so a report built on a handful of replies reads differently from one built on a company-wide conversation.
Technology landscape maps the systems your business runs on, grouped by category, with lines drawn between the ones that connect. A solid line means data moves automatically. A dashed one means somebody moves it by hand. A marker on a system flags where people described it as a real source of friction. This is how a spreadsheet wedged between two official tools finally ends up on the page.
Process inventory lists the recurring work the interviews turned up, grouped by the department that holds it. Each row shows how often the work happens and how many people independently described it, and is flagged if it was also scored as an opportunity elsewhere in the report. Take one entry: running the monthly supplier invoice batch. It names the role that owns it, and records that it adds up to about ninety invoices a month, most of them landing in the first week. The work itself: downloading the invoices from a shared mailbox, checking each one against the purchase order, keying them into the finance system, filing the paperwork. It notes that invoices with no purchase order behind them have to be traced back to whoever placed the order. The keying itself is mechanical, but whether an out-of-policy invoice gets paid is still somebody's judgment call. Behind every row sits the fuller description, in employees' own words.
Opportunity map plots every opportunity on two axes: how well it fits AI, and how important it is. Importance is pain multiplied by volume, how much something hurts, multiplied by how often it happens, so a small daily annoyance and a big rare one can land in the same place. Each dot falls into one of four zones: a quick win, a strategic bet worth the investment, work to fix without any AI at all, or something scored honestly and left with no proposal attached.
Recommendations sequence a path forward into what to start now, what follows once the first moves land, and what to revisit once the foundations are stronger. Each one says what kind of move it is: adopting a tool that already exists, a lighter layer of automation, a custom build, or simply fixing the process yourselves, no AI required.
What not to do lists everything the scan genuinely considered and deliberately did not recommend, and why. Clients tell us this is the section they read first. A parked idea is not a permanent no. It is a prompt to come back to once the company, the evidence, or the underlying data changes.
What AI Scan is not
AI Scan is not a survey, an IT audit, or a performance review. It reads nothing beyond the Slack or Teams channel carrying the interview, and every finding traces back to something an employee actually said. The agent interviews rather than consults: it never suggests a fix mid-conversation, a refusal to answer is final, and a stated time limit is respected. If something cannot be known, the report says so, instead of guessing.
The short version
Every company has that spreadsheet. AI Scan finds it, and everything like it, straight from the people who do the work. Five to ten days of ordinary conversation turn into one ranked report: no guessing, no boardroom debates, just evidence for where AI should go first. Where do you start? Now you'll actually know.