Choosing the process for your first pilot

Which process should the flywheel start on?

Starting on the wrong process is the most expensive mistake of an early AI transformation: the pilot runs for a year, nothing pays back, and belief in the whole idea drains away. Two ways to make that choice well: a free ten-minute self-assessment, or a one- to two-week process audit that ends in front of your executive team.

A 30-minute call about two or three of your processes →Take the self-assessment — ten minutes, free →

Run by the authors of the method — Sergei Lipchanskii and Askhat Urazbaev.

Two ways in

From a light first step to a session with your board.

Different depths for different situations. You can start with the self-assessment and decide about the audit afterwards, or go straight to the audit if you already know the company is ready.

Are we ready at all?

Self-assessment

Take this if you want a quick read on how mature the company is, with no budget and no commitment.

How it works: Twenty questions across five dimensions of maturity: awareness, tooling, process thinking, a culture of measurement, and readiness to change. One person answers online in ten minutes and the PDF report arrives by email.

What you get

  • A read on the company’s AI maturity across five dimensions, visualised
  • A comparison with your industry, from aggregated client data
  • Three to five recommendations for what to do next
  • A gap map: where you are strong and where the holes are

Time and format

Ten minutes online

Free

Take the self-assessment →

Which process comes first?

Process audit

Take this if you have three to seven candidate processes and need to choose one or two for the pilot, ideally with an argument the board will accept.

How it works: On the one or two finalists we design what the flywheel should look like for that specific process: what goes into the Knowledge Base, which fields the Decision Log needs, which autonomy level to start at, how quality is checked, which metrics to watch.

What you get

  • A map of 15 to 20 of your processes, each marked as flywheel, speed-up without redesign, or leave alone — one page with a status and a reason for every one
  • A flywheel design for one or two key processes — a specification of the five elements of the flywheel for that process: what to build, where the data comes from, which autonomy level to start at
  • A roadmap for six to twelve months — the pilot, a network of three to five champions, and explicit go / no-go criteria
  • A 60-minute session with the executive team — a working session where we go through the findings and make the roadmap decisions

Time and format

One to two weeks from signature to the final session

Standard (one week) or extended (two weeks, two processes)

Request the audit →A 30-minute call →

When an audit is the right thing

How to tell whether an audit is what you need.

The audit closes one specific gap: AI is clearly needed, there are several candidate processes, and it is not obvious which to start with. If you are somewhere else on the road, we will say so and point you at the format that fits.

  • You have three to seven candidate processes for AI and need to pick one or two for the pilot.

    ✓ Audit

  • You are going to the board for a budget and need the choice of process argued, with a portfolio of priorities behind it.

    ✓ Audit

  • Three executives have three different AI ideas and no shared priority.

    ✓ Audit

  • The process is already chosen and your team is ready to build the flywheel themselves.

    → Workshop

  • The process is chosen but you have no team to build it — you need a working system delivered.

    → Implementation

  • Under a hundred people in the company — a one- to two-week audit is usually more than you need.

    → Self-assessment or a 30-minute call

The main filter

The flywheel only applies to processes that have a feedback loop — you can tell whether the AI’s decision was right. Without that loop the flywheel does not turn: the system has nothing to learn from. If no such process exists in the company, the audit will show it, and save you the cost of a pilot that would otherwise have gone into the wrong place.

How the audit runs

What happens over one to two weeks.

A close look at real processes: four to six interviews with process owners, observation of the work itself, and analysis of data from your systems — ERP, CRM, spreadsheets, logs.

Week 1

A map of the processes, with a decision on each.

We gather 15 to 20 key processes from every level of the company (CEO and COO, four to six process owners, two or three senior people accountable for the outcome). Each is checked against four things: does it repeat at least fifty times a week, is there a feedback loop, does a mistake cost something, and is the team behind it willing to try. Every process then gets a mark: flywheel (start this quarter), speed-up without redesign (apply AI to the existing flow without the Data Flywheel), leave alone (not relevant, or not technically ready).

Week 2

Designing the flywheel for one or two finalists, plus the roadmap.

We take one or two processes marked as flywheel and design the five gears for them. That design is the brief for the pilot:

  • Knowledge Base: which rules, context and examples the AI needs for this process, where they come from (documents, what people know, the history of customer contact), what already exists and what has to be built.
  • Autonomy: which level from A1 to A5 the pilot starts at, which decisions the AI makes on its own, which stay with a person, and the rule by which autonomy is raised from there.
  • Decision Log: which fields to capture (the incoming context, the decision the AI made, the human correction, the outcome) and how that log turns into a training signal for the next iteration.
  • Quality check: which tests run before launch and on every iteration, who is accountable for quality, and how decay gets caught.
  • Metrics: three to five metrics that show the outcome of this process rather than the activity of the AI, and which signals to treat as counter-metrics.

Alongside it we build a roadmap for six to twelve months: pilot → extension into adjacent work → the next process. Who sits in the network of three to five champions, and a go / no-go point at every checkpoint.

The end of it

A 60-minute session with your executive team.

It is a working session, not a presentation: we go through the findings and make the roadmap decisions in the room.

What the audit does not cover

Said plainly — this is the work of implementation, not of one or two weeks:

  • building an automated eval pipeline on your data
  • assembling the champions network
  • agreeing data security policy
  • a full compliance checklist

Time and commitment

What it takes, and what we need from you.

Timeline
One to two weeks from signature to the session with your executive team. At least a week before the start we settle who is interviewed and what data we get access to.
The hidden cost to your team
Seven to eleven interviews of 60 to 90 minutes: one or two with the CEO or COO, four to six with process owners, two or three with senior staff. Plus the 60-minute session at the end. In total around 10 to 15 hours of your senior people’s time. Without that time set aside the audit drags out and the look at the gears stays shallow.
What to prepare in advance
A list of 15 to 20 key processes from your side (if there is no list, we build one in the first meeting, which adds a day or two to week one), access to exports from your systems (ERP, CRM, spreadsheets, logs), and agreement that the 60-minute session with the executive team will happen.
Request the audit →

The authors of the method

Run by the authors of the method.

Sergei Lipchanskii, Askhat Urazbaev. Sergei founded Kiberry (AI consulting and platform work since 2023). Askhat co-founded ScrumTrek (product and Agile transformations since 2008).

Where the audit format came from

Across three years of AI projects we started every time with the same question: which process to begin with. On some projects the choice was made intuitively, at the client’s request — and in several of those the pilot went into a process with no feedback loop. The flywheel there had stalled within three months — we took the lesson and began checking candidates methodically against four criteria before any pilot started. That check is what became the process audit. The programme is three years of practice, repackaged into one or two weeks.

About the company and the team

FAQ

Common questions.

What is the difference between the self-assessment and the audit?

The self-assessment is a ten-minute online form: one person answers twenty questions and gets a general read on the company’s maturity across five dimensions. The audit is one to two weeks of work — interviews with the executive team and process owners, analysis of your data, and a closing session with the board. The self-assessment answers “Are we ready at all?” The audit answers “Which two processes do we take first, and why?”

What if the audit says we do not need AI right now?

Then that is what we will say. In our experience, in some companies of 50 to 500 people not one of the 15 to 20 processes is ready for a flywheel as things stand: one needs standardising first, another has no feedback loop, a third has no data. Finding that out in a week costs a fraction of a pilot that never had a chance.

Can we have the audit without the session with the executive team?

You can, but we advise against it. That session is not a presentation — it is where the executive team makes the roadmap decisions. Without it, the audit’s output usually sits in a document and never turns into action. If a 60-minute slot with the CEO or COO cannot be found at all, that is a strong signal the initiative will not have support where decisions are made, and the pilot will most likely stall.

Request

Request the audit, or talk a process through.

Within one working day, a 30-minute call with one of the leads. On the call we check that the audit is actually what you need, and agree the format.

Company size *
What you need *
How many candidate processes for AI do you see right now *

We reply within one working day. We do not pass your contact details to anyone else.

Related reading

The Oper8 method

The open method: the Data Flywheel, the five elements of the flywheel, the autonomy spectrum.

Read the method

AI implementation

The full cycle from spec to a running pilot with our team. Eight to twelve weeks.

See the approach

Client case studies

Haleon, Astana Hub and others — where the method has already been applied.

See the cases

Let’s
work together

We’ll look at your problem, pick the right format, and show where AI actually strengthens your operations.

Book a consultationTelegram

50+

companies have gone through AI transformation with us — from diagnostics to working solutions.

Kiberry

© 2026 Kiberry — the European practice of ScrumTrek

Privacy policy