Accountable AI. A clear number for every AI idea, before anyone writes code.

Give us one process from your plant. We work out what it costs to build, to run and to keep safe, and what it saves. Then we say go, or don’t go, and why. Accountable means every number has a source, and a named person signs the answer. Assay, our own tool, makes that possible.

Book an AI Strategy Sprint → See what you get
4weeksFixed scope, fixed price. The sprint does not get longer to make an idea pass.
1workshopA few hours of your people’s time. The rest of the work is ours.
1decisionGo, or don’t go. Both come with the number and the reasons.
0guessed numbersEvery figure has a source. A number we did not get is asked for, not invented.
Why most AI ideas fail their own business case

Three things go wrong before the first line of code. All three can be fixed.

In most plants the list of AI ideas is long and the budget is short. The maths is not the hard part. The problem is this: the inputs are guessed, the reasons are not written down, and the most expensive cost line is often missing.

One

The number is built on guesses.

Six numbers decide any payback: volume, hourly cost, headcount, cost per person, error rate and cost per error. Most business cases simply guess them. Assay leaves them empty and asks you. A confident number built on six guesses is worse than no number.

Two

Nobody wrote down the reasons.

Notes become a summary. The summary becomes a slide. Six weeks later nobody can find the moment a step was called “a lot of work”. Assay links every claim to a quote and a time in the recording. “Where does this come from?” takes one click.

Three

The governance cost is missing.

Build cost and licences look fine next to the saving. Then add the people who still check the output, the evidence you must keep, and the monitoring. A clear yes can turn into a no. We add that line from day one. It is often the line that decides.

No guessed numbers. Every claim has a source. Governance is in the number. That is what we mean by accountable AI, and it is what Assay was built for.

Where AI ideas come from in a plant

The ideas usually come from the same twelve places.

These are the processes we hear about most in workshops with manufacturers. Some are quick wins. Some look attractive and are not. The sprint tells you which is which, for your plant, with a number.

Engineering and order intake

  • Drawing data typed into the ERPHigh volume, no judgement needed. Often a quick win.
  • Change requests and impact analysisA big saving, but an engineer must sign. Needs a plan first.
  • Quotations and tendersWorks when past quotes are on file. Depends on your data.
  • Customer order intake and checksReading and rules. Often a quick win with a person checking samples.

Plant and quality

  • Daily production reportingThe data exists in three systems. The work is collecting it.
  • Shift handover notesA template usually fixes it. No AI needed, and we say so.
  • Quality release decisionsA person signs every release anyway. The checks cost more than the saving.
  • Scrap and rework predictionOnly when the sensor data is really there. Often it is not yet.

Supply and service

  • Supplier invoice matchingRules in the ERP do most of it. Saying so saves a budget.
  • Maintenance tickets and spare partsNeeds a history of past decisions. Start keeping it, then look again.
  • Supplier certificates and documentsReading and checking documents. Usually a safe first step.
  • Complaints and warranty claimsValuable, but judgement is involved. A copilot, not an autopilot.
Often a go Not yet · needs a plan Often a don’t These are the patterns we see most often. Your plant will differ. That is what the sprint finds out.
What you get at the end

Four things you can take to a board. Every number with its source.

The sprint ends with a document, not a slide deck. It starts with the recommendation. The evidence is at the back. A named consultant has read it and signed it. Your CFO, your IT lead and your works council can read the same document.

The business case

Four lines: what it costs to build, to run, and to keep safe, set against what it saves. Then a verdict: go, or don’t go, and why.

from the workshopassumedfrom you

A scored shortlist

Every idea from the workshop is scored and ranked. Most stop here. In a typical sprint, 2 ideas out of 8 are worth building. Saying so is a result too.

GoNot yetDon’t build

The first build, designed and priced

What we would build first, on your systems. The roles and the weeks it takes, and what it costs to run. Small enough to start the week after the sprint.

€120k2 engineers · 1 analyst
9 weeks · example figures

A roadmap and a governance plan

Three phases, in order. What must be true before each one starts. And the rules that apply to your case, so the risk review has no surprises.

Three things the report does not contain, on purpose
  • No timeline chart. The roadmap has phases but no dates. A timeline would invent them.
  • No before-and-after comparison. There is no measured baseline yet, so there is nothing to compare.
  • No guessed payback. If a number we need did not come up in the workshop, the report asks for it. It does not invent it.
A worked example

Two ideas from the same plant. One go. One don’t.

Example figures from an invented mid-size manufacturer. The numbers are made up. The shape of the answer is the one you get: what the team said, what it costs, what it saves, and the verdict.

Idea 1

Drawing data typed into the ERP

Planners copy part data from supplier drawings into the ERP by hand. Twice, when a revision arrives.
“We read it off the PDF and type it in. Twice, if the supplier sends a revision.”Planning team · kickoff workshop · 14:22
Build one-off€120kfrom the workshop
Running per year€25kassumed
Governance & checks per year€30kfrom the workshop
Saving per year€190kfrom your numbers
Go · pays back in year 1High volume, no judgement needed, mistakes are easy to see. A person checks a sample, so the checking cost stays small. Net +€135k a year once live.
Idea 2

Quality release decisions

Let a model propose the release decision for each finished batch, with the test results attached.
“Every release is signed by a quality engineer. That will not change.”Quality manager · plant workshop · 41:10
Build one-off€140kfrom the workshop
Running per year€35kassumed
Governance & checks per year€160kfrom the workshop
Saving per year€120kfrom your numbers
Don’t build · the checks cost more than the savingSafety-critical. A person must check every output anyway, so the governance line is bigger than the saving. The record says so. The number follows. Net −€75k a year.

Both answers are results. The second one saved a budget and a year of work. In most sprints, 6 ideas out of 8 stop here. The two that go through are the ones worth defending.

Assay · our own tool · it runs the Decide step

The workshop is the easy part. Defending the plan is the hard part.

Assay is the tool behind every AI Business Case we make. It reads one conversation and turns it into a case you can defend in front of a board, a works council, a risk committee or an auditor. We show the tool itself in a private demo.

Assay · built and run by Focaloid

Assay keeps what your team said and what we concluded in two separate places. Everything we conclude must point to something that was said. If it cannot, it does not go in the report.

Nothing is invented

A number we did not hear is asked for, not guessed. The report carries the question instead of a figure.

Every claim has a source

Open any finding and see the sentence it came from, who said it, and when in the recording.

A person decides

Assay does the work. Our consultant reads it, checks it and signs it. The name on the report is a person’s.

assay · business case · one idea
One real screen. The business case for one idea, on an invented example company. No client data.
A working tool, not a slide

This is what the business case looks like inside Assay.

The cost side is filled in from the design of the solution. The numbers only you can give us are marked as needed, not guessed, until you give them to us.

  • Real product. Built and run by Focaloid, used on every sprint.
  • Shown in private. How it works inside is something we show you, not something we publish.
  • Yours to keep. You get the analysis and the report. The tool stays with us.
Ask for a private demo →
Made for European manufacturing

Rules, people and systems are in the number from day one.

A business case that ignores the EU AI Act, the works council or the state of your ERP is not a business case. It is a wish. We count those things before you decide, not after.

The EU AI Act, counted

Each idea gets a risk position under the EU AI Act. Higher risk means more checks, and checks cost money. That cost goes into the governance line before you see the verdict. So the risk review brings no surprises.

Your people are on the record

The workshop is with the people who do the work, not only with managers. What they said is written down, with the time it was said. That makes the case easier to explain to a works council, and easier to defend when someone asks who said that.

Your systems as they are

ERP, PLM, MES, the quality system and the spreadsheets between them. We look at what your process really runs on and what data it keeps today. An idea that needs data nobody keeps is stopped before it costs anything.

Your data stays where you need it

Assay runs in our ISO/IEC 27001 certified environment. For clients who need it, it can run inside yours. Your recording is used for your sprint and for nothing else, and you keep every document we make from it.

Who gets what out of it
Plant director, COOA ranked list of ideas, and a first build small enough to start the week after the sprint.
CFOA number with a source on every line, and the cost of running it, not only of building it.
CIO, ITHow it would be built on your systems, and a build, buy or partner answer for each idea.
Quality, complianceThe risk position of each idea, and the checks that will still be done by a person.
How the 4 weeks run

One conversation in. A decision you can defend out.

The AI Strategy Sprint has a fixed scope and a fixed price. Assay does the heavy lifting. So our consultants spend the weeks on judgement, not on typing up notes.

Week 1

Listen

A discovery call. Then a workshop with the people who do the work: planners, operators, quality, IT. Recorded. We also look at your systems and your data.

  • You bring the people and the process
  • We run the sessions
Week 2

Analyse

Assay turns the sessions into the analysis. Our consultants read every finding against its evidence and collect the open questions.

  • You get a short list of open questions
  • We answer what we can ourselves
Week 3

Check

Your answers go into the analysis and it runs again. Rankings can change. We review the shortlist and the build, buy or partner question with your teams.

  • You answer, and challenge the scores
  • We argue the alternatives
Week 4

Decide

The business case, the design for the chosen ideas, the roadmap and the priced first build. The report starts with the recommendation.

  • You keep every document
  • We say go, or don’t go, and why
What we need from you
  • One conversation. A workshop or interview recording, or the notes from it. A Teams or Zoom recording works as it is.
  • The people who do the work, for a couple of hours. Four teams that each know their own part is fine.
  • A few numbers about the process. Volume, hours, people, error rate. If you do not have them, we tell you which ones matter and why.

The 4 weeks are the shape, not a fixed timetable. We agree the rhythm on the discovery call. The sprint does not get longer to make an idea pass.

When it is useful

When “we should use AI” has to become a plan that other people will read.

Not every question needs a sprint. These are the moments when the answer must hold up in front of a board, a works council, a risk committee or an auditor.

The board asked for an AI strategy

There is a budget and a deadline, but no agreed list. The output must be ranked and priced, not a list of ideas.

A process nobody can describe from start to end

Four teams each know their own part. The map is built from what each of them says. Exceptions and workarounds are recorded, not tidied away.

Build, buy, or partner?

The question is not “can AI help?”. It is “is building the right way to get it?”. Effort, risk and readiness must be on one page.

A regulated or safety-critical process

Someone who was not in the room will re-read the decision. Every judgement must show its evidence. The level of automation must be argued both ways.

A pilot that stopped

Something was built. The demo went well. It never went live. Usually the readiness or the governance work was skipped. The sprint checks exactly that before anything is built.

Two plants, the same work, two ways

After a merger, or across sites. The record captures both processes as described. The comparison rests on facts, not on who won the meeting.

Questions we get

Before your first sprint.

The questions that come up in the first call with a plant director or a CFO, answered the way we answer them in the room.

How much of our people's time does it take?

A few hours. One workshop with the people who do the work, usually two to three hours, and a short round of questions in week three. Managers see the results in week four. The rest of the work is ours.

Does the AI write our business case?

No. Assay does the analysis. A consultant with a name reads it, corrects it and writes the recommendation. The document you receive was signed off by a person. It is not a generated file that nobody owns.

The useful part is not the speed. It is that the reasoning is written down in a form you can argue with. So the argument happens before the steering meeting, not during it.

Can we see Assay?

Yes, in a private demo. We walk you through the tool on an example plant, so you can see what it reads, what it refuses to conclude without you, and what the report looks like. We do not publish how it works inside. Ask on the discovery call and we will set it up.

Does this fall under the EU AI Act?

Some ideas will. Most back-office and shop-floor automation is not high risk under the Act, but anything that touches safety, people's jobs or access to a service can be. Each idea in the report carries its risk position, and the checks it needs are counted in the number. So if an idea is high risk, you know before you spend, not after.

What happens to our recording?

It is used for your sprint and for nothing else. Assay runs inside our ISO/IEC 27001 certified environment. For clients who need it, it can run inside yours. You keep every document we produce from it.

What if we do not have all the numbers?

You still get the whole cost side: what the first build costs to make and to run. We cannot tell you what it saves until the missing numbers arrive, and we will not guess them. The report says so, in plain words, and lists the numbers as questions.

In practice, the numbers turn up during the sprint. Each question explains why it matters. That is usually enough to get it answered.

What if the analysis gets something wrong?

We correct it, and the analysis runs again. Your corrections are kept, and they are marked as yours. A fact that changes in week three really changes the result. It is not patched into a slide.

How is this different from a consultant with a chat tool?

Three things, and they all show up when someone challenges the result.

Every claim has a source: a finding that cannot point to something your team said does not go in.

Nothing is guessed: a number we never received comes back as a question, not as an estimate.

It can be repeated: the same conversation gives the same shape of answer.

A good consultant with a chat tool also gets a first draft quickly. The difference shows in week six, when the board asks where a number came from.

Start here

Do you have a process worth a real number?

Bring one conversation: a workshop, an interview, or the notes from either. On the discovery call we show you Assay on an example plant, and what a first build would cost.

The AI Strategy Sprint

4 weeks, fixed scope, fixed price. A discovery call, workshops with the people who do the work, a look at your systems and data. Then a scored shortlist, a business case, a roadmap and a priced first build. You keep every document, whatever the answer.

  • Go comes with the business case, the design and a first build that can start the week after the sprint.
  • Don’t go comes with the reason, the line that decided it, and what would have to change for the answer to change.
Manufacturing · regulated processes · Europe and the UK

The process you would like us to look at

Four fields. We reply within one working day with a time for the discovery call and a short list of what to bring.

No newsletter. No email sequence. One reply from a person who has read your note.

Thank you. We have it.

Someone who has read your note will reply within one working day with a time and a short list of what to bring. If you have a recording of the process being described, keep it ready. That is where we start.

That did not send. Please try once more — or email connect@focaloid.com with the same note and we will pick it up from there.