AI adoption ยท Discovery
Find where AI can make a useful difference
Most companies know AI would help somewhere and cannot say where first. Discovery settles that. It samples the work as it actually runs, measures the before state, ranks the candidates on value and effort, and comes back with a pilot you can start or decline on the evidence.
The first conversation is free. Discovery itself is scoped and paid, because it involves measuring your work rather than discussing it.
What we look at
What does discovery look at?
Five things, and the order matters. A baseline taken after somebody has started using a new tool is no longer a baseline. None of this grades anyone: the point is to understand the work, not to mark it.
- Sponsor and objective
- One named person who answers for the result, and the constraint in a sentence they would defend in a management meeting. Without both, there is nothing to measure against later.
- Workflow sampling
- A representative sample of the work as it really runs, not the best run of it. Volume per period, labour time including review and exceptions, the rework rate and current spend.
- Current tools and permissions
- What is already licensed, what the assistant would be able to see, and whether the permissions on those sources are right. This is where most timelines actually go.
- Candidate opportunities
- Every task somebody nominates, scored on value, effort and how much of it a person would still have to check.
- Readiness and constraints
- Data quality, contractual and regulatory limits, who has to approve output, and whether the process is stable enough to be worth automating at all.
What you keep
What do you keep at the end?
Six documents, and they are yours whether or not you build anything with us.
| Deliverable | What is in it | Why it is there |
|---|---|---|
| Prioritised opportunity list | Candidates ranked by value and effort, with the ones we recommend against and why. | Stops the loudest request becoming the first project. |
| Baseline plan | What gets measured, on which sample, by whom, before anything is switched on. | Without it, every later number is an argument rather than a finding. |
| Provisional platform shortlist | One or two routes worth investigating, with the questions that would settle the choice. | A shortlist, not a purchase order. |
| Defined pilot scope | The workflow, the people, the sources, the period and who reviews the output. | A pilot with no edge is a pilot that never ends. |
| Success and stop criteria | The numbers that would justify expanding, and the ones that would mean stopping. | Agreed before the pilot, not after the results arrive. |
| Phased roadmap | What follows the pilot, in what order, and what each phase depends on. | So the second decision is easier than the first. |
Scope and fee are set against the work in front of us and agreed before it starts. We do not publish a standard length or a standard price for this, and we do not produce a bespoke specification for free.
A worked example
How would discovery examine a management pack?
A hypothetical, written to show the shape of the reasoning. It is not a client and there are no measured figures in it.
Task
The monthly management pack
Someone senior spends several days pulling numbers out of four systems, writing the commentary and chasing the two people whose figures arrive late.
The finance director nominates it as the obvious AI candidate, because the writing part looks automatable.
Checks
Where the time actually goes
Sample three months rather than take the nomination at face value. If most of it is chasing and reconciling rather than writing, a drafting assistant improves the part that was never the problem.
Whether the sources are fit to be read
A spreadsheet whose column headings change every quarter can defeat an agent quietly and produce a confident wrong answer.
Who reviews the output
Board commentary carries a name on it, so somebody senior reads every word whatever produced the draft. That review time belongs in the after state, and leaving it out is how a case that looked like a week of savings turns out to be worth two days.
Possible scope
The likely output, which is not one project
- A small assistant task for the commentary that a person still signs.
- A data preparation job on the two unreliable sources.
- A note that the chasing problem is a process change rather than an AI one.
That is a roadmap a finance director can defend.
Is this the right starting point for us?
Several teams or tasks
You cannot name one process, because the work is reading, drafting, analysis and client contact across the board. Discovery looks across the company and ranks the candidates against each other.
Discuss AI adoptionOne defined workflow
You can name the costly process in one sentence. Process discovery applies the same discipline to that single workflow and comes back with a scoped build, which is the shorter route.
Explore process discovery
Either way the arithmetic is the same, and you can do a first pass on it yourself before speaking to anyone.
- Check the business caseYour own baseline, and the month it pays back.
- Compare platform criteriaIf the platform question is the one blocking you.
Common questions about AI discovery
What happens on the first call, and does it cost anything?
The first call is free and it is a conversation, not an assessment. We ask what you want to improve, what makes it difficult today, what you already run and who would own the result. At the end of it we say whether a scoped paid discovery is the right next step, or whether something cheaper would answer your question. Sometimes that is reading one of the guides, and sometimes it is fixing a process before any tool goes near it.
Why is discovery paid when the first call is not?
Because the two produce different things. A call produces a view, and it costs us an hour. Discovery produces measured baselines taken from a real sample of your work, a ranked opportunity list, a defined pilot scope and the criteria for stopping, and producing those means days of work inside your business with the people who do the job. That is what the fee covers, and it is why the output is yours to keep and act on with or without us.
How long does discovery take?
It depends on how many workflows are in scope, how reachable the data is and how quickly the people who do the work can give us time. We scope and quote it before it starts rather than publishing a standard length, because a fortnight quoted against a month of work helps nobody.
What if discovery says the project is not worth doing?
Then that is the finding and you have it in writing, with the baseline that produced it. It is a cheaper answer than the same conclusion nine months into a build. Situations where we would expect that answer: the process is about to be replaced anyway, the team is already fully booked with no queue behind it so released hours have nowhere to go, or nobody can write down the rules the work follows.
Do we need to have chosen a platform first?
No. Discovery produces a shortlist rather than consuming one. If you already hold Microsoft 365 licences that narrows the question, and we will say so, but the decision should follow the work rather than lead it.
Who from our side needs to be involved?
The person who owns the constraint, a few of the people who actually do the work, and whoever can answer questions about permissions and data. Not a project team. The people who do the work matter most, because the process as documented and the process as run can differ in ways only they can tell you about.
Tell us what you want to improve
Bring the constraint rather than the tool. We will say whether a scoped discovery is the right next step, and what it would need to cover.
The first call covers what you want to improve, the tools you already run, and whether a scoped paid discovery is the right next step.