A pilot without a thesis
The board wants the advantage AI promises, so something gets built quickly. It demos well and proves nothing, because no one agreed in advance which number it was meant to move. The next budget conversation starts from zero.
Find where AI actually pays, before you buy a single tool.
Most AI programmes start with a licence and end with a pilot nobody uses. We start with the work your company already does: a ranked list of use cases, the numbers behind each one, and the order to build them in.
Where it stalls
The technology is ready. The decision is not.
Tools without a process
Licences land in inboxes and the work stays exactly as it was. A handful of people get good at prompting, everyone else goes back to the old way, and the spend shows up as a line item with no counterpart anywhere in the results.
The cost of deciding slowly
AI does not change companies through announcements. It changes them through processes that run without being asked. Every quarter spent deliberating is a quarter a competitor spends compounding a head start you then have to buy back.
What the numbers say
The gap is not access. It is redesign.
One person
45 min
a day, returned to someone who genuinely knows the tools
Not saved in the abstract. Returned: drafting, rewriting, summarising and reformatting collapse from an afternoon into a few minutes. Across a forty person team that is roughly seven thousand hours a year moved off admin.
OpenAI and MIT Sloan Management Review
One dot is five minutes of an eight hour day. Nine of them come back.
One task
25% faster, 40% higher quality
on the tasks that were actually redesigned around the model
The same study found the opposite effect outside that boundary: on work the model was wrong for, performance dropped. Which tasks qualify is not a matter of taste, and picking them is most of the job.
Harvard Business School and Boston Consulting Group
- Time to finish+25%
- Output quality+40%
Measured against the same task done the old way. The whole company
1.7x revenue growth, 1.6x EBIT margin
for the companies that changed the operating model, not just the toolset
Buying licences correlates with nothing. The separation shows up in companies that rebuilt processes around the capability, gave the results an owner, and measured them the way they measure everything else.
BCG Global Study
MarketAI leaders 11.711.6Revenue growthEBIT margin
What you get
A decision you can defend
A ranked map of where AI pays
Every candidate process scored on value and on how hard it is to change, with the ones not worth doing marked as such. The order matters more than the list: the first build decides whether there is a second.
A business case per use case
Hours today, hours after, build cost and the point at which it pays for itself. Sourced from your own volumes and rates, so finance can check the arithmetic instead of trusting a vendor benchmark.
Rules your legal team can sign
What data may go where, which categories never leave the building, where the human decision stays mandatory and what gets logged. Written as policy people can follow, not as a caveat slide.
A plan your team can build from
Sequenced work, named owners, the integrations each step needs and what has to be true before the next one starts. Written so another supplier could execute it, which is the only honest test of a plan.
How we work
Scoped so you can stop after any level
Level 1 - Opportunity audit
Where AI pays in your company, ranked and costed
Two weeks from kickoff to a decision the board can actually take: what to build, in what order, and what it is worth.
- 2 weeks end to end
- 4 to 6 interviews
- One half day workshop
- No tooling or licences required
- Process inventory. We walk the work as it is done, not as the process map says. Volumes, handoffs, the spreadsheet everyone maintains privately and the approval that adds four days.
- Use case shortlist. Every candidate scored on value and on feasibility, including the ones we recommend against. A short honest list beats a long flattering one.
- The numbers behind each one. Hours today, hours after, build cost and payback, derived from your volumes and your rates rather than from an industry benchmark.
- Data and risk boundaries. What each use case would need to touch, what that means under GDPR, and which options are closed before anyone gets attached to them.
- Build sequence. The order to build in, and the reasoning. The first one has to be small enough to finish and visible enough to matter.
- Board readout. One session, the full argument, and the questions your CFO will ask answered before they are asked.
You leave with
A ranked, costed use case map with a build sequence and a risk boundary, in a form you can take to the board without a translation layer.
Level 2 - Business case and roadmap
The shortlist turned into a funded plan
For companies that already know roughly where the value sits and need the case, the sequence and the integration reality nailed down.
- 3 to 4 weeks
- Finance and IT in the room
- Vendor evaluation included
- Runs from an existing audit or from scratch
- Baseline that survives scrutiny. Current cost and cycle time measured rather than estimated, so the after figure has something honest to be compared against.
- Build, buy or wait. For each use case, the three options costed side by side. Waiting is a real answer and we will say so when it is the right one.
- Integration reality check. What your systems actually expose, what that costs to work with, and which use cases quietly depend on a migration nobody has scheduled.
- Roadmap with owners. Sequenced work with named owners and the dependencies between steps, sized so each one can be cancelled without stranding the next.
- Measurement plan. The metric per use case, where the number comes from, who reads it and what happens when it does not move.
You leave with
A funded roadmap with owners, dependencies and a measurement plan, plus a build or buy recommendation per use case.
Level 3 - Governance and operating model
Rules, owners and a cadence that outlast the project
The part that decides whether adoption survives the first quarter in which nobody is watching.
- 4 to 6 weeks
- Policy and training materials included
- Legal and compliance workshops
- Designed to be maintained without us
- Usage policy. What is allowed, what is forbidden and what needs review, written in the language your employees actually use rather than in legal register.
- Review and escalation. Where a human decision stays mandatory, who signs it, and how an exception is raised without going around the rule.
- Ownership model. Who owns each workflow after go-live, what they are accountable for, and what happens when that person changes role.
- Operating cadence. The recurring review that keeps results visible, with an agenda short enough that it still happens in month nine.
- Enablement. Materials your own people can teach from, so competence does not stay concentrated in whoever happened to be in the pilot.
You leave with
A signed usage policy, a named owner per workflow and a review cadence that keeps working after the engagement ends.
Level 1 - Process deep dive
One process, measured properly, end to end
For teams that already know which process hurts and need it taken apart before anything is built on top of it.
- 1 to 2 weeks
- One process, one team
- Observation over interviews
- No commitment to build
- Walk the work. We sit with the people doing it. What the process map claims and what happens on a Tuesday afternoon are rarely the same document.
- Time and volume baseline. Measured, not estimated. Without this the after number is a story rather than a result.
- Failure modes. Where the process breaks today, how often, and what the rework costs. Some of it turns out not to need AI at all.
- Redesign options. Two or three ways the work could run instead, each with what it would take and what it would give back.
- Recommendation. Including, when it applies, the recommendation to fix the process first and revisit the model later.
You leave with
A measured baseline, a mapped set of failure modes and a costed redesign recommendation for one real process.
Level 2 - Pilot design
A pilot with a thesis, a boundary and a stop condition
Designed so that a negative result is still a useful result, which is what separates a pilot from a demo.
- 2 to 3 weeks of design
- Success criteria agreed up front
- Runs with your data and your team
- Stop condition written down
- The thesis. One sentence stating what this pilot would prove, written before anything is built and signed by whoever pays for it.
- Scope boundary. What is deliberately excluded and why. A pilot that grows during the pilot cannot be evaluated at the end of it.
- Evaluation design. How output quality gets judged, by whom, against what, and with which sample sizes. Agreed before the first result arrives.
- Data handling. What the pilot may touch, how it is stored and how it is disposed of, cleared with legal before kickoff rather than after.
- Stop condition. The result at which we recommend stopping. Naming it in advance is what makes the eventual answer credible.
You leave with
A pilot design with a written thesis, agreed evaluation criteria, a data handling boundary and a stop condition.
Level 3 - Scale readiness
What has to be true before this runs for everyone
The audit between a workflow that works for one team and a workflow the company depends on.
- 3 to 4 weeks
- IT and security involved
- Cost model at full volume
- Handover documentation included
- Load and cost at volume. What the workflow costs at ten times the pilot volume, which is where a per unit price that looked trivial stops being trivial.
- Failure and fallback. What happens when the model is unavailable or wrong, and who notices. A workflow with no fallback is an outage waiting for a date.
- Access and audit. Who can run it, who can change it, and what record exists afterwards. Written to satisfy an auditor, not to satisfy us.
- Support model. Where a user goes when it misbehaves, and how that route stays staffed once the project has a completion date behind it.
- Handover pack. Documentation written so your team, or another supplier, can maintain the workflow without calling us.
You leave with
A scale readiness assessment with a cost model at full volume, a fallback design, an access and audit policy and a handover pack.
Book a working session
Ready to see where your numbers move?
Pick a time. We map the two or three processes worth measuring first, and tell you honestly whether consulting is what you need or whether you already know enough to build.