AI strategy
Put AI where it pays, and prove it.
We help you decide where AI is worth using, test it on real work against today’s numbers, and put what works into everyday use. Sometimes the better answer is a process change, a tool you already own, or no AI at all.
A large company, simplified Example
AI strategy view
Customers
- Marketing
- SalesPilot
- ServiceLater
- Online channels
Operations
- PlanningPilot
- Supply chain
- Fulfilment
- Procurement
Corporate
- Finance
- People
- Legal and risk
- ReportingLater
Foundations
- Data
- Integration
- Infrastructure
- Security
- You get
- A baseline sheet, a scored use-case register, a pilot sheet, a business case and a results report.
- Typical length
- Use-case assessment: 3 to 4 weeks for one function. Pilot: 4 to 8 weeks.
- Priced as
- A fixed fee for the use-case assessment, and for each pilot.
When you need it
- The board wants something real with AI, and a result we can show within a year.
- Our pilots looked good in the demo and never reached production.
- We have a long list of AI ideas and no way to choose between them.
- People are using public AI tools with company information.
- A vendor is selling us an AI product, and we can’t test its claims.
- We have to show what the money already spent on AI has returned.
What you get
- Baseline sheet
- Today’s volumes, times, error rates and costs, with how each was measured.
- Use-case register
- Every candidate scored on value, feasibility and risk, with the reasons.
- Pilot sheet
- Baseline, success line, stop line, decision date, cost at volume and scope, agreed before the start.
- Business case
- Three-year cost for buying, building or neither, including model usage, with benefits split into cash and capacity. Three years, because model prices and products change faster than other software.
- Production checklist
- Integration, checks, monitoring, ownership and the way out.
- Results report
- The same measures as the baseline, after the change.
Pilot sheetIllustrative
| Item | Agreed before the pilot starts |
|---|---|
| Task | Order intake from email, about 300 orders a week |
| Baseline | 6 minutes per order, and 3% of orders with errors, from a timed and checked sample of real orders |
| Success line | At most 2 minutes per order, including the check, and at most 1% with errors |
| Stop line | Errors above today’s 3%, or an average time per order above today’s 6 minutes |
| Between the lines | Decided on the decision date, on the readings |
| Scope | 500 real orders over four weeks. A person confirms every order before it reaches the ERP. |
| Decision date | End of week 4, by the operations director |
| Cost at volume | About 2,000 a month at the success line, against about 5,200 a month for the work today. To switch it off, the desk goes back to typing orders in. |
- Money in the client’s currency. People’s time at 45 an hour; the model and software at 0.05 an order; four hours of monitoring a month. A month is 46 working weeks divided by 12.
Six steps, each with a written result
- 1
Baseline
Measure the work before changing it: volumes, time per case, error rates and cost, from your systems and a sample of real cases.
Result Baseline sheet
- 2
Use cases
Score each candidate on value, feasibility and risk. Some fall away here, which saves money.
Result Scored use-case register
- 3
Pilot
Test on real cases, with a person checking, against the success line and the stop line agreed in advance.
Result Pilot sheet and weekly readings
- 4
Decide
Buy an AI product, build on a model provider’s platform, or neither. The options are compared on three-year cost including model usage, fit, lock-in and the cost of leaving.
Result Business case and decision memo
- 5
Production
Connected to your systems, with a person checking where mistakes are expensive, monitoring of quality and cost, and an owner who answers for it.
Result Production checklist
- 6
Measure again
The same numbers as the baseline, measured the same way.
Result Results report
The work can end at any step, and that is a result too.
Governance, data and people
- Governance
- We work inside your AI governance and model-risk processes. Where you don’t have them yet, we help you set the minimum: an owner for each use, a record of what it does, a review date, and a rule for staff on public AI tools.
- Data
- Before any real data is used, we check the contract and the settings: where the data is processed, whether it may be used for training, and whether it must stay in a given country.
- People
- A person checks the results where mistakes are expensive. Where a system records who corrected what, and employee representatives have a say over such systems, the design is agreed with them before the pilot starts.
- The way out
- Every business case says how you would switch provider or switch it off, and what you keep if you do.
How it runs
- Use-case assessment
3 to 4 weeks for one function
The baseline, the candidates scored, and one or two pilots defined.
- Decision point
Which pilots
You choose the pilots and approve their success and stop lines.
- Pilot
4 to 8 weeks
Real cases, a person checking, readings every week.
- Decision point
Continue or stop
The result is read against the lines agreed at the start.
Business case
Buy, build or neither, with the three-year cost at volume.
- Decision point
Production
You decide whether to put it into everyday use.
Production and result
Planned in the business case, then measured again.
What you provide
- An owner for each process in scope, who decides at each decision point.
- A sample of real cases, anonymised where needed.
- Your data protection and security contacts, and your AI governance contacts where you have them.
- The people who do the work today, for the baseline and to check the pilot’s results.
Length and pricing
- Typical length: a use-case assessment of 3 to 4 weeks for one function, then pilots of 4 to 8 weeks. Production is planned in the business case.
- Priced as: a fixed fee for the use-case assessment and for each pilot; production by release or by milestone.
Questions
Do you build your own AI models?
No. We use established models and platforms, and build what surrounds them: the workflow, the integrations, the checks and the monitoring.
Will our data be used to train someone else’s model?
Not if the contract and the settings forbid it, and we check both before any real data is used. Our pilots keep personal data out, or to the minimum the task needs. If your data must stay in a given country, the pilot uses a provider and a region that keep it there, or it runs without that data.
What happens if a pilot fails?
It stops at the point agreed before it started. You keep the measurements and know why it fell short, which costs far less than finding out after a full rollout.
Is AI strategy separate from our technology plan?
It works best inside it: AI is one of the options in the same plan, judged by the same numbers. When AI is the question on the table, the work starts on its own, with a use-case assessment and then a pilot.
How do you count the benefits of AI?
As in any business case, in two columns: cash and capacity. For AI, cash usually comes from lower overtime, agency costs or licences; hours freed count as cash only when spending falls.
Tell us what you’re trying to decide or fix.
A line or two is enough. We read every enquiry ourselves and reply within two working days.
What happens next
- We reply to arrange a call. If we’re not the right fit, we say so.
- A first call of 30 to 45 minutes with the person who would lead the work.
- A written proposal for the first stage: scope, deliverables, length and fee.
We can sign a non-disclosure agreement, yours or ours, before you share anything sensitive.