Automation strategy and advisory that ends in a priced build plan.
Each idea is tested against your own records, to show where a model will pay for itself in your operation and where it will not. Every yes comes with a build price and a monthly running cost.
AI strategy consulting should end in a written plan that ranks the processes worth automating and gives each one its data, its model, the cost of a mistake and its build and running costs. Ours takes two to three weeks at one fixed price. The plan is yours, and any team may build from it.
One plan in six parts, all of it yours.
A strategy document earns its fee when every idea carries five facts: the process, the data it already produces, the model that would do the work, the cost of a wrong answer, and the figure it saves or earns.
Ours is one document in six parts, and none of it depends on hiring us next.
- A process map of the real work. What people do all week, from enquiry to paid, on one page.
- A ranked candidate list. Each with the hours it takes today, the share a model can reliably do and the share that stays with a person.
- Build and running costs. The build price, then the monthly bill once it is live, covering inference, hosting and the maintenance nobody quotes for.
- A data readiness read. Whether the records and documents you hold can answer your questions, and what fixing them would take.
- A risk and refusal list. What an error costs at each step, and where the system must refuse to guess.
- A build plan with milestones. Sequenced so the first piece works on its own, and priced as one fixed number if we build it.
Pilots stall for reasons Gartner has named.
In July 2024 Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, and named four causes: poor data quality, inadequate risk controls, escalating costs and unclear business value. A second release, in February 2025, put at 60% the share of AI projects without AI-ready data that organisations will abandon through 2026.
Demos never meet any of them. The plan answers each one before a build is priced, and adds a fifth cause Gartner leaves out: output that never reaches the system of record.
| Why pilots stall | Where the plan answers it |
|---|---|
| Poor data quality | The data readiness read, run on raw exports of your real records. |
| Inadequate risk controls | The risk and refusal list, plus the actions a model never takes alone: spending money, deleting records, first contact with a customer. |
| Escalating costs | A monthly running cost worked out from your real volumes and printed beside the build cost. |
| Unclear business value | The hours each candidate takes today and what remains once it is built. A candidate with no number behind it is dropped. |
| Output nobody uses | Every candidate writes back into the system you already run, so nobody retypes answers from a chat window. |
The end date is written into the proposal.
Late access, records or decisions on your side move it day for day, and nothing else does (§02 of the agreement).
- Day 1, the call. Thirty minutes on what the business does, what a normal week holds and what a win would look like.
- Days 2 to 5, watching the work. Interviews with the people who do it and those who own it, plus exports of the records that matter, under your NDA or ours.
- Days 6 to 8, testing the candidates. The strongest one or two are prototyped on your real data, where a weak candidate fails cheaply.
- Days 9 to 12, the numbers. Build cost, running cost, the saving and the risk. Where the numbers fail, the plan says no and says why.
- Day 13, the handover. We walk through the priced plan with you, and you decide what happens next, including nothing.
Every candidate faces the same five questions.
The order never changes, so you can argue with the ranking.
The plan also names a model for each one. Hard multi-step reasoning goes to a hosted API such as Anthropic Claude or OpenAI GPT when the data may leave your company, and to an open-weight model such as Llama, Mistral or Qwen when it may not. Model choice can stay reversible. Few other decisions in a build can, and the plan keeps this one open.
| Question | What we look at | What rules a candidate out |
|---|---|---|
| Is the work repetitive and checkable? | Volume, how much cases vary, and how fast a person can check a result | Rare judgement calls with no cheap way to check the output |
| Does the data exist? | The records, documents and logs the process already produces | Knowledge held only in someone's head, with no plan to write it down |
| Is the data allowed off your premises? | Contracts, confidentiality clauses, and where you hold your licence | Nothing on its own. It moves the choice to open-weight models on hardware you control |
| What does a mistake cost? | Who is affected, whether it can be undone, and how anyone would notice | Irreversible actions with no person in the loop |
| What does it cost per month? | Your real volumes, priced against the model that fits | A running cost larger than the time it saves |
Included and excluded, line by line.
'Consulting' covers a dozen different services, so ours is set out in full.
| Included | Not included | |
|---|---|---|
| Discovery | Interviews, system access, record exports, process mapping | A six-week audit of every department |
| Prototyping | One or two candidates tested on your real data | A production system |
| Costing | Build cost, monthly running cost, saving and payback | A business case written for investors |
| Deliverable | One written plan with a priced, sequenced build | A slide deck |
| Law and regulation | Every candidate that touches personal data, flagged with the question your lawyer needs to answer | Legal advice |
| Afterwards | You own the plan and may build it anywhere | An exclusivity clause |
For businesses where a mistake is expensive.
The fit is a business whose systems already carry its work, where a wrong figure or a missed step costs real money, and where people spend hours moving data from one system to the next.
Where the records underneath are not yet in order, putting them in order is usually the better first step, and you hear that on the call before you have paid anything. If you already know what you want built, skip this and ask for a build proposal.
The plan leaves you three routes.
Each is fine with us.
- Build it with us at the price the plan already states, with milestone billing, a 12-month warranty while we host it and the code yours on full payment.
- Hand it to another studio. The plan is written for any competent team to quote from, and it works as the brief.
- Wait. The plan keeps, and its numbers tell you when the work becomes worth doing.
The plan says which kind of build comes first: a private model, a knowledge assistant, or plain integration work between systems you already pay for, since the model is rarely the whole answer. Once the first piece is live, a short team enablement block prepares the people who will use it.
Frequently asked questions
What should an automation strategy engagement produce, and how long does it take?
AI strategy consulting should end in a written plan that ranks the processes worth automating and gives each one its data, its model, the cost of a mistake and its build and running costs. Ours takes two to three weeks at one fixed price. The plan is yours, and any team may build from it.
What does a strategy engagement cost?
One fixed price for the scope on this page, stated in the proposal before you commit. It is deliberately a small fraction of a build, since its job is to tell you whether the build is worth doing.
Will you tell us not to use a model at all?
Yes, when that is the answer. A clean CRM, a workflow rule or a few hours of database work often beats a model, and where it does, the plan says so. We decline to sell a build that would never pay for itself.
What is a readiness assessment, and does this include one?
An AI readiness assessment checks whether your data, systems and processes can support a model before anything is built. The plan's data readiness read does that per candidate on your real records, and adds what a readiness score leaves out: costs, risks and a build sequence.
How do you handle our data during discovery?
Under your NDA, or ours if you have none. Exports sit in an isolated environment and train no model outside your company, and on request everything is deleted when the work ends.
We already ran a pilot and it went nowhere. What is different?
A common reason is that the pilot never reached the system of record, and somebody retyped its output from a chat window. Every candidate in our plan writes back into the system you already run, with its running and failure costs stated, or it is left out.
Do you advise on data protection law?
No. We are not lawyers and give no legal advice. The plan flags each candidate that touches personal data, notes which regime may apply, such as the GDPR, Federal Decree-Law No. 45 of 2021 in the UAE or a free-zone law such as those of the DIFC or ADGM, and leaves that question with your lawyer.