An AI plan that says which ideas to fund and which to drop.
Most companies already have a long list of AI ideas and a couple of pilots that never left the demo stage. We sort the list by value and by whether the data and access controls exist to support each idea, then take a small number into production.
AI strategy works out where AI will pay off in your business and what has to be true first.
We score your candidate use cases on value, data readiness, risk and effort, check that the data and access controls really exist, and say whether to build, buy or configure each one. You get an operating model with owners and governance, and a ranked roadmap whose first use case is scoped with an agreed success measure.
you probably need this if
Every department's got an AI idea, and nobody's agreed how to pick which goes first.
A vendor's pitching an AI add-on and nobody's checked whether your data supports it.
Pilots keep starting and none has a success measure written down.
From the first call to a setup your own team can run.
Here's what usually prompts a call about your AI Strategy, what we do about it and in which order, and where you end up.
01 · where it hurts02 · what we change03 · how it runs04 · where you end up
incident board · ai strategy3 open
openEvery department's got an AI idea, and nobody's agreed how to pick which goes first.
recurringA vendor's pitching an AI add-on and nobody's checked whether your data supports it.
blockedPilots keep starting and none has a success measure written down.
exposure
Use case intake and scoringData and access readinessBuild, buy or configureRisk classificationRunning-cost estimatesOperating model
startrehearsehandover
Collect
Score
Prove
Roadmap
we are here
ready to hand over
The steering group works from a ranked list with reasons attached.
The recommendation rests on a side-by-side test on your own cases, with running costs…
Use case register with value, effort and risk scores
Data and access readiness findings per use case
chapter 01 · where it hurts
Why teams reach out
Every department's got an AI idea, and nobody's agreed how to pick which goes first.
A vendor's pitching an AI add-on and nobody's checked whether your data supports it.
Pilots keep starting and none has a success measure written down.
chapter 02 · what we change
How we solve it
Use case intake and scoring
Data and access readiness
Build, buy or configure
Risk classification
Running-cost estimates
Operating model
chapter 03 · how it runs
Our approach
Collect
Score
Prove
Roadmap
chapter 04 · where you end up
What changes
The steering group works from a ranked list with reasons attached.
The recommendation rests on a side-by-side test on your own cases, with running costs for each option.
Use case register with value, effort and risk scores
Data and access readiness findings per use case
what we do
Where we spend the hours
6 areas of work, each one done by an engineer who has run it in production.
We use one intake form and one scoring method. Each idea gets estimates for value and effort, plus a risk rating, so people argue about the inputs instead of their opinions.
Is the data available, clean enough and allowed to be used for this purpose? Do permissions in the source systems reflect who should see what? Retrieval over a SharePoint site with broken permissions just exposes the breakage faster.
A lot of use cases are already covered by AI features in the CRM, ERP or productivity suite you pay for. Before anyone writes code we compare that option with a custom build, on cost, control and lock-in.
For every use case we note which rules could bite. EU AI Act risk tiers, Colorado's automated decision-making law for consequential decisions, and the sector rules if you're in finance, insurance or healthcare.
Token volumes, model prices, retrieval and evaluation costs, estimated per use case. A successful pilot shouldn't turn into a budget surprise at scale.
We settle who approves new AI systems, who owns each one in production, how incidents are handled, and where the AI inventory lives.
capability 01 / 06in production
Use case intake and scoring
We use one intake form and one scoring method. Each idea gets estimates for value and effort, plus a risk rating, so people argue about the inputs instead of their opinions.
Is the data available, clean enough and allowed to be used for this purpose? Do permissions in the source systems reflect who should see what? Retrieval over a SharePoint site with broken permissions just exposes the breakage faster.
A lot of use cases are already covered by AI features in the CRM, ERP or productivity suite you pay for. Before anyone writes code we compare that option with a custom build, on cost, control and lock-in.
For every use case we note which rules could bite. EU AI Act risk tiers, Colorado's automated decision-making law for consequential decisions, and the sector rules if you're in finance, insurance or healthcare.
Token volumes, model prices, retrieval and evaluation costs, estimated per use case. A successful pilot shouldn't turn into a budget surprise at scale.
We pieced these together from AI Strategy situations that come up a lot. They're patterns. You won't find a specific client in them.
case file 01
A survey that produced 120 AI ideas and no order to do them in
composite · not a client story
Beforethe situation
An internal survey produced about 120 AI ideas in a spreadsheet. Every business unit wants its own first, nothing is scored, and the pilots already built from the list have stalled waiting on data access approvals.
What we dothe work
We merged duplicates and scored each idea on value, data readiness, risk and effort, using weights the steering group agreed before scoring started. For the top ten we checked access and classification in the actual systems. One use case was scoped with a success measure, an evaluation set and a named owner.
Afterwhere you end up
The steering group works from a ranked list with reasons attached. The first use case has data access approved before the build starts, which is exactly where the earlier pilots got stuck.
case file 02
Build on a hosted model, or switch on the CRM vendor's AI?
composite · not a client story
Beforethe situation
Service wants AI case summaries and suggested replies. The CRM vendor has an AI add-on to sell, an internal team would rather build on a hosted model, and procurement wants a recommendation in hand before renewal.
What we dothe work
We write down the real requirements: data residency, which fields the AI may read and expected volume. Both options run on the same sample of real cases and are scored on quality, cost per case and admin effort. Licensing and data terms go through review with procurement.
Afterwhere you end up
The recommendation rests on a side-by-side test on your own cases, with running costs for each option. If the vendor feature wins, there's nothing to build.
How AI use cases go from a long list to a funded first project
We scored ideas from the business on data readiness, risk and running cost. Each use case that survived got a build, buy or configure decision and went onto a ranked roadmap with an owner and a success measure.
01 · Inputs
Use case ideasbusiness teams · vendor pitches → Use case scoring · intake
Data readinessSnowflake · Databricks · Fabric → Use case scoring · readiness check
Access controlswho can see which records → Use case scoring · control gaps
02 · Scoring
Use case scoringvalue · readiness · risk · effort → Build · custom need→ Buy or configure · product fit
Risk classificationEU AI Act tiers · internal policy → Use case scoring · risk tier
03 · Decision
BuildClaude · OpenAI API → Ranked roadmap
Buy or configureSalesforce · Dynamics 365 AI → Ranked roadmap
04 · Output
Ranked roadmapowners · success measures → Operating model · governance
Operating modelintake · review · funding
Read the diagram as text
Use case ideas → Use case scoring: intake
Data readiness → Use case scoring: readiness check
Access controls → Use case scoring: control gaps
Risk classification → Use case scoring: risk tier
Use case scoring → Build: custom need
Use case scoring → Buy or configure: product fit
Build → Ranked roadmap
Buy or configure → Ranked roadmap
Ranked roadmap → Operating model: governance
how it runs
A typical engagement
We don't start a step until the last one has produced something you can check.
step 01
Collect
Interviews with business owners and a review of the pilots running now. Then an inventory of AI features you're already paying for.
step 02
Score
We score and rank use cases and share the scoring workbook, so you can re-run it yourselves.
step 03
Prove
A short test of the top candidate, time-boxed, against real data and an evaluation set everyone agreed on beforehand.
step 04
Roadmap
We hand over a sequenced plan, budget ranges, an operating model and the first production release's governance steps.
you keep
What you're left with
We write these for your AI Strategy setup, and they're working documents. Your team should be able to keep things running from them long after we've gone.
inventory01
Use case register with value, effort and risk scores
yours to keep
report02
Data and access readiness findings per use case
yours to keep
document03
Build, buy or configure recommendation per use case
Some clients want a project team, some want a couple of engineers on theirs. Either works. Onsite, nearshore or offshore, with handovers across time zones that keep things moving.
Onsite
Engineers at your sites in the US
Nearshore
Overlapping hours with US teams
Offshore
Overnight progress on your backlog
All time zones
Coverage handed over across time zones
24-hour coverage · US Central time
OnsiteUS business hours
Nearshoreoverlapping US hours
Offshoreovernight progress
Combinedhandover, not hand-off
00:0006:0012:0018:0000:00
roles we staff
AI strategy consultant
AI solutions architect
Data platform architect
AI product owner
Business analyst for AI use cases
due diligence · ai strategy
Questions to ask any partner, including us.
Put these to anyone bidding on your AI Strategy work. The ones who've done it answer differently from the ones who've read about it.
Q1
How will you score use cases, and can our leadership set the weights?
The scoring criteria decide the roadmap. They should be agreed before ideas are scored, so the ranking holds up when a business unit pushes back.
Q2
For the top use cases, has anyone checked the data is reachable and classified?
Data access is where most pilots get stuck. So readiness has to be checked in the real systems, and interviews can only tell you what people think is there.
Q3
When would you tell us to buy or configure instead of build?
If a firm earns its fees by building, expect it to lean that way. Ask to see the criteria that would point to a vendor feature or a config change.
Q4
What is the success measure for the first use case, and who signs it off?
Without an agreed measure, pilots end with a demo and a debate. A number the business owner accepts in advance turns the result into a decision.
Q5
How will you estimate the running cost as well as the build?
Over a year, tokens, retrieval, evaluation and support often cost more than the build. The roadmap should show both figures.
01How long does an AI strategy engagement take?
It depends on how many business units take part and how many pilots exist already. One division, focused, is short. Company-wide takes longer, mainly because of the interviews and the data access reviews.
02Does the EU AI Act apply to us if we're a US company?
Yes. Being based outside the EU doesn't settle it. If your system's output is used in the EU, the AI Act can cover you as a provider or deployer, and placing a system on the EU market brings providers in too. We flag the use cases that might be in scope for your counsel to confirm.
03How do we estimate what an LLM application will cost to run?
Start with expected requests times tokens per request times the model's price, then add retrieval, evaluation and hosting on top. Do it with real prompts. Token counts on real documents tend to come in higher than the guesses.
04What makes AI pilots stall?
Usually data access, unclear ownership or no agreed measure of success. We settle all three before a pilot starts.
05Do you need access to our systems during the strategy work?
Read-only access to metadata and small samples helps a lot. It lets us check data readiness against the real systems instead of relying on interviews. We agree what we can see, and from where, with your security team before we start.
related
Often paired with
Projects on AI Strategy usually end up touching one of these, so they're worth a look.
