AI Strategy & Readiness
Find the two use cases that actually pay back.
Six weeks to a ranked portfolio of AI opportunities, each with a cost model, a data readiness score and a decision on whether to build it at all.
- 6-week engagement
- Costed use cases
- Build/buy/skip verdict
What it is
Most AI portfolios are a list of demos
The common failure is not technical. It is a portfolio of twenty pilots, none of which has an owner, a budget line or a number it is supposed to move.
We size opportunities against three things you can verify: the cost of the work being done today, the quality of the data required, and whether a model is genuinely a better tool than a rule. Most items fail at least one test, and saying so early is the point.
- Ranked by payback, not novelty
- Data readiness scored per use case
- Explicit build / buy / skip call
- Risk and compliance assessed up front
Model families we work with
- Claude
- GPT-class models
- Open-weight Llama and Mistral
- Domain-specific classical ML
End to end
Frame to board-ready portfolio
Of candidates we recommend skipping
Median across readiness reviews
Capabilities
What the engagement covers
Opportunity mapping
Workshops with the teams doing the work, producing a long list grounded in real process cost rather than vendor categories.
- Process cost baseline
- Stakeholder interviews
- Long-list generation
Data readiness
An honest score per use case: is the data present, accurate, accessible and legally usable for this purpose?
- Availability audit
- Quality assessment
- Usage rights review
Cost modelling
Inference, retrieval, evaluation and human review costed at realistic volumes — including the parts vendors leave out.
- Token + infra modelling
- Human-in-loop cost
- Sensitivity analysis
Risk assessment
Where a wrong answer is expensive, and what controls the use case therefore needs before it can go near a customer.
- Failure-cost analysis
- Regulatory mapping
- Control requirements
Reference architecture
The platform decisions — models, retrieval, evaluation, observability — that the shortlisted use cases share.
- Model strategy
- Platform blueprint
- Build sequencing
Operating model
Who owns a model in production, who reviews its outputs, and what happens when it degrades.
- Ownership model
- Review process
- Skills gap plan
Use cases
What this looks like in practice
Deployments we have built or scoped, with the sector they landed in.
Support deflection
Sizing how much of a contact centre's volume a retrieval system can genuinely absorb.
- Retail & Commerce
Underwriting assistance
Where a model can accelerate a regulated decision without making it.
- Financial Services
Clinical documentation
Reducing administrative load without touching the clinical record itself.
- Healthcare & Life Sciences
What you leave with
A ranked portfolio
Every candidate scored on payback, readiness and risk.
A defensible number
Cost and benefit modelling you can take to a board.
A first build
One use case scoped in enough detail to start immediately.
A stop list
The ideas we recommend not doing, with reasons.
Process
Six weeks, four phases
01
Frame
Agree the business outcomes in play and the constraints that are genuinely fixed.
- Outcome definitions
- Constraint register
02
Discover
Interviews and process observation across the candidate areas, plus a data readiness audit.
- Long list
- Readiness scores
03
Model
Cost, benefit and risk modelled for the shortlist, with sensitivity on the assumptions that matter.
- Cost models
- Risk assessment
04
Decide
A ranked portfolio, a reference architecture and a scoped first build presented to your leadership.
- Ranked portfolio
- Scoped pilot
Stack
Technologies we use here
Sectors
Where this lands first
Also relevant
Related services
Assurance
How we keep this honest
The commitments that matter when the system is making or shaping decisions.
No vendor incentive
We do not resell model capacity or take referral fees, so the recommendation is not funded by the answer.
Skip is a valid outcome
Several of our readiness reviews have concluded that the right first step is data work, not AI.
FAQ
Questions we are asked
Usually more so. The common finding is that three of the eight pilots are the same use case with different owners, and consolidating them is the fastest available win.
Find out which use cases survive scrutiny
Six weeks to a portfolio you can fund with a straight face.