AI Readiness Assessment
Most AI projects fail on data rather than on models. The algorithm is rarely the hard part; having information that is complete, consistent and reachable usually is. We assess what your data and processes can actually support before anybody buys anything, and most clients leave with a shorter list of use cases than they arrived with and a much clearer view of which ones will pay.
1
Data
Quality and reachability
2
Process
Where AI would fit
3
Capability
Skills and capacity
4
Cost
Benefit per use case
5
Prioritise
A ranked shortlist
How We Assess
01
Data Quality and Availability
We examine the data you actually hold rather than the data described in a system manual. Completeness, consistency of coding, historical depth and whether records can be reached programmatically all determine what is feasible.
- Completeness and consistency profiling of key datasets
- Historical depth available for training and validation
- Whether data is reachable by API, database or export only
- Labelling and structure required for supervised approaches
02
Process Assessment
AI earns its place where volume is high, rules are stable and the current process is expensive in human time. We look for those conditions rather than starting from a tool and searching for somewhere to apply it.
- Processes with volume and repetition worth automating
- Stability of the rules governing each process
- Current cost in time and error rate
- Processes where automation would be inappropriate
03
Capability and Governance Readiness
Capability is not only technical. Somebody has to own the output, judge whether it is right and act when it is not. Governance has to exist before deployment rather than being retrofitted afterwards.
- Internal skills available to operate and maintain
- Ownership of model output and decisions
- Governance and approval framework in place
- Change readiness within affected teams
04
Cost and Benefit Modelling
Each candidate use case is costed against realistic benefit. Build cost is usually the smaller number; ongoing operation, monitoring and retraining are what determine whether it repays over time.
- Build and implementation cost per use case
- Ongoing running, monitoring and retraining cost
- Benefit quantified in time, error reduction or capacity
- Payback period and sensitivity to volume
05
Prioritised Recommendation
A ranked shortlist with an honest recommendation, including the use cases we would advise against. Telling a client not to build something is frequently the most valuable output of the assessment.
- Use cases ranked by return and feasibility
- Recommended starting point with rationale
- Use cases explicitly not recommended, with reasons
- Sequenced roadmap with dependencies
What You Receive
- Data quality profile across the key datasets
- Process assessment identifying viable candidates
- Capability and governance readiness findings
- Cost and benefit model per candidate use case
- Ranked shortlist with a recommended starting point
- Use cases explicitly not recommended, with reasons
Indicative Timeline
A readiness assessment normally runs three to four weeks. Data profiling takes the time; obtaining extracts from source systems is usually the constraint rather than the analysis.
- Scoping and stakeholder interviews: three to five days
- Data profiling and quality analysis: one to two weeks
- Process and capability assessment: one week
- Cost modelling and reporting: one week
What We Examine
Readiness spans data, process, people and governance, and a weakness in any one of them determines the outcome.
Data Quality
Completeness, consistency and historical depth, which set the ceiling on what is achievable.
Integration
Whether systems expose data programmatically or only through manual export.
Process Fit
Whether volume, repetition and rule stability justify automating at all.
Capability
Whether anyone internally can operate, judge and maintain the result.
Governance
Whether policy, approval and accountability exist before anything is deployed.
Economics
Total cost including running and retraining, weighed against quantified benefit.
Frequently Asked Questions
Do we need AI at all?
Frequently not, or not yet. Plenty of problems presented as AI opportunities are better solved by fixing a process, integrating two systems or automating a rule. We say so when that is the case, because a failed AI project is more expensive than not starting one.
How much data do we need?
It depends entirely on the use case. Document extraction can work from a few hundred examples. Forecasting generally needs several years of history to capture seasonality. Part of the assessment is establishing whether you have enough for what you want.
Our data is messy. Is that a blocker?
It is a cost rather than a blocker, and it is better discovered now. Data cleansing is usually the largest line item in an AI project, and quantifying it early is what stops a project stalling halfway through.
Can we use off the shelf AI instead of building?
Often yes, and it is usually cheaper. Where an existing product does the job we will say so. Building is warranted where the process is genuinely specific to you or where the data cannot leave your environment.
What does an AI project actually cost to run?
More than most expect, because inference, monitoring and periodic retraining continue indefinitely. That is why the cost model includes running cost rather than only build cost, and why some use cases fail the test on economics alone.
How does this relate to AI governance?
Readiness asks whether you can. Governance asks on what terms you should. Organisations deploying anything material need both, and the governance framework should exist before the first model goes live.
Related Services
This sits inside our AI Services practice. Related work: AI Governance and Policy for the framework that has to exist before deployment, and Data Integration and Migration where the data first has to be made reachable.
