Self-Assessment · Seven Dimensions

Would your AI project survive contact with reality?

Most don’t, and usually not because the technology fails. 22 short questions across the seven areas that decide whether AI sticks, and you will see exactly where the business stands. We will show you your rating and give you the option to have a detailed PDF sent to you.

95% of generative AI pilots show no measurable return (MIT NANDA, 2025)

This assessment needs JavaScript to score itself. The questions are all below if you would like to read them, and we are happy to walk you through it instead: book a call and we will score it with you.

Dimension 1 of 7 Strategy

Strategy, Sponsorship & Value Case

Who owns your AI agenda? No one specific, it's ad hoc IT owns it as a technology project, not a business one An enthusiastic individual champions it, without formal authority A functional or departmental leader owns it, part-time A named board-level sponsor owns it, with dedicated budget and authority

How do you measure AI success across your initiatives, generally? We haven't defined what success looks like We notice informally if people are using the tools We track usage or adoption numbers, not business impact We track some financial or operational KPIs, inconsistently Every initiative has a pre-agreed target, reviewed after launch

Do you have a clear, documented AI strategy? No, we're reacting to individual opportunities as they arise We have some informal thinking, nothing written down We have a written strategy, but it isn't linked to business priorities We have a written strategy aligned to one or two business priorities We have a documented, board-approved strategy with a costed roadmap

Before your most recent AI pilot started, did you define what success would look like? We haven't run a pilot yet, or didn't define anything We had a rough sense, nothing specific We agreed a goal, but not how we'd measure it We defined a measurable target before starting We defined a measurable target and a go/no-go review date before starting

Dimension 2 of 7 Process

Use Case & Process Suitability

How well documented is the process you want AI to touch? Not documented, it lives in people's heads Documented informally, e.g. in emails or someone's notes Loosely documented, varies a lot by person Documented, with a fair number of exceptions Standardised, documented, and rules-based

How was your AI use case chosen? It followed a trend or a vendor pitch It looked easy to demo It was requested by one team, without wider comparison We shortlisted a few options by effort and impact We prioritised systematically against value and feasibility

How prepared is the business to change its operating model or processes if AI reveals a better way of working? Not prepared, we'd expect AI to fit around the current process unchanged Open in principle, but no one owns making that change happen Willing to make small adjustments, not a redesign Willing to redesign the process, with the right business case Process change is expected and planned for as part of any AI initiative

Dimension 3 of 7 Data Quality

Data Quality

How accurate is the data behind your top use case? We haven't checked We suspect it's poor, but haven't measured it We've spot-checked it informally Mostly clean, with known gaps Measured and monitored, with clear ownership

How consistent is that data across systems? Different systems regularly disagree We notice mismatches but rarely resolve them Manual reconciliation is needed often Mostly consistent, occasional fixes One source of truth, reconciled automatically

How complete is the historical data you'd need to train or ground an AI model? We don't know what we have Large, known gaps in the history Partial history, patchy in places Reasonably complete, with minor gaps Complete, with a defined process for filling gaps

Dimension 4 of 7 Data Structure

Data Architecture & Structure

Can the data you need actually be accessed and combined? It's siloed across systems that don't talk to each other Technically possible, but no one's tried Possible, but only with a lot of manual effort Integrated for some systems Integrated via APIs, with clear lineage

How much of the relevant data is unstructured, such as documents, emails, calls and PDFs? Almost all of it, and untouched Most of it, and only skimmed when there's time Most of it, extracted manually when needed A mix, partially processed Structured or well-indexed and ready to use

Who controls access to the data your AI use case needs? No one, access is informal and inconsistent IT can grant access, but there's no clear policy Access is controlled, but slowly and manually Role-based access exists for most systems Clear ownership and role-based access across the systems involved

Dimension 5 of 7 Technology

Application & Technology Architecture

How would you describe your core technology estate? Legacy, fragmented, hard to change Legacy, but we're aware of the problem A loosely connected mix of old and new Modernising, and reasonably well integrated Cloud-based, API-first, straightforward to integrate

Could you run AI beyond a demo, covering hosting, monitoring and scaling? No idea what that would take We'd need external help to even scope it We'd be building it from nothing Some of the foundations exist Yes, proven and already in place

How do your existing systems integrate with new tools or platforms? Rarely, most integration is manual copy-paste work Point-to-point integrations exist, built one at a time Some systems integrate, others don't Most core systems integrate via standard APIs Integration is a designed capability, not a one-off project

Dimension 6 of 7 Delivery

Delivery Capability

What's your track record turning pilots into adopted tools? We've never got past a pilot Pilots tend to quietly stop, without a clear decision One or two pilots, none of them scaled Some have scaled successfully We reliably ship and adopt what we pilot

Do you have a delivery method capable of running an AI project? No formal method Methods exist for other work, not really used here Ad hoc, it depends who's running it A defined method, not yet used for AI A proven method, already adapted for AI/data work

How do you decide whether to keep going, change course, or stop an AI project? We don't, projects continue until someone objects Whoever's closest to it decides informally We review progress occasionally, without a clear trigger We review at agreed milestones Go/no-go decisions are built in at defined checkpoints

Dimension 7 of 7 Governance

AI Governance & Risk

Do you have a policy covering acceptable AI use and data handling? No policy Guidance exists somewhere, but no one reads it Informal guidance only A written policy, not consistently followed A written, enforced policy with clear ownership

Have you assessed your AI use against UK/EU rules such as GDPR and the EU AI Act? Not looked at it Aware it matters, no time set aside for it Aware of it, not assessed Assessed informally Formally assessed and reviewed regularly

If your AI system produced a wrong or biased output, would you know? No, nothing is set up to catch that Only if a customer or employee complained We'd catch obvious errors, not subtle ones Spot-checks happen periodically Ongoing monitoring is in place, with a clear escalation route