An Edge151 Institute standard
The AI Capability Standard
What an organisation needs in place to adopt AI and keep adopting it, and what the people in it need to be able to do. Published in full, applied by the organisation itself, and awarded by nobody.
- VERSION
- 0.1, draft for comment
- ASSESSED BY
- The organisation applying it
- COST
- Free to read and to apply
1. Purpose and scope
Most organisations can reach AI tools easily and still get very little from them. The difference between the ones that do and the ones that do not is rarely the tools. This standard describes what has to be true of the organisation, and of the people in it, for AI to be worth anything.
1.1 Who it is for
Organisations adopting AI in their own operations, particularly those without the scale to buy the expertise. It assumes no specialist technical capability and no dedicated AI function.
1.2 What it covers
How an organisation prepares for, applies and sustains AI use in its own work, and the capability its people need to do that safely and well.
1.3 What it does not cover
The technical evaluation of AI systems, model safety, building AI products for sale, and sector-specific regulatory obligations. Employment and data protection law sit outside it and take precedence over anything stated here.
1.4 Status
This is a published standard, not a certification scheme. Applying it is a matter for the organisation. Nothing in it is awarded, issued or verified by the Institute. Section 5 states what it does not entitle anyone to claim.
2. How the standard is structured
Two parts, applied in parallel. Part one sets requirements on the organisation, in three stages. Part two sets requirements on its people, in three tiers with one endorsement.
Neither part is sufficient alone. An organisation with prepared systems and unprepared people gets poor and risky output. Capable people in an unprepared organisation produce isolated gains that leave when they do.
2.1 What both parts lead to
A business that can put AI to work, and keep doing so as it changes.
The second clause carries the weight. Adopting AI once is a project. Continuing to adopt it as the tools, the work and the rules change is a capability, and that is what this standard describes.
2.2 Terms
| Term | Meaning here |
|---|---|
| Stage | One of the three organisational stages in part one. Stages are sequential. |
| Tier | One of the three levels of individual capability in part two. Tiers are held by people. |
| Endorsement | An additional capability held alongside a tier, not above it. |
| Approved tool | An AI tool the organisation has evaluated and permitted, with stated limits on the data it may handle. |
| Evidence | Something that could be shown if asked: a document, a record, a worked example, a number. |
| Element | A component of the Business Evolution and Resilience Model, which this standard uses as its measurement frame. |
This standard does not introduce a model of organisational capability. It uses the Business Evolution and Resilience Model, and its three stages are a defined route through that model for one purpose.
3. Part one: the organisation
Three stages, taken in order, because each depends on the one before. The failure this standard is written to prevent is attempting stage three with stage one unresolved, where automation inherits unclear work and untrusted data and multiplies both.
3.1 Foundation, which builds confidence
| Element | Requirement |
|---|---|
| Purpose | Leadership can state what the business is trying to become, independently of any technology. |
| Workflow | The work that matters is described well enough that someone could follow it without asking. |
| Information | The data the business relies on has a known source and a known level of trust. |
| Resources | Systems in use are fit for the work, and it is known which are not. |
| Decisions | It is clear who decides what, including who may approve a new tool or a new use. |
Also required at this stage: a published AI use policy with a named owner, a defined set of approved tools with data rules attached to each, and a stated position on when AI assistance is disclosed.
3.2 Augmentation, which builds practice
| Element | Requirement |
|---|---|
| Actors | People hold the tier their role requires, assessed rather than assumed. |
| Work | Tasks where AI pays off have been identified in the actual work, not in the abstract. |
| Information | People can reach what they need to know, and know what they may put into a tool. |
| Decisions | AI-assisted output is checked by someone accountable before it is relied on. |
| Value | At least one change has a recorded position beforehand and an account of what AI contributed. |
This stage runs alongside part two. It cannot be cleared without Actors, which is where the two parts meet.
3.3 Orchestration, which builds momentum
| Element | Requirement |
|---|---|
| Workflow | Work crosses systems and teams by design, not by handover. |
| Resources | Systems can be connected, and something owns the connection. |
| Work | Multi-step work runs with checks built in, not remembered. |
| Measurement | Outcomes are measured continuously, not reviewed occasionally. |
| Learning and Evolution | The business changes how it works as a matter of course. |
Also required at this stage: new uses and new tools pass through a register, a risk screen and an approval; leadership reviews adoption outcomes at a set interval; and the organisation reassesses on a cycle.
3.4 Reading the stages against BERM
Where an organisation has a capability profile, a stage is worth attempting when the elements it names reach Developing, and is cleared when they reach Strength. One element in Critical Constraint stops progress wherever it appears and should be addressed before anything else.
4. Part two: the people
Three tiers and one endorsement. A tier describes what a person can do, not their seniority. Roles are matched to the tier they require, and most people in most organisations need Tier 1 and no more.
4.1 Tier 1, Aware
Held by every employee. The baseline reached before relying on a tool for anything that leaves their own desk. A person at this tier can:
- explain what current AI tools are useful for and where they fail, including that an output can be wrong and confident at the same time
- identify which categories of information must never enter a tool, and apply that to examples from their own work
- name the approved tools for their role and use them on a company account
- check an output before relying on it, and say afterwards what they checked and what they found
- state when AI assistance has to be disclosed, and who to ask when unsure
Assessed by asking where they would not trust it and what they would never paste in, looking for examples from their own work. Someone who treats the tool as reliable, or who sees no data risk, is not at this tier.
4.2 Tier 2, Practitioner
Held by anyone using AI in real work, including managers. The entry outcomes are the floor for holding the tier. A person at this tier can:
- structure a prompt carrying the task, the context, the constraints and the output format required
- judge an output against the standard their work demands, and improve it through another round
- apply the data rules to a case the policy does not cover explicitly, and explain the reasoning used
The established outcomes describe someone whose work others have begun to rely on. They can also:
- build a reusable prompt or pattern so the same task produces consistent quality whoever runs it
- produce structured output that goes straight into another document, system or process
- show a real change in time, quality or consistency on a task they have altered
- get a colleague to a better result than that colleague would have reached alone
Assessed by a worked example start to finish, including the check step and at least one thing the tool got wrong. Tool names without a worked example, or a workflow with no checking step, do not meet the tier.
4.3 Tier 3, Leader
Held by executives and directors who direct and govern AI use. Heavy personal use of the tools is not required at this tier and is not evidence for it. A person at this tier can:
- decide where AI is worth investing in their area, and state what they would stop doing to fund it
- weigh risk, governance and the cost of change alongside feasibility
- set expectations for their team and demonstrate the behaviour they have asked for
- connect AI use to outcomes the business already measures
- interrogate an AI-assisted recommendation, including asking what was verified and by whom
- sponsor adoption by naming an owner, decision rights and a review point
Assessed by an investment decision, real or hypothetical, with the trade-offs stated and a governance position attached. Enthusiasm with no view on risk, or a view on risk with no view on value, does not meet the tier.
4.4 The Builder endorsement
Held alongside Tier 2 or Tier 3 by people who build things others depend on: tools, automations, agents and reusable methods. It is an endorsement, not a fourth tier, because building capability does not carry the accountability Tier 3 does. A person who builds agents is not thereby senior to a director. A holder can:
- choose the right depth of build, weighing effort against the chance of being overtaken by a native feature in a platform already in use
- design so that data rules, human sign-off and disclosure hold by default, not by the user remembering them
- test against realistic failure and fix what breaks before handover
- document a method so someone else can run it, check it and change it
- hand over with a guide, a named owner and a review point, not ongoing personal support
- raise the standard of people at the tiers below
Assessed by something they have built that other people use, with an account of what went wrong during the pilot and what changed as a result. A build nobody else has used is not yet evidence.
5. Applying it, and what it does not mean
5.1 Who assesses
The organisation assesses itself against this standard, and assesses its own people against part two. No external party is involved, and none is required.
5.2 The evidence test
A requirement is met when the organisation could show the proof if asked: the document, the record, the number, the worked example. Anything that could not be shown is not met, whatever anyone believes about it.
For individual tiers, evidence means a worked example, not a claim. A person describing tools they use has not demonstrated a tier. A person walking through a real task, including the check step and something the tool got wrong, has.
5.3 Judging fairly
Assess judgement, not vocabulary. Someone who has never heard the word prompt but checks every output carefully is further along than someone fluent in the language who checks nothing. Record what was seen, not only the tier concluded, so the judgement can be revisited.
5.4 Reassessment
A self-assessment is a position at a date, and its value comes from the second reading. Reassess at a stated interval and compare.
5.5 Partial application
An organisation may apply any part of this standard without the rest. Adopting the policy and the Tier 1 outcomes alone is a legitimate use of it, and is how most organisations should start.
What this does not entitle anyone to claim
Applying this standard does not make an organisation certified, accredited, approved or assured by anyone. There is no award behind it, no register, no badge and no verification.
An organisation may state that it works to this standard and describe what it has in place. It may not state or imply that it has been assessed, approved or recognised by the Edge151 Institute or any other body. An individual may describe their capability in the terms this standard uses, but does not thereby hold a qualification.
This standard exists for the organisation applying it, not for anyone displaying it. Its value is in the work it prompts, and that value does not depend on recognition by a third party.
The Institute defines and maintains this standard. It does not deliver training or consulting against it, and it does not assess conformance. How the Institute works. Enquiries and proposed changes go to hello@edge151.com or through the contact page.