Solution design

PUBLISHED

Deciding what to build, whether to build it, and how it fits together.

Statements

—

Version

v1.29

Last reviewed

30.08.2026

In scope

Framing a problem in terms that admit an answer

Assessing whether AI is suitable, including establishing that it is not

Requirements elicitation for probabilistic systems

Build, buy and adapt decisions

Architecture choices and the trade-offs between them

Triangulating cost, latency and quality

Designing user experience for systems that are sometimes wrong

Designing for failure modes rather than for the success path

Scoping a pilot so that its result is informative

Total cost of ownership, including evaluation and maintenance

Out of scope

Excluded	Where it lives
Project and product management in general	`Out of framework`
Implementation	`D4`
Assurance that the design was met	`D6`

Boundary notes

Where this domain abuts another, and how the line is drawn

Against D6 Evaluation and Assurance

D5 sets the criteria. D6 establishes whether they were met. A statement about defining fitness is D5; about demonstrating it, D6. The two domains cross-reference heavily at L3.

Against D7 Ethics and Governance

D5 covers whether to use AI. D7 covers whether one is permitted to.

Statements by level

Every identifier is a permanent address. Indicators are normative; they state what would be observed in a person who meets the statement.

L1 Aware

4 statements

D5.L1.01Judgement

State that not every problem is improved by adding an AI system.

Indicators

Identifies a problem in their own work better solved another way

States the question that decides it: what evidence would show AI is the right tool

Recognises that a solution seeking a problem is a common failure

D5.L1.02Knowledge

Describe what a requirement is, and how it differs from a preference.

Indicators

States a requirement in terms that can be checked

Distinguishes what must happen from what is wanted

Identifies a requirement stated so vaguely it cannot be met or failed

D5.L1.03Judgement

Identify who is affected by a system before it is built.

Indicators

Names the people who use it and the people it acts upon

Distinguishes the buyer from the affected party

Recognises an affected party with no voice in the decision

D5.L1.04Knowledge

State why the cost of a system continues after it is built.

Indicators

Identifies evaluation, monitoring and maintenance as continuing costs

States that a model or dependency change forces work

Recognises that an unmaintained system degrades

L2 Applied

6 statements

D5.L2.01Skill

Write requirements for an AI-assisted task that can be checked.

Indicators

States each requirement so that it can pass or fail

Separates requirements from implementation choices

Records who agreed them and when

D5.L2.02Skill

Identify the constraints a solution must work within.

Indicators

Establishes legal, contractual and policy constraints before design

States the constraint’s source, not only the constraint

Refers a constraint that appears to prohibit the task

D5.L2.03Skill

Describe an existing process accurately before proposing to change it.

Indicators

Describes what happens rather than what is supposed to happen

Identifies the informal steps people actually perform

Confirms the description with those who do the work

D5.L2.04Judgement

Compare a proposed AI solution against a non-AI alternative.

Indicators

States the alternative honestly rather than as a straw comparison

Compares on the requirements, including maintenance

Recommends the alternative where it wins

D5.L2.05Skill

Identify the evidence a decision to build would require.

Indicators

States what would have to be true for the solution to work

Identifies which assumptions are untested

Proposes the cheapest test of the riskiest assumption

D5.L2.06Skill

Document a design decision and the reason for it.

Indicators

Records the options considered and the reason for the choice

Records the assumption on which the choice depends

Keeps the record with the system rather than in a message thread

L3 Proficient

8 statements

D5.L3.01Judgement

Decide whether to build, buy, adapt or decline.

Indicators

States the criteria before comparing the options

Includes the cost of dependence and of exit

Records a decision not to proceed with the same rigour as a decision to proceed

D5.L3.02Skill

Design the architecture of an AI-enabled solution, stating its boundaries.

Indicators

States what each component is responsible for

Places the boundary where failure can be contained

Documents the interfaces that other work will depend on

D5.L3.03Skill

Specify the human role in a system before specifying its automation.

Indicators

States what the person is accountable for and what they are given to do it with

Avoids a role that requires vigilance the person cannot sustain

Tests the human step as part of the design

D5.L3.04Judgement

Assess total cost of ownership, including evaluation, monitoring and change.

Indicators

Quantifies the continuing cost, not only the build

States the cost of a forced model or provider change

Identifies the point at which the solution stops being worth maintaining

D5.L3.05Skill

Design for failure, degradation and unavailability.

Indicators

States the behaviour when the model is unavailable or degraded

Distinguishes degradation the user must be told about from degradation that can be absorbed

Provides a path that does not depend on the AI component

D5.L3.06Skill

Establish the acceptance criteria a solution must meet before release.

Indicators

Derives criteria from the requirements and the consequence of error

States who accepts, on what evidence

Provides for the criteria not being met

D5.L3.07Judgement

Assess the effect of a solution on the people who do the work.

Indicators

Identifies work that is removed, added or made harder

Consults those affected before the design is fixed

States the deskilling risk where one exists

D5.L3.08Judgement

Advise on a proposal that should not proceed, and state why.

Indicators

States the ground for declining in terms of evidence, not preference

Offers what would change the answer

Records the advice where it can be reviewed later

L4 Advanced

4 statements

D5.L4.01Practice

Set the organisation’s approach to designing and approving AI-enabled work.

Indicators

Defines the stages at which a proposal is reviewed and by whom

Requires evidence proportionate to consequence

Provides a route for small work that the process would otherwise smother

D5.L4.02Judgement

Establish which architectural dependencies the organisation will accept.

Indicators

States acceptable dependence on a provider, model or hosting arrangement

Requires substitution to be tested rather than asserted

Reviews the position as the market changes

D5.L4.03Judgement

Decide between competing AI investments on stated criteria.

Indicators

Applies criteria known in advance to all candidates

States what is given up by the choice made

Publishes the reasoning to those affected

D5.L4.04Practice

Establish when a system is retired, and act on it.

Indicators

Defines the conditions for retirement at the point of approval

Assigns the authority and obligation to retire

Provides for the work the system was doing

Editorial notes

Known gaps, open questions and contested points

Published because the record is more useful than the appearance of completeness.

The competence employers most consistently report missing, and the hardest to assess by written examination.

Expect this domain to drive the case for practical assessment in credential specifications.

All eight

Domains index

Filtered view

Statements in D5

Proficiency

Levels

This page displays version 1.29 · last reviewed 30.08.2026