Examining AI participation in organisational work

Representing the work

An assessment begins with a defined role or workflow and the information available about it. Sources may include role descriptions, process documents and practitioner accounts.

Language models help organise this information into tasks and steps. The result is a provisional representation of the work. A map generated from a job title alone can suggest questions to investigate, but cannot establish how a particular organisation operates.

Useful steps have recognisable outcomes and enough context to understand who acts, what information they need and what authority they exercise. Work may branch, repeat, run in parallel or change according to the situation.

The level of detail matters. Broad steps can conceal important differences; very fine steps can fragment meaningful work and exaggerate the number of handovers.

What is examined

The methodology first examines characteristics of the work in its stated context, separately from assumptions about a particular AI system. The assessment includes questions such as:

• What are the realistic consequences of poor performance?
• What current and historical context does the work require?
• How stable is the procedure, and how explicit is its reasoning?
• What knowledge, judgement or creative contribution is involved?
• How do relationships and interpersonal interactions affect the work?

These characteristics should not be collapsed prematurely. High-consequence work can be rule-bound. A stable process can still rely on tacit knowledge. Relationship demands can matter even where the information-processing task is straightforward.

The scoring framework and its definitions are being tested and refined. Scores are analytical observations made under specified assumptions, rather than objective measurements of a role or direct recommendations to automate it.

Systems and authority

A candidate arrangement depends on the system being considered: its model, tools, information access, permissions and operating environment. A general chatbot and a system connected to reliable organisational records are different propositions.

Capability assumptions

What the specified system might perform, under what conditions and to what standard. These assumptions need evidence from testing the actual arrangement.

Deployment policy

The organisation’s requirements for review, permitted actions, escalation and decision authority. Policy cannot override legal or institutional restrictions on delegation.

A profile applies declared assumptions and rules to the work map. It can help identify candidates for assistance or automation and the conditions attached to them. Passing a rule or threshold is not proof that a system performs the work adequately.

New technology may justify revisiting an assessment. Existing maps can be reused where their context and assumptions still hold, but changes to the work, evidence, rubric or scoring model may also require reassessment.

Practitioner input

People who perform or oversee the work are essential to checking whether the representation is recognisable, what is missing and where exceptional cases change the process.

The research includes developing practitioner surveys and targeted questions. The aim is to obtain facts and perspectives that can materially change an assessment, rather than ask people to review every score.

A practitioner might be asked whether an action can be reversed, where the required information is held, who has authority to approve an outcome or what makes a particular interaction sensitive.

Agreement that a map resembles the work, acceptance of a proposed arrangement and evidence of actual system performance are different findings. Disagreements can reveal missing context or legitimate differences in responsibilities and priorities.

From a map to an investigation

An assessment is intended to produce a documented basis for deciding what to investigate next. This may include:

• A provisional map with assumptions and unresolved gaps.
• Candidate uses of AI or conventional automation.
• Human roles, authority boundaries and review requirements.
• Conditions that must hold for a proposed arrangement.
• Practitioner questions and priorities for technical testing.

Connected sequences of work matter. Several individually promising steps may be unattractive if they introduce expensive handovers, repeated checking or loss of context.

Human review also needs to be workable. A reviewer needs sufficient evidence, time and expertise to recognise problems, together with authority to intervene. Adding an approval step does not by itself establish a reliable control.

Research and evidence

Appropriate use

The immediate application is structured investigation and pilot planning: helping an organisation examine its work, question assumptions and identify where further evidence would be valuable.

The focus is on work and proposed operating arrangements, not on assessing individual employees. Counts or classifications of steps cannot be translated directly into working hours, staffing requirements or the value of a person’s contribution.

Decisions about deployment require the relevant technical, operational, legal and professional expertise. Privacy, security, employee participation, customer experience and implementation effort must be considered in the actual organisational setting.