Worked example · entirely fictional
One workflow. A practical next-step decision.
A fictional organisation wants staff to resolve internal equipment-service requests with less rework. The discovery explores its existing systems before recommending a bounded change.
Illustrative scenario: a fictional composite written to make the point concrete. It does not describe a real organisation, and its figures are not reported outcomes.
The question is about service quality, not the number of agents
A service team receives equipment requests through a portal and email. Staff consult an older asset application, a modern service platform, policy documents and a reporting warehouse. Different teams use “available”, “assigned” and “approved” differently. No baseline has yet been measured.
The sponsor asks: Can we reduce avoidable rework while preserving the required approvals and service quality? Discovery excludes live employee data, purchasing, production changes and automated approval. All examples use synthetic records.
What discovery would establish
| Stage | Fictional deliverable | Evidence still needed and human decision |
|---|---|---|
| Intent | Outcome brief: fewer returned requests without poorer service or inappropriate allocations. | Sponsor agrees definitions; analyst establishes baseline, sample window and quality measure. No improvement is claimed. |
| Current work | Request → staff triage → asset check → manager decision → allocation → service closure, with exceptions. | Frontline staff validate the map and identify undocumented hand-offs. |
| Context | Request, equipment item, availability state, policy, decision, role and evidence definitions linked to their systems of record. | Asset and process owners resolve conflicting definitions and source precedence. |
| Readiness | Data-quality and access gaps; no approved direct write path for AI. | Security and privacy reviewers establish permitted data handling; service owner agrees recovery and support. |
| Options | Compare clearer intake, fixed validation, reporting, a draft-only assistant and a bounded agent. | Sponsor chooses the smallest option justified by evidence and operating effort. |
| Transition | First improve deterministic completeness checks; trial draft summaries only if the information basis and evaluations pass. | Engineers verify in a test environment. Client approves any later release. |
| Measurement | Returned-request rate, resolution time, user effort, quality, exception rate, full operating cost and recovery evidence. | Owners define matched comparisons and review dates before delivery. |
A small contextual model makes dependencies visible
The model says that a request concerns an equipment item, an item has an authoritative asset record, a policy constrains allocation, a manager has a limited decision remit, and a decision cites evidence. Each relationship carries its source and review status. The model also records unknowns, including whether “available” includes items awaiting inspection.
An assistant could draft a summary from information the reviewer may access. It would not decide that an item is safe, approve an allocation or change the asset register. A later agent would need separate justification, explicit authority, tested adapters and recovery.
Compare six options before selecting one
| Option | When it may fit | Trade-off and verification |
|---|---|---|
| Keep current work | Change costs exceed demonstrated benefit. | Establish the baseline and accept existing limitations explicitly. |
| Process improvement | Ambiguous terms and ownership cause rework. | Staff time to agree the process; verify through walkthroughs. |
| Deterministic automation | Required fields and rules can be specified. | Maintain the rules; test exceptions and accessible error recovery. |
| Analytics | Owners lack visibility of delays and failure patterns. | Improve measurement and definitions; reporting does not fix the process by itself. |
| AI assistance | Staff need synthesis across permitted documents. | Evaluate factual support, omissions and reviewer effort. Keep approval and action separate. |
| Bounded agent | A variable sequence of tools offers enough value to justify complexity. | Requires permitted tools, budgets, stop conditions, exact-action authority, evaluation and recovery. Not selected by default. |
The discovery outcome is a decision-ready plan
The proposed first increment is clearer definitions and fixed intake checks using existing systems. A draft-only assistant remains conditional. There are no fabricated savings, delivery dates, customer results or model benchmarks. “Scale” means a design that can be reviewed and reused across other workflows after the first increment supplies evidence.
Explore the full animated enterprise model for a broader fictional estate. It is an explanatory visualisation, not a running integration.
For a separate purchasing sequence, explore procurement, policy gates and MAEB.
YOUR ENTERPRISE. YOUR NEXT CHAPTER.
Turn complexity into
a clear next move.
Bring a business question. We can discuss a discovery scope and the decisions it should support.
Start a discovery conversation