Find the repeated handoff, not just the unpopular tool.
Ask staff where they re-enter information, wait for acceptance, chase a status or correct an unclear request. Record the start event, intended finish, people involved, current tools and known exceptions. A complaint about software may reveal an ownership problem that buying another application would leave untouched.
Collect a sample of actual process evidence over a stated period. Count relevant cases, distinguish active handling from waiting and record rework. Do not multiply a guessed time saving by an assumed annual volume and present the result as a business case.
A low-volume process can still deserve attention when an error is consequential. Conversely, a frequent task may be a poor first automation if its rules are disputed or its inputs are too inconsistent. Assess both usefulness and readiness.
Apply readiness gates before assigning a score.
- Ownership: a service owner can define the task and decide how exceptions are handled.
- Permission: the proposed data use, system access and actions have an approved path.
- Recoverability: staff can stop the automation and reconcile pending work.
- Observability: receipt, ownership, errors and completion can be distinguished.
- Service continuity: an approved fallback exists when a connection or tool fails.
If a gate is unresolved, record the missing decision. Do not award enough points elsewhere to disguise that blocker. For example, unclear authority to publish an official response is not balanced out by high request volume.

Compare evidence with a simple priority matrix.
| Criterion | Evidence to record | Favorable starting condition |
|---|---|---|
| Repeated work | Observed case volume and handling steps | A recurring task with repeatable inputs. |
| Current friction | Re-entry, waits, missed handoffs and rework in the sample | A specific failure the change can address. |
| Rule stability | Exceptions, disagreement and frequency of rule changes | The usual route and exception owner are understood. |
| Delivery effort | Access, integration, training and support dependencies | A bounded implementation with available support. |
| Failure consequence | What a wrong or missing action would do | Errors can be detected and contained before harm. |
| Measurement | Baseline events and agreed result definition | A way to compare like cases and inspect unfinished work. |
On a small screen, scroll the table sideways to see every column.
For an initial workshop, rate each criterion Favorable, Uncertain or Unfavorable and attach the supporting note. These labels are a discussion aid, not a validated scoring instrument. If numeric weights are useful, agree on them before comparing projects and show how the choice changes under different weights.
Compare three fictional candidates.
Fictional Cedar Bay planning comparison. Candidate A routes routine internal equipment requests to a named team, with staff accepting ownership and a manual queue as fallback. Its rules are understood, so a pilot could test whether acceptance becomes more visible.
Candidate B automatically decides eligibility for a public benefit. Its consequences and review requirements make it unsuitable for the same routine routing pilot. A separate specialist-led assessment would be needed. Calling both candidates “automation” does not make their risks comparable.
Candidate C adds reminders to a spreadsheet whose departments disagree about the meaning of “complete.” First define the completion event and who confirms it. A reminder can prompt action, but it cannot repair the disputed definition.
The comparison leads to a proposed first experiment, not a claimed success. None of these are Civicfield client outcomes. Use the workflow workbook to make the chosen process explicit.
Choose the smallest useful intervention.
Options include clarifying an owner, improving a shared template, configuring an existing tool, adding a supported integration or building a focused interface. Choose based on the task and constraints. A simpler process change may let staff learn what software is actually needed.
The GOV.UK technology-selection guidance (opens in a new tab) is a useful reference for evaluating technology in its surrounding context. It is UK guidance, not a municipal procurement instruction. Review the alternatives in the build-or-buy guide.
Separate deterministic routing from AI-assisted drafting or classification. If ordinary rules can reliably perform the step, adding model-generated interpretation introduces a different evaluation burden.
Record the pilot decision and stop conditions.
Keep one record containing the selected process, baseline period, included cases, excluded cases, owner, proposed change, fallback and evaluation date. Define a stop condition such as missing requests, incorrect access or unacknowledged failures. Staff must know who can pause the pilot.
Set success criteria before launch and inspect both the result and the work needed to operate it. Faster routing may still leave fieldwork backlogs unchanged. Use the request measurement guide to keep acknowledgement, ownership and operational completion distinct.
Scope implementation and recurring responsibility with the workflow cost model. A sensible first project is one the team can support, evaluate and stop safely, with enough evidence to decide what comes next.


