Questions 1–5: Is there a worthwhile, well-defined problem?

1. Can you describe the first workflow without saying “AI transformation”?

• A — 0: We want AI, but have not selected a specific process.
• B — 1: We have identified a broad problem, such as missed calls or excessive administration.
• C — 2: We have selected one workflow and documented its trigger, inputs, output and responsible person.
• D — 3: We also have explicit exclusions and a measurable outcome—for example, reducing office handling time for service enquiries without increasing booking errors.

Evidence to collect: A one-page workflow map. Write down what the system must never do, as well as what it should do.

2. Is there enough usable demand or administrative work to justify investigation?

• A — 0: We do not know the volume or value, or have already established that there is no worthwhile opportunity.
• B — 1: Staff report a problem, but we have not checked suitable enquiry volume, workload or delivery capacity.
• C — 2: We have reviewed actual examples and identified either recoverable work we can deliver or repetitive administration we can remove.
• D — 3: We have estimated the addressable volume, excluded work already recovered by existing processes, and checked engineer capacity or a specific use for released office time.

Evidence to collect: For call recovery, separate unique suitable enquiries from repeat calls, existing-job queries and enquiries your team already recovers. For administration, identify the actual task being removed—not an assumed whole-job replacement.

Buying rule: If investigation establishes that the opportunity is too small, “do not buy” is the correct outcome, not a readiness failure to overcome.

3. Can you measure the current process before changing it?

• A — 0: We have no usable baseline.
• B — 1: We have impressions or totals, such as calls answered or quotations sent, but cannot connect them to outcomes.
• C — 2: We have a dated baseline for workflow volume, outcomes, handling time and errors.
• D — 3: We also have agreed definitions, a repeatable collection method and a way to account for differences in demand, job mix, staffing or season.

Evidence to collect: A baseline worksheet showing the reporting period and denominators. For example, record qualified bookings per unique eligible enquiry—not just the number of bookings.

Government guidance on evaluating AI interventions recommends establishing the baseline early and documenting what business-as-usual actually involves. We apply that principle here as a measurement recommendation, not as a requirement to conduct a government-scale study. (gov.uk)

4. Are the records this workflow depends on accurate enough to use?

• A — 0: Important information is missing, contradictory or held mainly in people’s heads.
• B — 1: Records exist, but duplicates, outdated details or inconsistent statuses are common in the sample we reviewed.
• C — 2: We have chosen an authoritative system, checked a representative sample and assigned responsibility for corrections.
• D — 3: Required fields, record matching, update ownership and stale-data handling are defined; missing or conflicting information stops the action or goes to a person.

Evidence to collect: A small, representative record audit. Check customer versus site, payer versus occupier, current quote status, asset identity, service dates and booking restrictions as applicable.

Buyer recommendation: Do not let an AI-generated summary silently become the authoritative record. Specify which system wins when records disagree.

5. Have you compared native software and human alternatives before an agency build?

• A — 0: We have only considered an AI agency proposal.
• B — 1: We know alternatives exist, but have not checked our subscription or current process.
• C — 2: We have compared native configuration, a standard product, human support and bespoke implementation against the same requirements.
• D — 3: We have tested the leading alternatives and can justify the least complex adequate option—even if that means no new AI or no agency.

Evidence to collect: An options table with capability, restrictions, implementation effort and full-year cost.

This comparison is practical: ServiceM8 documents automated quote follow-ups and booking reminders; Commusoft advertises maintenance reminders linked to online booking; Spruce advertises enquiry, estimate and survey workflows for heat-pump installers. These are documented or advertised capabilities, not independently verified savings or proof of suitability for your account. (support.servicem8.com)