An AI opportunity map for a 20-person service business should show where client work moves, where it stalls, and which improvement can be tested without destabilizing delivery. It is not a catalogue of tools for sales, operations, support, and finance. The useful map follows a client case from first contact to completed work and payment, then separates immediate workflow candidates from prerequisites and later options.
Draw the business as connected work lanes
Bring together the people who own intake, sales, delivery, support, and finance. Use one recent client engagement as the reference. On a shared map, place the event that starts each lane, the system holding the record, the handoff to the next role, and the accepted end state. Keep the organization chart off the page unless a role makes or approves a decision.
- Intake lane: inquiry received, qualification performed, and next action accepted.
- Sales lane: need clarified, proposal assembled, decision recorded, and delivery handoff completed.
- Delivery lane: work requested, inputs checked, output reviewed, and client acceptance retained.
- Support lane: issue classified, owner assigned, exception escalated, and resolution confirmed.
- Finance lane: billable state confirmed, invoice prepared, discrepancy handled, and payment status updated.
Mark where the same fact is copied, where a queue has no owner, and where an exception returns upstream. Feasibility depends on the limiting system and permission boundary, not just on model capability. The infrastructure bottleneck analysis offers a useful constraint-first analogy for identifying the limiting condition before adding another layer of automation.
Mark friction with evidence, not enthusiasm
For every marked problem, attach a small evidence card. It should name the recent cases reviewed, observable waiting or rework, the person who accepts the result, and any missing fact. Do not estimate savings simply because a task looks manual. A queue with timestamps supports a delay claim. Corrected documents support a rework claim. A complaint without records remains a hypothesis.
Use three labels on the map. A workflow candidate has a stable trigger, usable evidence, and an owner. A prerequisite is missing data, permission, integration, or acceptance practice that must be fixed first. A later option may be valuable but depends on an earlier change. This prevents the map from turning every pain point into a simultaneous project.
Rank the map by sequence, not by department
Compare candidates on consequence, frequency, evidence quality, reversibility, owner capacity, and the cost of failure. Then apply a hard veto: if an unsafe action cannot be held for review, or a sensitive record lacks an allowed handling path, the candidate does not advance. The score organizes discussion; it does not manufacture ROI.
Economic framing still matters. The payback-period article helps turn a feature conversation into a question about when an observable benefit could reasonably offset implementation and operating burden. Use that logic to record assumptions, not to invent a forecast.
Sequence one start, one follow-on and one hold
Consider a hypothetical compact agency whose proposals, delivery notes, support cases, and invoices live in separate systems. The map may show that proposal drafting is visible but not the main delay. Intake classification may be the better start because it has frequent cases, a sales owner, and a reversible recommendation step. A follow-on could reuse accepted classifications in proposal preparation. Invoice automation stays on hold until delivery acceptance is captured consistently.
Frequently Asked Questions
Include workflow triggers, systems, handoffs, decisions, exceptions, accepted end states, evidence cards, owners, prerequisites, and explicit hold conditions.
No. Rank connected workflow changes across the client journey. Some departments may only own a prerequisite or a later follow-on.
Compare consequence, frequency, evidence, reversibility, owner capacity, and failure impact, then veto candidates without a safe permission or review path.
Produce a sequenced map with one starting workflow, a logical follow-on, held options, evidence gaps, accountable owners, and the next decision.
Capability building belongs on the map as a separate prerequisite. The free AI university course roundup is useful when the evidence reveals a skills gap and the team needs low-cost learning options; it does not prove that a workflow will pay back. The final map should name the start, follow-on, held options, evidence gaps, owners, and next decision. For an independent version of that exercise, request an AI Opportunity Report.