Sample AI workflow audit: one enquiry journey

Fictional example: a service enquiry is incomplete

This is a worked demonstration using fictional information. It is not a customer case, an audit of your business or evidence of savings. It illustrates the kind of decision material that can be agreed for an advisory engagement. The download has the same boundary.

The fictional business is a small service firm. The task begins when a customer asks for help and ends when a staff member has enough information to choose the next step. Quotations, professional assessment, final pricing and any appointment are outside this scope.

Process map: from message to a responsible person

In the current process, a staff member finds the message, asks for missing information and locates the right colleague. The proposed process collects a contact route, task type, service area and preferred timing first. The customer then sees the agreed next step, not an invented price or available slot.

The process map below is an illustration, not a live booking or messaging system. Approved text can help, but unknown facts and binding decisions go to a person. A calendar appointment can only be described as confirmed after a confirmed result in the agreed system.

Information sources and ownership

The source for opening hours and service area is the business's approved service description. A named owner must date and maintain it. The customer's current message is the source for their request, but does not prove ownership of an email address or grant access to another customer's history.

The example uses only a name, contact route, task type, service area and preferred time. It needs no payment cards, sensitive case files or passwords. If later work needs other data or integrations, assess necessity, permissions and responsibilities first.

Exceptions belong in the deliverable

An unknown price becomes a question for the owner. Conflicting opening hours become a source issue to correct. A request outside the approved service area goes to manual review. A failed connection must not produce a false claim that something has been saved or booked.

When the customer requests a person, the handover needs a clear owner. The team must be able to stop automation and locate the problem. Automated follow-up needs its own permitted contact route, stopping rule and checks; it is not automatically included in this process map.

Tests before deciding on a pilot

Use a fixed set of fictional messages before using customer data. Record expected behaviour, actual results and corrections. The checklist proposes acceptance criteria; it does not claim that these customer tests have already been run.

• Complete enquiry: collect the details once and assign an owner.
• Missing area or contact route: ask for what is needed; do not guess.
• Unknown price or preferred time: give no binding price or booking confirmation.
• Human-help request: show the right contact route and test the handover.
• Duplicate or interrupted connection: avoid duplicate records and show an honest status.
• Incorrect or stale source: stop the affected action, correct the source and retest.

Measurement sheet: baseline, quality and actual effort

Before a trial, record the period, relevant enquiry count, median time to first human follow-up, missing-information rate and staff handling time. After the trial, compare the same task types and disclose changes in staffing and volume.

Also record incorrect answers, corrections and failed handovers. No result fields are filled in because this example has no customer measurement. Released time and an economic calculation are different things; use the workload calculator's scenarios to examine the assumptions.

The sample recommendation and completion point

Start with an approved contact process and one responsible employee. Test anonymised examples and correct uncertain sources. Only then decide on a bounded pilot. A better form may be enough; a new AI system is not a goal in itself.

The deliverable is ready for review when the process, sources, responsibilities, exceptions, open questions and next check are described. You agree the actual scope and fee with Procora. Integrations, implementation and ongoing operation are separate decisions.