Do you need AI, automation, or a better process?

For an engineering workshop, an assistant could turn a customer email into a draft quote request using an approved template. A salesperson still checks the specifications, price and commitments.

For a contractor, document extraction could suggest expiry dates and missing fields for a reviewer. For a processing business, a reporting assistant could assemble approved updates into a draft report. These are examples to assess, not installed products or demonstrated customer outcomes.

Keep the original email, document or approved record next to the AI suggestion. Check names, job references, dates, quantities, amounts and conditions that may have been omitted. Leave unclear fields unresolved and route them to a person.

Try varied and unfamiliar examples, not just the example used to prepare the prompt. Include conflicting dates, missing information and a document from the wrong job. A fluent answer or a high confidence claim is not proof that the result is correct.

Count the time spent reviewing, correcting and handling failures. Keep consequential decisions with an authorised person and agree what happens when the tool or connection is unavailable.

For a first demonstration, use synthetic or approved non-sensitive examples. Name the tools and accounts staff may use, and the information they may enter. Do not treat a paid subscription as permission to upload every document.

OpenAI states that its business products and API do not use business data for model training by default. That is one provider control, not a conclusion about your whole workflow. Assess the relevant contract, retention, access, connected apps and cross-border arrangements.

Count software seats, model usage, integrations, hosting, monitoring, staff review and ongoing maintenance. Use current supplier quotes rather than assuming a fixed rand price.

For an illustrative draft-quote test, compare the time from request to reviewed quote before and after the pilot. Include corrections and failed attempts. If drafting becomes faster but checking takes longer, the net saving may be small. Agree beforehand what evidence would justify expanding, revising or stopping.

Use the Before you automate worksheet to map one process, name its owner and choose a bounded test. The fictional fabrication-shop example shows a useful first workflow that does not need AI.

A short practice exercise can reveal whether the immediate need is clearer instructions, staff training or a system integration. Ask a team member to use the proposed approach on an unfamiliar example and explain how they checked the result.

Aitsa offers separately scoped leadership training, employee workshops, corporate AI roadmaps and implementation. The initial conversation is free; paid work, outputs and terms are agreed in a proposal.