AI workflows

Offer an AI-assisted service with human accountability

AI can help draft, organize and transform information. A service becomes useful when a person checks the output, understands the client’s problem and takes responsibility for the final result.

Sell an outcome, not a prompt

Consider a source-checked research brief, an organized knowledge base or a first draft of product descriptions reviewed against approved product facts. Define the input, finished output and acceptance criteria. Avoid offering regulated advice or expert work beyond your competence. Tool familiarity does not substitute for subject knowledge.

Build a reviewable workflow

Start with a small test using public or synthetic data. Separate source collection, drafting, factual verification and final editing. Keep links to evidence next to factual claims. If the system cannot produce a reliable result, narrow the task or complete that step manually. Clients should be able to understand where automation is involved.

Handle information deliberately

Obtain permission before putting client data into external tools. Read the relevant provider’s current data controls and contractual terms. Remove unnecessary identifiers and use approved systems. Free access does not establish permission to process confidential material. A client may require that no AI service be used at all.

Estimate actual labor

Track preparation, corrections and delivery, not just generation time. Start with existing free access if it fits the task, then purchase a plan only when its limits block validated work. An initial pilot might take a few focused sessions; complex integrations require substantially more time and maintenance. Use the time value calculator after the pilot.

Set honest boundaries

AI output can contain invented facts, inconsistent formatting or inappropriate content. Do not promise perfect accuracy or unattended operation. Agree on revision scope and the client’s own approval responsibilities. If repeated checking takes longer than a manual approach, use the manual approach and revise your offer.

Example: a checked supplier research brief

A pilot could compare five suppliers using only their public product documentation. Agree on the fields first: capabilities, geographic coverage, documented constraints and official links. Use AI to organize notes, not to invent missing information. Mark unanswered questions as unknown and let the client decide whether to contact a supplier. Check each factual statement against its cited page before delivery. Record how long the checking took. If the sources are incomplete, explain the limits instead of filling the gaps with confident-sounding guesses.

Examples are illustrative. This guide does not promise income or provide individualized financial, tax or legal advice.

Keep exploring