The Alphonse approach

Better thinking.
Better work.

Eight principles for using AI effectively in a business. Apply them with the tools you already have.

Start with the outcome.

Define the business result before choosing a tool. Give the work a clear trigger, an owner, and a definition of done.

In practice

A quote brief is useful when it gives the estimator complete requirements and visible open questions.

Ask: what would a useful result let someone do?

Understand how the work really happens.

Observe the people, messages, records, and exceptions involved. Resolve unclear ownership and conflicting policies before turning them into automation.

In practice

A missed inquiry may be an assignment problem even when every message was delivered successfully.

Map one real case from request to completion.

Make the context explicit.

Provide the goal, current facts, constraints, and examples. Identify authoritative sources and give the model a way to handle missing information.

In practice

Attach the current refund policy when asking for a customer reply, and mark details still awaiting confirmation.

Write the brief you would give a capable new colleague.

Use AI to improve your thinking.

Explore alternative explanations, distinguish facts from assumptions, and examine tradeoffs. A useful conversation can change the question you started with.

In practice

Before buying new software, investigate why the current workflow loses information.

Ask for the strongest alternative explanation.

Challenge the first answer.

Confidence and fluency are not evidence. Ask what would make the recommendation wrong, then check important claims against the world outside the conversation.

In practice

Inspect several missed follow-ups before accepting a model’s explanation of the cause.

Find the observation that could change your decision.

Keep evidence close to the work.

Retain source references, decisions, corrections, and unresolved questions. Keep important business state in a maintained record with an accountable owner.

In practice

An email requesting a Friday delivery stays a request until an authorized person confirms it.

Make every important commitment traceable to a source.

Expand autonomy when results justify it.

Begin with a reviewable result. Test difficult cases, define approval boundaries, and keep a way to stop or recover before allowing actions.

In practice

A useful reply draft can still require review before it promises a refund or is sent to a customer.

Evaluate reading, drafting, and acting separately.

Turn useful learning into a repeatable practice.

Save strong briefs, examples, procedures, and corrections. Give someone responsibility for keeping them current, and measure whether the work actually improves.

In practice

A reviewed handoff format becomes a shared procedure the next team member can use.

Keep the instructions and the evidence from the pilot.