What to prepare before your business adopts AI

Choose the work before the tool

I want AI to help people deliver work more efficiently and provide a more personal service. A useful place to start is a task your team already understands. Describe what happens today, where time is lost and what a better result would look like.

For example, a team might test AI to help draft responses to common customer questions. That is a hypothetical starting point, not a client case study. The aim would be to give staff more time to understand the customer, while keeping them responsible for the response.

Know which information the task needs

Work out where the relevant information lives and whether it is current. A response based on an old policy or the wrong price list can create more work than it saves. Decide which sources the team should use and who keeps them up to date.

For the customer response example, start with approved service information. Identify whether the task actually needs customer records. Avoid adding personal or confidential information simply because it is available.

Check access and security before connecting systems

Connecting AI to business systems is also an access decision. Understand what the tool can read or change, which accounts it uses, and who is responsible for those permissions. Check the provider’s terms and settings for how submitted information is handled.

The controls need to suit the task and the information involved. A trial using public product information is different from a system handling sensitive customer records. If the team cannot explain what information a tool can access, resolve that before expanding the trial.

Keep a person responsible for the result

Agree who checks the output and what they need to check. For a customer response, that could include factual accuracy, the proposed action and whether the language suits the person receiving it. Decide which requests should go straight to a person.

Human review needs context. Someone needs to understand the work well enough to recognise an error or a missing detail. Critical thinking, empathy and accountability still belong to people; approving an answer without reading it does not provide that oversight.

Measure the whole task

Compare the trial with how the task was done before. Include the time spent checking and correcting the output, rather than timing only the first draft. Look at service quality and mistakes as well as time and cost.

Decide whether to continue, change the approach or stop. A faster draft is useful only if the finished work still meets the needs of the business and the person it serves.

Give the team a way to keep learning

Record the agreed tools, information sources and review responsibilities. Give people a way to raise problems and ask for help. Revisit the approach when the tool, the task or the information changes.

This is the connection I want between technology education and adoption. People need enough understanding to use the tool, question the result and recognise when a different approach is needed.

Further reading on AI security.

The Australian Cyber Security Centre’s Engaging with artificial intelligence guidance covers risks and considerations when using AI systems. Use it alongside advice specific to your systems and the information you handle.