we/need/a/dev

Practical automation · 4 min read

Which part of your business should you automate first?

Still copying customer details between apps or chasing the same updates? Learn how to pick one task, decide whether it needs AI, and check that it saves time.

Your team copies the same customer details into two apps, sorts messages by hand, or chases updates every Friday. Those jobs are worth looking at before a big AI project. Pick one repeated task, find out how it works today, and decide what a better version would do.

What to take away

  • Start with one task that takes time each week.
  • Use fixed rules for predictable work and AI when it helps make sense of varied text.
  • Count the time spent checking and fixing results, not just the time saved.

Start with a job you can explain in a few sentences

Imagine a body shop that copies each estimate request from its website into its customer system. An automation could create the record and assign someone to reply. That is a concrete first job. It is an example, not a claim about a client project.

Watch how the team does the job now. Gather a few ordinary requests and a few awkward ones, such as a missing phone number or a request from an existing customer. Agree what should happen in each case.

Start where a person can spot and fix a mistake. Preparing a draft for someone to review is easier to check than letting software quote a repair price on its own.

Does this job need AI?

Copying a phone number into another app usually needs a simple connection between the two systems. It does not need an AI agent. Fixed rules can also assign a request based on a service area or selected service.

AI may help when the work involves reading varied messages, writing a summary, or drafting a reply from your business information. Show the original message alongside the draft so a person can check it.

Decide what the software may do on its own, what needs approval, and when it should ask for help. Keep those rules in the application so they still apply if you change the AI model.

Keep a record of what has already happened

The next run needs to know whether someone already replied, a task is complete, or a customer asked to stop receiving messages. Keep that history in your business system, not only in an agent’s conversation.

Each task should show the customer, the request, what has been done, and who handles the next step. Link back to the source message so the team can check the details.

Choose where each record belongs. Customer details might live in your CRM and work assignments in your project tool. Avoid making the team update two copies of the same information.

What happens if an app is down or a request arrives twice?

An app can accept a request and then fail to send a response. If your automation tries again, it could create the same record twice. AWS explains one way to avoid this: give a request a stable ID so the receiving service can recognize a repeat.

Ask your developer how the automation checks for duplicates and handles a task that stopped halfway through. The answer depends on what each connected app supports.

Put failed tasks somewhere the team can see them, with a reason and a person responsible for fixing them. Repeating a failed action forever will not fix a missing permission or an incorrect email address.

Sources: AWS Builders’ Library: Making retries safe with idempotent APIs

Check whether it actually saves work

Try a small set of examples before turning it on for every request. Include duplicates, missing details, unusual messages, and a connected app being unavailable. Write down what should happen before you run the test.

Count the time people spend reviewing and correcting the result. If a draft takes longer to fix than to write, it needs more work. Check for missed requests as well as completed tasks.

Once the first job works reliably, add more volume or another task. Change one thing at a time so you can tell what caused a problem.

Keep this handy

Your pre-project checklist

  • Which repeated task takes time each week?
  • What starts it, and what should happen at the end?
  • Which steps follow fixed rules, and which need judgment?
  • Where will the record of completed work live?
  • What needs a person’s approval?
  • Does the time saved exceed the time spent checking and fixing results?

Put it into practice

What is your team tired of doing by hand?

Tell us about one repeated task and the apps involved. We can help decide whether a simple connection or an AI agent would make the work easier.

Prefer to write it down? Tell us about your project. No portal account required.

These examples explain how we would approach the work. They are not client results. How we write our guides or send a correction.

Keep exploring

All guides