Try sorting customer enquiries before automating replies

What changed this week?

On 22 October, OpenAI announced general availability of its Decisions API. We reviewed the announcement and developer guidance on 25 October. The service had already been available in public beta, so this is a release update rather than the first appearance of the idea.

What could this mean for a small business?

Our interpretation is that enquiry sorting offers a manageable place to test this type of tool. The proposed output is small enough for someone to inspect: a label and a route to the right person.

Consider an illustrative catering business. It starts with four categories: new event enquiry, existing booking change, supplier message and needs review. The owner decides what each label means and who checks that queue. These are examples, not results from a client deployment.

A message saying 'Can we add ten guests to Friday's booking?' belongs with an existing job. A message asking 'Do you cater for groups of fifty?' may be a new enquiry. But a forwarded email containing both an old booking and a new request needs more care. During the trial, a colleague reviews the suggested label before anything is moved or assigned.

Keep the scope clear. Recognising a booking change does not establish whether the business has capacity, what the revised price should be or whether a dietary request can be met. Those questions depend on current records and the people responsible for the event.

What does the developer discussion add?

X access was limited: OpenAI's profile returned an access error, and targeted discovery did not expose a relevant accessible current discussion. We have not used X reactions as evidence.

Test the sorting before expanding the workflow

We suggest starting with a review-only trial. The system adds a proposed label while the team continues using its normal inbox. No customer message is hidden, deleted or answered because of that suggestion.

Choose a modest sample of past enquiries, with unnecessary personal details removed. Include short messages, forwarded threads, changes to existing work and messages that do not fit a category. Have someone who understands the work label them first. If colleagues disagree, clarify the categories before treating their labels as a reliable comparison.

Count the mistakes by type. Mixing up two routine enquiries may be easy to correct. Sending a change to tomorrow's event into a low-priority queue could be more disruptive. Record missed booking changes separately, along with the time spent checking and correcting suggestions.

Also plan what happens when the service is unavailable. Messages should remain in the ordinary inbox with a clear route to a person. The business should be able to switch off the sorting without losing the original enquiries.

If the trial is useful, a later step might assign routine enquiries to a colleague or prepare an internal task. Keep each extension separate so its effect can be checked. Customer replies, pricing and booking changes can continue to require approval.

For a business receiving only a few enquiries a week, named inbox folders or a clearer website form may be sufficient. The relevant comparison is the total effort of handling the work, including checking the automation and maintaining its connections.

What can you do this week?

Pick one incoming channel and one sorting task. You can prepare this review before deciding whether to commission an integration.

• Write down a few distinct categories and who handles each. Include a needs-review category for messages that do not fit.
• Collect a small set of representative examples, remove unnecessary personal information and record the category a colleague expects for each.
• Test the proposed labels without sending replies or changing bookings. Keep original messages visible in the normal inbox.
• Track wrong labels, missed important requests and correction time. Include service failures and ambiguous messages in the review.
• Decide whether the result is useful enough to continue. If it is, agree on one limited next action, its owner and how to turn it off.

How Context Aware can help

We can help map your enquiry types, write the sorting rules and prepare examples that reflect the work your team receives. From there, we can compare a simple form or inbox rule with an AI-assisted approach.

If an integration makes sense, we can connect the relevant systems, keep customer-facing actions within the agreed limits and make failures visible. We can also help measure the checking and maintenance work, so you have a practical basis for deciding whether to keep it.

Bring us one inbox or enquiry process that is becoming difficult to manage. We can start by identifying where messages get delayed and what a small trial should measure.