An enquiry about office cleaning may include the service, the town and a contact detail, but lack the date and the floor area needed to assess the job. What happens then? In our local demo the answer is not an invented quote, but two simple follow-up questions that a person can review.
The example is entirely fictional. A test enquiry concerns office cleaning in Tyresö and contains a made-up contact address. The prototype gathers the details into an overview, flags that the date and the area are missing and creates a draft: which date would you like, and how large is the area in square metres? The draft is clearly marked as a demo. It is not sent anywhere.
Enquiries like this are everyday life for many small businesses. To see what they look like in reality, have a look at the Önskemål service marketplace, where the jobs are written out in plain text.
We also tried an enquiry where all five details are present: service, town, date, area and contact. Even then a human judgement remains. The prototype does not calculate a price, does not issue a quote and does not book a time.
The most useful part may be knowing when the flow should stop. A relative date such as “tomorrow”, a date in the past, contradictory details or text that tries to instruct the system all lead to a manual check. An identical duplicate does not produce a new draft reply. If the same case ID is used for different content, both drafts are withdrawn. That way uncertain information never looks like a finished answer.
This version uses explicit rules and fixed text templates. No AI model has been run, and no inbox or CRM is connected. We tried 15 synthetic input cases and ran 16 automated tests, which passed. That says something about these particular test cases, but nothing about accuracy on real customer enquiries or how much time a business would save. We explain the difference between rules, auto-replies and a real chatbot in a separate article.
For a business that wants to look at its own flow, a sensible pilot starts with one recurring type of enquiry and anonymised examples. We agree on which details are needed, who reviews the draft replies and how the results will be followed up. The work is billed by the hour, with the hourly rate, the scope and a cap on hours agreed before we start. Any AI interpretation of freer text, or a connection to email and a CRM, is tested separately if it is actually needed.
What automated customer replies usually cost is covered in another article, and three ready-made starters at a fixed price are on our pricing page.