Key Takeaways

  • Sort your inquiries into three kinds: answer automatically, draft for review, always to a person.
  • Decide for each kind by asking what a wrong answer would cost and how standard the question is.
  • Start small: a handful of frequent questions answered automatically, everything else drafted or routed.
  • Replies should come from your own texts and rules, in your tone, and say openly that they were sent automatically.
  • Measure how fast you answer and how often a request reaches the right person first time.

How much should the system do on its own?

When businesses think about automating their inbox, they usually picture one of two things: a chatbot that answers everything, or nothing at all. Neither tends to work. Many inquiries repeat themselves, like the status of a job, a missing document, your opening hours or an appointment. A smaller share needs judgement, care or a decision. So the useful question is which inquiries a system may answer by itself, which it should only prepare for you, and which should always go straight to a person.

Three kinds of inquiry

  1. Answer automatically.Standard questions with a stable, correct answer, where a wrong reply wouldn't do real harm. Think opening hours, how to send documents, or what happens next in a routine process.
  2. Draft for review.Inquiries where the answer is usually clear but someone should read it first: a complaint, a question about a specific invoice, anything involving money or a commitment.
  3. Always to a person.Anything urgent, sensitive or unusual, like a damage report, a legal matter, an unhappy key customer or a question the system hasn't seen before. These go straight to the right person, with a summary.

How to decide which is which

Take a few weeks of real inquiries and group them. For each group, ask yourself three things. What would a wrong answer cost you: an annoyed customer, lost money, a legal problem? How standard is the question, and does it have one right answer or does it depend? And do you have the information to answer it, and are you allowed to use it? The cheaper a mistake and the more standard the question, the more the system can do by itself.

Start small

In a first pilot, let the system answer only a handful of your most frequent, lowest-risk questions by itself. It sorts and summarises everything else and either drafts a reply for review or passes it to the right person. That alone usually changes the day. Nobody reads the whole inbox from top to bottom any more, urgent requests sit at the top, and routine answers stop eating your morning. Once your team trusts the drafts, you can let the system do more.

Tone, sources and being open about it

Build the replies from your own texts, rules and templates, so they sound like you and say what you'd say. Say openly that a reply was written automatically: the Federal Data Protection and Information Commissioner (FDPIC) takes the view that people have a right to know whether they are corresponding with a machine. Customers also mind a wrong answer far more than an automatic one. And every automatic reply should make it easy to reach a person.

What to measure

  • How long customers wait for a first answer, before and after.
  • How many inquiries reach the right person the first time.
  • How many automatic answers needed correcting, and why.
  • How many hours your team spends on the inbox each week.

The share that's automated matters least. What you're after is faster, better answers and time back for the work only your team can do. The customer inquiries page shows what this looks like in property management, fiduciary firms, trades and law firms. If you'd like to talk it through for your own inbox, get in touch with Artifex.