The AI that answers your customers

The question always comes in the same shape: «Can you put a robot on there to answer my customers?». Yes, it can be done. The point is that it is almost never the thing worth doing first, and after a year of genuinely trying I have some concrete reasons to give you.

Automate what nobody wants to do, not what you get paid for.

What we kept switched on

Sorting. A hundred enquiries land at once, and somebody has to work out which are urgent, which are for sales, and which are advertising in disguise. A job nobody wants, that a machine does in a second, and where getting it wrong costs little: when in doubt, it passes it to a person.

Summarising. A forty-minute phone call becomes ten lines covering what the client asked for and what we promised them. Before, nobody wrote those ten lines, and six months later nobody remembered a thing.

Drafting. The first version of a reply, the outline of a quote, the first pass at some copy. It is not the finished thing: it is the page no longer blank, which is the part that wastes everybody's time!

The three places where we put people back

Answering the customer unfiltered. The problem is not the quality of the answer: it is that a machine does not know when it is about to say something wrong. It answers an easy question and a question it knows nothing about with exactly the same confidence, and the customer has no way of telling them apart. One wrong answer delivered confidently costs you a client.

Prices. It looks like the easiest thing to automate and it is the most dangerous: one special condition, a discount already agreed, an old price list left lying around, and you have promised a figure you cannot honour. A person signs the quote.

Bad news. A delay, a mistake of ours, a job that cannot be done. These are the moments when a client decides whether to stay, and they need somebody who puts their own face to it, quickly. An automated message apologising does more damage than the delay itself.

The cost nobody puts in the quote

It is not the software licence, which is cheap. It is the time spent checking what comes out, and the discipline to keep checking after three months of everything going fine. Because that is when it happens: you stop watching, and the thing breaks quietly.

The same rule applies as with a new hire: you delegate and you check. The only difference is that a person tells you «I don't know», and a machine does not. If you want to work out what is worth automating in your own business, let's talk.

Where to start, if you want to try

• Take the job you all hate most and repeat every day. That is the first one to automate, not the most visible one.
• Keep a person in the middle for the first two months. Watch how often they have to correct it.
• Measure one thing only, before and after: time saved, or mistakes avoided.
• Do not start with the customer. Start inside, where a mistake gets fixed without anybody noticing.