Knowledge & practice

LAYER

AI in your customer contact: when does it pay off?

When it comes to AI in customer contact, everyone thinks of a chatbot. That is the riskiest of three places where AI can help.

The short version.

AI in your customer contact pays off when lots of messages come in that someone has to read, sort and answer, and a large share of them are similar. It doesn’t pay off when little comes in, or when every answer is tailor-made.

Don’t think of a chatbot straight away. There are three places where AI can help, and the chatbot is the riskiest.

1. Sorting what comes in.

An email can be a new enquiry, a complaint, a question about an invoice or a receipt from a supplier. A language model reads along and puts it ready in the right place: a lead, a ticket, a task. Someone checks that it’s right.

This is where AI delivers the most with the least risk. Nothing goes to the customer, and you spot a wrongly sorted message straight away.

2. Preparing a reply.

For questions that keep coming back, the model prepares a reply from what you have recorded: agreements, ways of working, earlier answers. Someone reads it, adjusts it and sends it. That saves typing, and every customer gets the same answer.

This only works if the knowledge is recorded somewhere. If the answer isn’t anywhere, the model should say so instead of making something up.

3. Answering on its own: the chatbot.

A chatbot answers without anyone looking over its shoulder. So a mistake goes straight to your customer. That can work for well-defined questions, such as opening hours, the status of an order or how to book an appointment. For advice, custom pricing or a complaint, you want a person.

If you choose a chatbot, make sure it says what it doesn’t know and passes the question on to someone in your team.

We also build a chatbot like that, which only answers from knowledge you have approved and passes the rest on as a lead, separately for your website. Read more about the AI chatbot (Breda) (in Dutch).

Speed matters, but a person can be quick too. A short message saying the question has arrived and when an answer will follow already goes a long way. You don’t need AI for that.

When not to start.

What TrompTech does here.

In TrompTech Layer, a language model reads incoming email and prepares proposals: a lead, a ticket, a task or an expense. An agent can prepare a reply, but a reply to an incoming email or ticket always goes through a person: you can’t switch that to automatic later either. Tickets, communication, meetings and the knowledge base such a reply comes from are in the Service module, from the Core step, €895 a month, excluding VAT.

The chat on our own site is deliberately not AI. It answers a few set questions, says so when it doesn’t know something, and passes the rest on as an enquiry.

TrompTech Layer starts at €399 per month, excluding VAT and with no setup costs.