To AI or not to AI – that is the question

“It’s so expensive.”

I have heard this again and again over the past months whenever I talked to prospective customers about AI. Let me put the question differently.

“Give me EUR 17,000 and I will give you back at least EUR 100,000 within the first year – what do you say?”

“You’re crazy – that can only be a risky, insane investment.”

But it isn’t … and in this blog series I will show you why, using real use cases. There will be a bit of maths, but only basic school maths. And it will be a little controversial …

Here is our first equation:

Time = Money

Let us first work this equation out for humans – the people who answer the calls, handle the e-mails and so on. We also have to make a few assumptions.

Let us start with the right-hand side of the equation for a human:

In Germany, employer contributions amount to roughly 20 % on top of the gross salary.

What do I mean by overheads? Things like office rent, the occasional new PC, electricity, telephone, software, office cleaning and so on. In short, everything that has to exist and be paid for so that the company can exist and people can actually work there, and which can be apportioned to all employees in some way. Here I assumed 20 %. Sure – that can vary, but I think EUR 600 per month is fairly reasonable.

Now to the left-hand side of the equation – how much time do we get from a human? Again a few assumptions, but reasonable ones:

With our first equation and some simple maths we get:

215 days per year = EUR 50,400 · 1 day = EUR 234.32

Assuming an 8-hour day without breaks and 100 % productivity during those 8 hours (you can decide whether that is reasonable!) we get:

1 hour = EUR 29.29 · 1 minute = EUR 0.49

Now to our first use case. Typical conversations in a contact center for print media could include the following:

• I did not receive my newspaper
• I am on holiday, please do not deliver my newspaper from/until (dates)
• Please send my newspaper to this address

Or in a technical support center, first-level support might simply mean creating a ticket in the ticket system containing the problem and a description of it. And perhaps reading out a few FAQs that could help solve the problem.

Or, for a restaurant, booking a table. Or making an appointment at a doctor’s practice.

Almost every line of business has simple, trivial tasks that have to be done dozens of times a day or more; the list goes on and on.

These usually end up as typical 3-minute calls in a call center, handled by the people there. So each such call handled by a human would cost:

1 call (3 minutes) = EUR 1.47

A well-trained jtel AI voice bot can easily handle all of the tasks above. Let us calculate what a 4-channel jtel voice bot would cost:

* The cost of the speech services is approximate, because it cannot be calculated exactly – it is based on the actual usage time of the service. For speech recognition that is the time the bot “listens” to the caller, and for text to speech how long it “speaks” to the caller. This figure is based on real voice bots in production handling around 5,000 calls per month, and it will do for our purposes, as you will soon see.

Here we also assume the customer has no jtel system. For existing customers the cost would be even lower, since not all of the licences above would be required.

Let us run a few numbers:

Wow! What is going on here?

First of all, I assume the bot is dumb and can do only one job. If only one call comes in per month and that is all the bot can do, then the entire cost of the bot is applied to that one call. If two calls come in, the figure is 50 % of the bot’s total cost, and so on.

For the humans I assume they will do other things for the rest of the day. So the cost of the human for just one call per month is EUR 1.47, and for the bot it is a colossal EUR 1476.

In the real world, however, it looks different. A jtel bot can handle thousands of calls per month. But humans cannot – and you can see that for yourself. Somewhere between 2,500 and 5,000 calls per month we will clearly need more than one human, and above 10,000 even more. Not so with the bot.

Assuming an 8-hour day and the same assumptions as for the human (no breaks – 100 % productivity), the voice bot can handle 640 calls per day – or well over 13,000 calls per month based on 21 working days.

But in fact the voice bot needs no breaks and works 24 hours a day, 365 days a year. It can actually handle a maximum of 1,920 calls per day, an impressive 58,400 calls per month, or a gigantic 700,800 calls per year.

Looking at the table again, we see that the tipping point is at around 1,000 calls per month. So if you have 1,000 calls per month, you can let the voice bot do the work, and in doing so the bot has saved the humans 50 working hours.

Now for the controversial part. Based on our original equation:

Time = Money

you can choose what you want to do:

• use the bot to literally buy yourself more time, so that people can do more productive and more complex things
• or use the bot to save money by having fewer people in the first place

I am personally a fan of option 1.

In closing: is it a crazy, risky investment?

Well, I don’t think so. I think you would be crazy not to invest. Because you can always earn more money through hard work, but time is gone – forever.