Chatbot cost calculator

There Is No Invoice for Answering Messages

Every other cost in a business arrives as a bill. Rent, stock, software, the card machine. Answering messages arrives as somebody being slightly behind all day, which does not appear on any statement, so it never gets compared with anything and never gets a decision made about it. The chatbot cost calculator above exists to turn that into a figure you can put next to the other figures.

On its opening numbers, 400 conversations a month at six minutes each is 40 hours. An entire working week, every month, spent typing the same answers.

What the Chatbot Cost Calculator Compares

Two costs and the distance between them.

The first is what answering by hand costs you: your conversations, multiplied by how long each one takes, multiplied by what an hour of your team is worth. The second is what it costs once some share of those conversations are answered automatically, which is the remainder still done by a person plus the price of the software.

The headline is the gap between them, with the hours handed back beside it, and rows underneath showing the working: the manual cost, the automated cost, the hours, the annual figure and the cost of a single conversation answered by a person.

The Five Inputs

Five boxes, and two of them are genuine estimates rather than facts.

• Conversations a month. Opens on 400. Every thread you answer, not every message.
• Minutes to answer one. Opens on 6. The whole conversation, start to finish, not one reply.
• What an hour of your team costs. Opens on 22.
• Share a chatbot can answer. Opens on 70 per cent, and it is capped at 100.
• What the chatbot costs a month. Opens on 19, which is a figure in a box rather than a quote.

Everything reprices as you type. Nothing is submitted and nothing is stored.

The Arithmetic, Worked Through

The whole model is four multiplications and a subtraction, and the card shows enough of them that anybody can check it.

Four hundred conversations at six minutes is 2,400 minutes, which is 40 hours. At 22 an hour that is 880 a month to answer by hand, and 880 divided by 400 gives the 2.20 a conversation on the bottom row.

Now the automated side. If 70 per cent are handled, 120 conversations are left, which is 12 hours, which is 264. Add the 19 for the software and the automated cost is 283.

Eight hundred and eighty less two hundred and eighty three is 597 a month, or 7,164 a year, and 28 of those 40 hours come back.

Six Minutes Is Probably Too Low

Of the five inputs, the minutes field is the one people underestimate most, and it is worth being deliberate about.

Six minutes sounds generous for a conversation about opening hours. It is not, because a conversation is not one message. It is a question, a reply, a follow-up, a wait, another reply, and a booking, spread over an afternoon and interrupted four times.

The cost is the interruption rather than the typing. Somebody who stops what they are doing eleven times an hour is not doing either job well, and none of that shows up in a minutes field.

So time five real conversations before you trust the number. Not the typing, the whole thing: from the first notification to the point where you stopped thinking about it.

The Seventy Per Cent Is the Number to Argue With

Every other input is something you can measure. The share a chatbot can answer is a claim about the future, and it carries almost all of the uncertainty in the model.

The honest way to estimate it is to look at a week of real conversations and sort them into two piles. In one pile go the questions with a fixed answer: are you open, where are you, what does that cost, do you do this, can I bring a dog, is there parking. In the other go the ones that need a decision, a judgement or a person: my appointment went wrong, can you fit me in before Friday, what would you recommend for this.

The first pile is genuinely large for most businesses and is usually the majority. It is not seventy per cent for everybody, and it is nowhere near seventy for a business whose customers ask complicated questions.

Put your real proportion in the box. A model built on a share you have counted is worth arguing about at a bank. One built on a default is a screenshot.

What an Hour of Your Team Actually Costs

The hourly field is not the wage, and treating it as one understates everything below it.

For an employee it is the wage plus everything that goes with employing somebody, which is meaningfully more than the number on the payslip. If you are working that out properly, the figure to use is what the hour costs the business rather than what lands in the person’s account.

If the person answering is you, the honest figure is different again, and it is usually higher. An hour of the owner’s time is worth what the owner would otherwise be doing with it, which for most small businesses is the work that only they can do. Putting your own wage in that box is the polite version of the answer rather than the true one.

The Payback Line Is the Least Interesting Row

On the opening figures the card says the subscription pays for itself in under a day, and that is arithmetically true and slightly beside the point.

The software cost is trivial next to the labour cost, in almost every version of this model. Nineteen against eight hundred and eighty is not a close call, and it will not be a close call at fifty, or at a hundred, which means payback is never the question you are actually deciding.

The question you are deciding is whether the share is real, and whether the automated answers are good enough that the customers who get them stay. Those are the two things the payback row cannot see and the only two that matter.

What the Model Leaves Out

Three costs sit outside the automated column, and one of them is the one that decides whether any of this works.

• Setting it up. Writing the answers, deciding what happens when nobody matches, testing it. That is a real afternoon and it happens before any of the saving does.
• Keeping it right. Prices change, hours change, services get dropped. An automation nobody maintains starts telling customers things that are no longer true, and that is worse than not having one.
• A bad automated answer. This is the one. A customer who gets a confidently wrong reply and goes elsewhere costs more than the six minutes you saved, and nothing on this card knows about them.

The way to hold all three down is the same: automate the questions with fixed answers, and hand everything else to a person quickly and obviously. The saving on this card comes almost entirely from the boring questions anyway.

What the Chatbot Cost Calculator Cannot Know

Five boxes and some arithmetic, with no view of your business.

• It cannot see your inbox. No message history, no account, no counting. All five figures are yours.
• It cannot measure the share. The seventy per cent is a default in a box, not a finding about you.
• It does not price setup or maintenance. The automated column is the remaining labour plus the subscription, and nothing else.
• It cannot value a lost customer. The cost of a wrong answer is real and it is not in here.
• It assumes every conversation costs the same. In reality a handful take an hour and most take two minutes.
• Nothing is stored. Close the tab and the figures are gone.

The whole card is computed on the server before any script runs, so every row is already correct when the page appears.

Count a Week Before You Model a Year

The gap between a useful version of this and a screenshot is about forty minutes of work, done once.

Count your conversations for a week and multiply by four. Time five of them properly, from notification to finished. Sort a week’s worth into fixed-answer and needs-a-person. Then put those three numbers in, and the fourth, what an hour of the person doing it is really worth.

What usually comes out is a figure that surprises the owner, because the cost was always there and nobody had ever added it up.

Then start with the boring questions. With DMly the repeated ones get answered automatically and the rest land in a shared inbox with a person on them, which is the arrangement this model is really describing: not a robot instead of your team, but your team stopping typing the opening hours.

If most of those conversations arrive from a broadcast rather than day to day, price that side too. The WhatsApp marketing ROI calculator works out what a campaign returns, and the broadcast cost calculator what it costs to send.