It sounds sure. It is wrong.
Support teams keep using the same two words for it: confidently wrong. Steps for a desktop, sent to someone on a phone. A phone number that was never yours. The customer finds out after they act on it.
Yourbotwritesitsownanswers.Thatiswhyitgetsthemwrong.
Ours starts from what your team already said.
It will not make up a price, a policy, or a phone number. If nothing in your history fits the question, a person picks it up instead.
Right answers mean fewer people asking for a human. That is the whole saving.
We will get back to you about setting it up on your support history. By submitting you agree to our Privacy Policy and Terms of Use.
- Sits on top of your CRM
- Nothing to write first
Incoming conversation
Answer base 4,812 repliesCustomer
Can I still change the delivery address after ordering?
Closest answer in your history0.91
Same thing the customer asked aboutMatch
Answered
Yes, as long as it has not shipped. Reply with the new address and we will update it before it leaves the warehouse.
Written by an agent on your team. Sent 26 times since.
Why people ask for a human
Nobodyasksforapersonbecausetheyhatebots.Theyaskbecausetheygotburned.
So people stop trusting it
"Did that answer your question?" No. Again, no. A third time, no. Get burned once and you open the chat and type "human" before you read a word.
And you pay for every one
A person handles it in the end, and that one costs about seven times more. A bot that answers badly does not just fail to save money. It makes more work for the team behind it.
Why the usual fixes fall short
Youcannotfilteryourwayoutofananswerthatwasmadeup.
Turn the confidence score up
A score tells you something close exists. It does not tell you it is about the same thing. An answer about Sunday hours in one city scores high against a question about Wednesday in another, and sails past the cut-off.
Write better prompts
Prompts shape text that is still being written on the spot. The problem is not the tone. It is a fact the bot produced that nobody on your team ever approved.
Add escalation rules
They fire on keywords and mood. So they act after the customer is already annoyed. By then the wrong answer has been read and acted on.
We start from what your team has already said, instead of writing something new each time and hoping it holds.
How it works
Wedonotwritetheanswer.Webuildit.
The wording comes from replies your own team already sent. The details come out of your system, live. Neither part is left to a model to make up.
- 01
We work out what was really asked
"How much is it?" on its own means nothing. Read against the chat so far it means "how much is the yearly plan". That is the question we go looking for.
- 02
We find how your team answered it before
Not a help article. The reply one of your own agents sent to that question, out of your support history. There is nothing for you to write first.
- 03
We fill in this customer's real details
Their order, their plan, their delivery date, read live out of your system and dropped into the reply by code. The model never writes those values, so it cannot invent one.
- 04
Two checks before anything sends
Is there a close enough answer, and is it about the same thing this customer asked about. A similarity score can only tell you the first one.
Then one of two things
The customer is answered
It reads the way your team writes, because that is where it came from. No queue, no wait, and no reason to go asking for a person.
A person picks it up
Before the customer ever sees a guess. Your agent gets the draft, the history and the reason it stopped.
The numbers
Everyquestionthebotcannotsettleturnsintoaperson'sjob.
76-90%what bot vendors advertise41%what really gets solvedHow many customer questions an AI support bot actually settles end to end, next to the number on the brochure.
The rest land on a person. Closing that gap is the entire job, and every point we close is a wage you do not pay.
87%
of people say a company using AI for support must still let them reach a person
Gartner, 3,566 customers, published August 2026
96% vs 83%
of support bosses say the handover keeps the context. And 83% of customers say they end up repeating themselves anyway.
Five9 2026 Business Leaders CX Report
15%
of people say a handover from bot to person has ever gone smoothly for them
Gartner
$1.84 vs $13.50
what one chat costs when the bot handles it, against when a person does. Roughly seven times.
Gartner, Benchmarks to Assess Your Customer Service Costs
20-40 pts
how much better a deflection rate reads than a real resolution rate on the very same bot. Deflection counts the customers who gave up.
Aissist 2026 AI Customer Service Benchmark
26% to 55%
how much more customers trust a bot once a person is reachable. So we leave that door open and answer well enough that few people use it.
Five9 2026 Business Leaders CX Report
72% vs 52%
the same chats, counted two ways. A twenty point gap from wording alone.
Ada, published deployment data
A support bot invented a policy
In April 2025 a software firm's support bot told people about a login rule that did not exist. It said the logouts were normal. A cofounder went public to admit the answer was wrong. People cancelled over a rule nobody had ever written.
Klarna cut too far and went back
Klarna swapped about 700 support jobs for AI, then hired back in May 2025. The CEO said they had gone too far and quality dropped. The averages looked fine. The odd cases did the damage.
Sources: Aissist 2026 AI Customer Service Benchmark; Gartner press release, 4 August 2026, and Gartner customer service cost benchmarks; Five9 2026 Business Leaders CX Report; Ada published deployment data; contemporary reporting on the Cursor and Klarna incidents. Read the Gartner release.
What it costs you
Everychatthatreachesapersonisachatyoupayfor.
Tell us roughly how many you get in a month. We will show you what they cost when people answer all of them, and what is left once the bot takes eighty percent.
5,000
Roughly is fine. Everything that arrives, across every channel you answer on.
Every conversation handled by a person$67,500a month
What your team is paid to answer them
What is left once the bot takes 80%$13,500a month
Only the ones a person still needs to see
Wages you would not be paying
$648,000a year
$54,000 a month, going to people answering questions your team has already answered before.
Gartner puts a conversation a person handles at $13.50, wages included. The 80% is our working assumption, and it is a floor rather than a ceiling: every conversation your team answers goes into the base the bot draws on, so the share it can cover grows with use. This is arithmetic on the number you entered. It is not what we charge and not a promise about your account.
Two ways in
WeplugintoyourCRM,oryoumoveontoours.
Some tools let us plug in and answer your customers inside them. Some write their own replies and cannot be changed. We will tell you which one you are on.
We plug into your CRM
Your system stays where it is. We answer inside the chats your team already handles.
Already on a closed AI agent?
These write their own replies, so there is nothing to sit behind. Those teams switch to us.
Yourteamalreadywrotetheanswers.
- Your team keeps the same inbox and the same tools.
- Fewer chats get pushed through to a person.
- The ones that do arrive with the whole story attached.
We will get back to you about setting it up on your support history.