Nobody tracks what you promised the customer
Support tools measure how fast you replied and whether the ticket closed. Almost none record the promise you made in the middle, which is the thing customers actually judge you on.
Operations · 7 September 2026 · 6 min read · By The OctaDezx team, Builders of the OctaDezx AI customer care platform
Key takeaways
- Complaints escalate over broken promises far more often than over slow first replies.
- A promise is made mid-conversation and then lives nowhere, which is why it gets missed.
- Response time and resolution rate cannot see promises at all, so dashboards look healthy while trust erodes.
- Store the promise in the words it was made in. A paraphrase loses the part the customer will quote back.
- The customer always remembers. The only question is whether you do too.
Look at the last handful of complaints that turned genuinely hostile. Not the mildly irritated ones, the ones that escalated into refund demands and public reviews.
Most of them will not start with a slow reply. They will start with something you said you would do.
The moment nothing records
Support conversations are full of small undertakings. The replacement ships Tuesday. Someone will call you back this afternoon. We will refund that once the return scans. I will chase the courier and let you know.
Each of these is a real obligation that a real person is now planning around. And in almost every support tool, the moment that sentence is sent it stops existing as anything except a line in a transcript nobody will read again.
The ticket may well close. The metrics look fine. The promise is simply gone.
Your dashboard cannot show you a promise, because there is no field for one. So it shows you a healthy week while three people are waiting for a callback nobody remembers agreeing to.
Why the second complaint is worse
A slow first reply is a disappointment. A broken promise is a different category, because the customer did something with it.
They stayed in for the callback. They told a colleague the part would arrive Tuesday. They stopped looking for an alternative because you said the refund was coming. When the promise fails, it takes their arrangements with it, and it also tells them something worse than that you are slow: it tells them that what you say is unreliable.
That is the point at which a support problem becomes a trust problem, and trust problems are the ones that end in refunds and reviews.
Why the usual fixes do not hold
Most teams recognise the problem and reach for one of three things. Each fails in a predictable way.
- A task in a separate tool. Works, right up until the day it is busy, which is the day the promise matters most.
- A tag on the ticket. Records that something was promised, not what, to whom, or by when. Unactionable a week later.
- A follow-up reminder. Fires on a schedule rather than on the actual due date, so it arrives after the customer has already noticed.
What a commitment record needs
If the promise is going to be tracked properly, four things have to be captured, and one of them is regularly got wrong.
- The promise itself, in the words it was made in. This is the one people get wrong. A paraphrase like customer promised delivery update loses the specific thing the customer will quote back at you, which was Tuesday.
- What kind of promise it is: a shipment, a callback, a refund, a follow-up. Different kinds fail differently and different people fix them.
- When it falls due, so it can surface before it is broken rather than after.
- Whether it was kept, missed or cancelled, which is the only way anyone learns which promises the business routinely cannot keep.
The measurement nobody has
Once promises are recorded, a number becomes available that almost no business currently has: what proportion of what you told customers you would do, you actually did.
It is an uncomfortable number the first time. It is also more predictive of whether customers stay than response time, because it measures the gap between what you say and what happens, which is the thing people are really deciding about when they decide whether to trust a business.
You cannot improve it until you can see it, and you cannot see it until somebody writes the promises down. The advantage of an AI agent handling the conversation is that it can write each one down at the moment it makes it, rather than hoping the pattern gets spotted in a transcript afterwards.
Frequently asked questions
- What is commitment tracking in customer service?
- Recording the specific promises made to customers, such as a shipping date, a callback or a refund, along with when each is due and whether it was kept. Most support tools track how fast you replied and whether the ticket closed, but have no field at all for what you actually undertook to do in between.
- Why do broken promises matter more than slow replies?
- A slow reply is a disappointment. A broken promise is a betrayal of something specific the customer arranged their expectations around. They waited in for the callback or told somebody the part would arrive Tuesday. That is why the second complaint is angrier than the first, and why it is more likely to end in a refund.
- How can an AI agent track promises it made?
- By recording each one at the moment it is made, in the words it used, with the type of promise and its due date. Since the assistant is already handling the conversation, it can write the commitment down as it makes it rather than relying on somebody to spot it in a transcript later.
Where this fits in OctaDezx
Keep reading
- Will AI annoy your customers? Only if you set it up to
The fear that automation will irritate people is reasonable, because most of us have been on the wrong end of a bad bot. Here is what actually makes customers angry, and how to avoid every one of those things. - Can AI handle angry customers?
Emotional conversations are where automation is most likely to make things worse. Here is what AI can safely do with an upset customer, what it should never attempt, and how to draw the line. - Why customers hate repeating themselves, and what fixes it
Being asked to explain the same thing twice is the most common complaint in customer service. It is also completely avoidable. Here is why it happens, and what a shared customer context actually changes.
About the author
The OctaDezx team, Builders of the OctaDezx AI customer care platform. We build OctaDezx, an AI customer care platform used by online stores, restaurants, agencies and clinics to answer customers and take orders around the clock. Everything here comes from running that product and reading real support conversations across those businesses, not from a keyword brief.