Meaning, not vocabulary
Rules fire on the words a customer happened to choose, so they miss synonyms and misfire on unusual phrasing. Reading intent means “it will not turn on” and “no power light” reach the same conclusion.
Voice of the Customer
Ticket intent
What a customer types and what they actually need are frequently different things. "It stopped working" could be a dead power supply, a lapsed licence, or somebody who was never trained on it. Reading the intent rather than the wording is what decides whether the routing, the priority and the first response are right or wrong.
- Reads what the customer needs, not the words they used
- Separates requests that carry more than one intent
- Feeds routing, priority and the first response
- Works on free text you already collect, not a new form
The subject line is the worst description of the problem
Customers describe symptoms, not causes, in whatever language they have. The same fault arrives as “unit is dead”, “no power light” and “it will not turn on”, and triage working on wording treats them as three different things.
Category dropdowns move the guess to the customer, which is why almost every queue has a large “Other” bucket. The cost lands downstream: a misread request goes to the wrong team with the wrong priority, and unpicking that takes longer than handling it properly would have.
From free text to a decision
Three decisions get made before anyone works the case. All of them depend on this one.
- Step 1
Read what is being asked
The request is understood as language rather than matched on keywords, so different wordings resolve to the same intent.
- Step 2
Separate the intents
Real messages often carry more than one: a fault report with a billing question, or a how-to buried inside a complaint.
- Step 3
Set routing and priority
Intent rather than vocabulary drives where a request goes and how urgently, which stops a critical fault sitting in a general queue.
- Step 4
Shape the first response
The opening reply is matched to what was being asked, so the customer is not made to re-explain.
Why this is the layer everything else rests on
Get the reading wrong and every decision built on top of it inherits the error.
One message, several intents
A message reporting a fault, asking about a renewal and complaining about the last visit is three pieces of work, and treating it as one is how two get lost.
It is the input to everything downstream
Routing, prioritisation, automated triage and every trend report are only as good as what the request was understood to be.
Why the usual approaches fall short
Most queues are running two of these at once and getting the worst of both.
A category dropdown on the intake form
Moves the guess to the customer, who picks the nearest option and moves on.
Keyword rules and filters
Fire on words rather than meaning, and maintaining the rule set becomes a job nobody wants.
Manual triage by a senior agent
Accurate, and it spends your most expensive people on reading a queue.
Routing by product alone
Says nothing about urgency or what kind of help was wanted.
What things are called
Intent, category and abstraction level get used interchangeably and are three different things.
- Intent
- What the customer is trying to achieve, as distinct from the words they used to ask for it.
- Multi-intent request
- A single message containing more than one thing to be handled.
- Triage
- Deciding where a request goes and how urgent it is, before anyone works on it.
- Categorisation
- Assigning the request to an issue type. Intent is what makes that assignment reliable.
- Abstraction level
- How coarse or fine the categories are. Set separately from intent detection.
- First response
- The opening reply to the customer, and the point at which a misread request becomes obvious to them.
Frequently Asked Questions
Category is which issue type a request belongs to. Intent is what the customer is trying to achieve — reporting a fault, asking how to do something, chasing an order, escalating. The same category routinely carries several intents, and they need to be handled differently.
Where this fits
Auto categorisation tour
What happens once intent is understood: the issue type applied to the request on arrival.
Category abstraction levels
How coarse or fine those categories should be, and what that choice does to your reporting.
TopDrivers
The ranked view of what is driving volume, built on top of this layer.
More Voice of the Customer tours
Voice of the Customer
Run it against last month’s queue
The fastest test is on requests you have already triaged. Give us a sample and we will show you where the reading differs from what your team assigned, and which of the two was right.
Talk to us