Glossary
The words, defined.
This product uses a handful of terms precisely, and the precision is the point — a case is not a ticket, an incident is not a problem, and a deflection is not always a success. Here is what each one means here.
- Customer Service Graph
- The connected model of a customer's relationship with you: contacts, conversations, cases, purchases, invoices, promises, entitlements, knowledge, resolutions and health. It is called a graph because the relationships carry the meaning — a list of tickets with a customer name attached is not one.
- Case
- One customer's problem, with a type, an owner, a customer, related records, a commitment, a target and a timeline. Distinct from a conversation, which is the messages, and from a ticket, which is usually just the conversation with a number.
- Incident
- One live fault affecting several customers. Their individual cases attach to it, it is communicated once, and resolving it closes all of them. Distinct from a problem, which is the underlying cause.
- Problem
- The recurring root cause behind several incidents. It stays open after an incident closes, carries a countable cost in cases and customer value, and is closed only when contact on the topic actually stops.
- Promise
- A commitment made to a customer, with an owner and a due time, tracked as an object rather than left in a reply. Customers rarely leave because a problem was hard; they leave because a promise passed in silence.
- Internal action
- Work on a case that support cannot do alone — a refund approval, a correction, a fix. A tracked item with a named owner and a due date, so the customer's wait is visible rather than hidden in a private message.
- Entitlement
- What a customer is actually entitled to under their plan: support level, response targets, what is included and what is chargeable. It drives service targets and settles the question two agents would otherwise answer differently.
- Customer health
- One of three explainable states — healthy, needs attention, at risk — each shown with the specific records that produced it. Deliberately not a score, because a number nobody can explain is a number nobody acts on.
- Waiting on
- Whether a case is waiting on the customer, on us, or on a colleague. Three very different states that a single "open" status hides, and the difference decides whose problem the delay is.
- Repeat contact
- A customer getting in touch a second time about the same thing. The most honest measure of whether support is working, and the least flattering — it means the first attempt failed.
- Knowledge gap
- A question customers keep asking that has no verified answer. Counted from cases, help centre searches that found nothing, and questions the assistant declined, then ranked by how much demand each is generating.
- Verified knowledge
- An article with a named owner, a version and a date somebody last confirmed it is true. The assistant answers only from verified material, so lapsed content stops being quoted at customers automatically.
- Provenance
- The records an AI answer came from, named and openable, with the date each was last updated. When the subject is a customer, an unsourced answer is not evidence — it is a claim that will be repeated.
- Fact, inference, recommendation
- Three things the product keeps apart. "Four cases this month" is a fact. "This customer may be losing patience" is an inference and is labelled as one. "The account owner should call" is a recommendation. Presenting the second as the first is the failure that costs trust.
- Autonomy level
- How much the assistant may do without a person, from suggestions only through to handling selected routine categories. Set by you, per category, and reversible with one control — which exists because at some point it will be needed.
- Sensitive case
- A case containing financial, legal, health or comparable personal information a customer disclosed while asking for help. Marked as such, visible to fewer people, excluded from assistant answers by default, and subject to its own retention period.
- Deflection
- A question resolved without a person. Measured honestly here: a customer who read an article and left satisfied is a deflection, and a customer who searched three times and gave up is not — most help centres count both the same way.
- Case type
- Question, problem, request, complaint or incident-related. Not internal taxonomy: each changes what should happen, and a complaint logged as a generic ticket is a complaint nobody reviews.
- Service target
- The response, resolution and update commitments for a case, derived from the customer's entitlement and the priority. Time waiting on your own internal teams counts towards it; time genuinely waiting on the customer does not.
- Team member
- Somebody who works cases. The billing unit. Customers, their contacts and colleagues who only receive notifications or approve the occasional internal action are never counted — worth checking against any product in this category.
- Customer voice
- The aggregate of what customers are asking about, counted across every conversation rather than sampled. Most of what it surfaces is not a support problem: a confusing page, unclear pricing, a missing feature.
- Controller and processor
- For the people who contact your support team you are the controller and we process their data on your instructions. For your own agents' accounts we are the controller. Confusing the two changes who must answer a customer's subject access request, and it is you.
Take better care of every customer.
Give your team the context, knowledge and AI they need to resolve problems properly — and know who needs attention before they ask.
Keep the mailbox you already use · The AI is never metered · [email protected]