Short answer
AI coverage rate is the share of the work in scope that an AI system is set up to handle. In customer service, it is the share of conversations or request types the AI may handle: a measure of breadth, not of problems solved.- Also called
- AI coverage, automation coverage
- Measures
- How much of the work AI is allowed to take on
- Does not measure
- Whether the customer's problem was solved
- Read it with
- AI resolution rate
- Other senses
- Knowledge-base answer coverage; AI's share of tasks in labor research
Coverage is the first number a team watches when it starts with AI in support, because it moves quickly: switch AI on for one more request type and coverage jumps. That is also why it is the easiest number to misread. It says where AI is pointed, not how well it does there.
How to calculate AI coverage rate
There are two ways to count, and they give very different numbers.
- By volume. Conversations in request types where AI may answer, divided by all conversations, times 100.
- By type. Request types where AI may answer, divided by all request types, times 100.
A Shopify store gets 4,000 conversations a month across 12 request types. AI is switched on for order status, shipping questions and cancellations, which together are 2,600 conversations. Volume coverage is 2,600 divided by 4,000, or 65%. Type coverage is 3 of 12, or 25%. The volume figure shows the workload AI can take; the type figure shows how long the tail still is.
Decide up front what counts as covered. A request type where AI answers on its own and one where AI only drafts for a person are different, so report them separately.
With Aide, coverage also widens inbox by inbox: a connected Zendesk, Front or Gorgias inbox moves from Watch only, where Aide drafts in its Activity log and sends nothing, to test senders, then a share of conversations, then everyone.
AI coverage rate vs resolution rate
Coverage counts what AI may attempt. Resolution counts what it actually solves.
| Metric | Question it answers | In the example |
|---|---|---|
| AI coverage rate | How much is AI allowed to handle? | 65% of conversations |
| AI resolution rate | How much does AI solve end to end? | 1,900 of 4,000, or 47.5% |
| Deflection rate | How much never reached a person? | Counts abandoned chats as handled |
AI resolution rate is the number to lead with, because it reflects outcomes. Coverage explains it: if resolution is flat while coverage climbs, AI is being pointed at requests it cannot yet finish.
The practice this page rejects is reporting coverage as if it were results. Switching AI on for more request types is easy. Resolving them, without the customer coming back, is the work.
Answer coverage: does the AI have what it needs?
A second kind of coverage asks whether the AI has the information to answer. Answer coverage, or knowledge coverage, is the share of customer questions the knowledge sources can actually answer: help articles, policies, product pages, order data.
It matters because low answer coverage looks like an AI problem and is a content problem. An agent switched on for returns, with no article on the return window for sale items, will hand off or guess on every sale-item return.
In Aide, the weekly knowledge gaps list measures this directly: what customers ask that the team answers but knowledge doesn't cover, ordered by how many conversations raised it, each with the text and the article it belongs in.
AI coverage in labor research
Outside customer service, AI coverage describes how much of an occupation's work AI performs. Researchers compare theoretical coverage, the tasks AI could do, with observed coverage, the tasks people actually use it for. The gap between the two is usually about integration, trust and oversight rather than what the models can do.
Where Aide fits
In Aide, coverage is a decision made topic by topic. Each of the team's topics has an AI level, AI Agent, Drafts or None, and new topics start at Drafts, so a team widens what Aide handles one request type at a time. Aide's analytics report outcomes rather than coverage: the conversations Aide resolved on its own against those a person finished, with every number opening the conversations it counts.