Keywords #

While Topic Analysis and Sentiment are driven entirely by AI, Keywords give you direct control over what you want to monitor. A keyword rule defines one or more specific words or phrases that ~.Dimensions.~ should look for in transcribed calls. When a match is found, the call is tagged and the matching text is highlighted in the conversation view.

Keywords bridge the gap between open-ended AI discovery and targeted, policy-driven monitoring. They are particularly useful when you need to know whether something specific was said — or not said — on a call.

For information on how to create and configure keyword rules, refer to the Keywords configuration section.

Getting the Most Out of Keywords #

Be specific with names #

Give each keyword rule a clear, concise name — this name appears as a tag on every matched call, so it needs to be meaningful at a glance. 'Cancellation Mention' is more useful than 'KW Rule 3'.

Use the Speaker setting #

The 'Speaker' setting is one of the most powerful ways to make keyword rules precise. A phrase like 'I want to cancel' means something very different depending on whether the customer or the agent said it. Restricting a rule to 'Customer only' or 'User only' dramatically reduces false matches and makes results more actionable.

Use Location to catch openings and closings #

Many compliance and quality checks are specifically about what is said at the start or end of a call — a required greeting, a terms and conditions statement, or a closing offer. The 'Beginning' and 'End' location settings let you target just those portions of the call rather than searching the entire conversation.

Use Proximity for context-aware matching #

A single word in isolation can be ambiguous. The 'Proximity' setting lets you match a word only when it appears near another word or phrase — adding context that removes false positives. For example, matching 'cancel' only when it is in the same sentence as 'subscription' is far more precise than matching 'cancel' anywhere in the call.

Use Filter Conditions to target the right calls #

Not every keyword rule should apply to every call. Use filter conditions to limit which calls are searched — by DID, call tag, direction, or other criteria. This keeps results clean and avoids noise from irrelevant call types.

Keep phrase lists focused #

Each keyword rule supports multiple words and phrases, but resist the temptation to make a single rule too broad. A focused rule with a clear purpose is easier to act on than one with twenty loosely related phrases. If you find yourself adding many unrelated phrases to one rule, consider splitting them into separate rules with distinct names.

Example Configurations #

Compliance — required greeting check #

Use case: Ensure agents answer calls with the company name.

Flag any call where the agent does not say the company name or a standard greeting within the first 30 seconds.

Churn risk — cancellation language #

Use case: Identify calls where the customer mentions cancelling, for follow-up by a retention team.

Escalation detection — requests for a manager #

Use case: Flag calls where a customer asks to speak to someone more senior.

Competitor mention #

Use case: Track calls where a competitor is mentioned, to understand switching risk or competitive pressure.

Sales — closing language #

Use case: Check that agents are using a positive closing and offering next steps on sales calls, rather than just ending the call.

info
Changes to keyword rules only affect calls transcribed after the change is saved. Previously processed calls are not re-evaluated.