A caller asks whether the hostel fee includes the mess. The agent says it will find out. The call ends well enough, and nobody thinks about it again. The next day two more callers ask the same thing, on two different calls, and no single one of those calls looks like a problem.
This is the kind of gap that survives for months. A supervisor reading transcripts would have to read the right three out of thousands. Conversation Intelligence adds up the calls it has read and gives you a list: the topics, the objections, how callers felt, and the questions the agent or person on the call could not answer.
What makes the list useful
Each call that is read becomes a small record: topics, intent, sentiment, outcome, the caller's questions, objections, and anything nobody could answer. The list of unanswered questions is called knowledge gaps, and it is the part most people act on first, because every entry has an obvious next step.
You have two choices on each one:
- Answer it, by writing the answer into a knowledge base you pick. That teaches only the agents using that knowledge base, once the answer has loaded. If it fails to load, the gap comes back.
- Dismiss it, which hides it for everyone until you press Undo.
A gap stays on the list until someone does one of those, and it returns if a later call raises it again. Gaps for AI agents and for your team leave their lists separately.
A half-hour routine
Pick one person who owns the facts, usually whoever maintains your knowledge base, and give them a fixed half hour each week.
- Sort the gaps. Answer the ones you can answer in a sentence.
- Send the ones that need a decision to whoever owns that decision. “Do we refund the registration fee?” is a policy question, and an agent should not be guessing at it.
- Dismiss the ones that are out of scope, such as a caller asking about something you do not sell.
- Open one call behind any gap that looks strange. A question can be a misheard word.
Then look at the objections. If the top objection on your floor is one your script never mentions, that goes to whoever owns the script. A count of what callers actually said is worth more than the theory your team has been repeating.
Choosing how many calls to read
Reading a call is charged once per call, and only calls with a transcript can be read. So you choose how much to read in Call Settings:
- On demand is where a new workspace starts. Nothing is read until someone presses Get insights on a call or selects calls in Call Logs, and the cost is shown first.
- A sample each night reads up to 20 of the previous day's calls, at random.
- Automatic reads every answered call of 15 seconds or more where the caller spoke, and charges for each.
- Off stops it.
The nightly sample suits most teams starting out, because a random sample is fairer than the calls someone happened to pick. Move to Automatic when the list is being worked and you want the fuller picture. The current rate is on the pricing page.
How it can mislead you
Every figure is out of the calls that were read, and a line at the top says how many that is. Twelve calls hand-picked under On demand still fill the bars. Read that line first.
Dates are when the call happened, and not when it was read. Test calls can be read if you ask, and are charged, but never count here. A question a person could not answer names them only if they were alone on the call, so this is a poor tool for blaming someone. It works much better for finding out what your whole floor lacks.
It reads the transcript in the language the call was held in. If you need to read a call yourself in another language, use call translation. And answering a gap does not fix a script: if the agent says something wrong because its instructions told it to, the fix belongs in the script, followed by a run of your agent tests.
What it will not do
It won't decide what your answer should be, and it won't change anything by itself. Nothing reaches an agent until a person writes the answer and picks where it goes. That is the right way round, since a wrong fee quoted with confidence is worse than “I will find out”. Conversation Intelligence is not on the free trial and depends on your plan, which pricing shows.