Search Options
The first parameter we need to set is the date range. This can be as small as a single day, and as wide as all dates with data.
✅ Your Turn |
|---|
To ensure we are searching through all the calls currently loaded in the system, select the “All dates with data” option.  |
2. Now that we have opened the date range to all calls in the system, we see 24,559 calls rather than the 1,519 calls we saw before (you may show a different number of calls here).
3. Now we need to refine our search by adjusting the sliders in this section.
Calls below 1 minute 30 seconds provide little to no substantial value in analysis, and calls with low call clarity have less transcription accuracy.
✅ Your Turn |
|---|
Adjust the” Call Duration” to exclude calls shorter than 1:30. Adjust the ”Agent Voice Clarity” and “Client Voice Clarity” to exclude calls with less than 70% clarity.  |
4. Now we can see the results have been refined down to 6,258 calls.
5. Next, we will take a look at the different search options in the “Term” section. Let’s find a call where the client is both frustrated and confused.
✅ Your Turn |
|---|
Let’s look for instances where a customer expressed that they were confused and frustrated. First, select the “Client” option from the drop-down menu.  |
Phrases/Terms
✅ Your Turn |
|---|
Enter “frustrated” and “confused”, they will appear in the “Search Terms” section. |
Further to the right, there is an option to select between AND / OR.
AND: Results must contain all entered search terms.
OR: Results must contain at least one of the entered terms.
This defaults to AND, so we can leave on that.
2. The Search Terms “frustrated” and “confused” are in fact located in the call as indicated by the preview of the transcript. However, the terms we used limited the search too much due to variations of these terms like “confusing” or “frustration” not being included.
You could include all the variations of the terms manually, or you could use a special character to be more inclusive.
3. Changing the terms to “frustr*” and “confus*” has expanded the results to 6 calls by making use of the wildcard special character which allows any variations of words that start with “frustr” and ”confus”.
Examples seen in the results:
Frustrated
Frustrating
Frustration
Confusing
Confusion
✅ Your Turn |
|---|
Next, let’s try a different approach to locating calls of interest. Select the ”X” here to clear all search terms.  |
4. Let’s look for calls where the agent fails to use polite words. We would use the negative or minus “ – “ character to exclude calls where those terms were used.
5. You could further refine the search by looking specifically at calls where the AI has determined Agent Emotion to have an overall negative score.
✅ Your Turn |
|---|
Try entering the phrases and filters as seen on the example; take note of how many results show in this section. Once you’ve completed this, “X” out the search terms and return the Emotion filter to “All”.  |
6. Because speech is complex, we may need to learn how agents speak to build phrase libraries.
7. To see how clients talk about “call back” or “calling back”, we could use the terms here with the OR option, but this is restrictive.
8. To discover other ways clients bring up this topic and to expand the search, we could use a tilde “ ~ “ to allow words in between “call” and “back”.
We also want to exclude the original phrases so we only see the different ways clients bring up the topic.
9. We want to continue adding negative terms to the ”Search Terms” list when they relate to call backs; this will allow us to continue filtering through different ways calls backs are brought up.
10. When we see terms that are not related to call backs, we do not add them to the list; these are considered “false positives.”
In this case, “call log” does not relate to call backs, so we will use remember this when we build an application in Module 5: Application Building/Management.
Special Characters
Other special character you may find useful are listed here.
It is important to keep in mind that only one special character can be used per phrase. The exceptions to this rule are the “ - “ minus character which excludes calls, the OR, and AND characters listed here.
Character | Function | Examples | Meaning |
|---|---|---|---|
Phrase | Cancel my service | Returns files with the phrase words that must be in order | |
| | OR | Cancel | opt out | Returns files for cancel OR opt out |
& | AND | Frustrated & upset My attorney & filed claim | Returns files when both phrases are detected |
~n | Search for phrases appearing up to n words apart | Cancel account ~4 My attorney ~3 Verify your ~3 Concerns questions ~4 Resolved issue ~4 | Returns files containing “cancel” and “account” within the same speaker turn no more than 4 words apart. E.g. “cancel my daughter’s account” will be detected |
* | Wildcard | Harass* | Will return files with other words for harass such as harassment, harassing, harasser, harassed, etc. |
~s< | Only in beginning of call (by seconds) | Defer payment ~s<30 First and last ~s<45 Verify phone ~s<30 Can I verify ~s<60 Your birthday ~s<42 | Will return files when ‘defer payment’ is said within the first 30 seconds. In this case if trying to identify calls whereas the customer says: ‘defer payment,’ you might want to use only the CLIENT (customer) channel |
~e< | Only at end of call (by seconds) | additional questions ~e<30 twenty four ~e<20 call back ~e<25 available twenty ~e<25 concerns ~e<30 file a claim ~e<20 better business bureau ~e<35
In ad hoc area, create two rules and make sure you are using the AND function on the right side Cancel ~s<30 AND Resolved ~e<30 (or a positive utterance) | Will return files when ‘additional questions’ is detected, in this case within 30 seconds from the end of the call. Often to better understand the “resolution” of a call can be found by focusing in on the last 30 seconds. If someone says “cancel” in the first 30 seconds the agent may have saved the customer and detecting cancel at the end of the call may be a better indicator or a true cancellation.
You could create an ad hoc search of two phrases and bookmark them to be reviewed continually – cancel at the beginning of the call and a ‘save’ statement at the end of the call to analyze agents who have saved a call (see on left). |
~s> | Except in beginning of call | Cancel ~s>30 (probably use CLIENT channel) | Cancel that is said outside of the beginning of the call, in this case after 30 seconds |
~e> | Except in the end of call | Cancel ~e>30 (probably use CLIENT channel) | Cancel that is said outside of the end of the call and in this case not in the last 30 seconds |
-~e< | Not detected at end of call | -concerns or questions ~e<30 (use AGENT channel here) | Will return files that an agent did not state “concerns or questions” at the end of the call and in this case last 30 seconds
|
-~s< | Not detected at the beginning of call | -verify phone ~s<45 -last four ~s<30 (AGENT channel or a specific AGENT ID) | The is a good way to detect when agents are not stating something they should be. Script adherence or Mini-Miranda usage |
~t | In same speaker turn | Cancel account ~t | Will return files that “cancel” and “account” appear in the same speaker turn |
^ | Search for phrases found at the beginning of the speaker turn | ^ helpful  | Will return files that contain “helpful” at the beginning of the speaker turn |
$ | Search for phrases found at the end of the speaker turn | Upset $  | Will return files containing “upset” at the end of the speaker turn |
For special characters related to timing within a call, it is important to note that some recordings begin when a customer is placed on hold rather than when an agent joins the call; this means the first few minutes of a call could be taken by hold time.
This could also be the case for the end of a call, if an agent is transferring a customer or if a call is dropped while a customer is being transferred.
These timing special characters are best used for locating calls of interest rather than coring purposes, because an agent has no control over the format of a call recording.
The beginning/end of speaker turn special characters can be useful to narrow down searches, but is important to remember that conversations can have some overtalk and/or interruptions that could disrupt the expected conversation flow.
Keep in mind that both ^ and $ would return results like this.
