Guidelines for building trackers
  • 10 minute read
  • Contributors

Guidelines for building trackers

Article summary

Keyword trackers are based on specific words, terms or phrases that are mentioned in your conversations. You can add as many words as you like to a single keyword tracker, and it will surface all of the calls in which these words appear. For example, you can set a keyword tracker to track words like: discount, save, best price, and more and when any of these words is used in a call, you’ll track it.

Elements that make up a tracker

The tracker setup screen includes fields that specifically help tailor trackers to your business need. Check out the list below to refresh yourself on each element that you'll use to build your tracker:


  1. TRACKER NAME: Serves as a title for the phrases it contains. Note that the title itself is not tracked unless added as a phrase (if needed).

  2. LANGUAGE: The default language is English. Trackers can be set to any of the supported languages by clicking Add another language below #4 and adding phrases per language. Note that phrases may repeat across languages.

  3. Enter phrases (separated by commas): This is where you enter the actual phrases you want to track. Separate tracker words and phrases with a comma.

  4. Include related word forms: Select this checkbox to include related word forms, for example, stemming and inflections. For example, for renew, Gong will also surface renewal and renewing.

  5. Track when said by: Select whether the phrases are to be tracked when said by someone at your company, someone not at your company (for example, customer) or anyone.

    For example, you probably want your "Pricing" tracker to capture any pricing mentions and so should set it to track when said by anyone; whereas a "Competitor" tracker you’ll probably only want to track when competitors are brought up by the customer (not your own company).

  6. Additional filters: Use additional filters to choose the calls to which the tracker applies. For example, a tracker can apply to calls by a specific team member or team, specific stages, etc. The number of calls that the tracker applies to appears while choosing the filter.

  7. Show this tracker in: Trackers can appear in multiple places: on call pages, in emails, in the Gong API, and in the CRM export. You control where trackers are displayed.

  8. Advanced (not shown in screenshot above) Click to set more filters: when the tracker is said as part of a question only, or when in a call related to which topic the tracker is mentioned.

Checklist for building trackers

When you create a tracker, it's useful to review this list to help refine your thinking:

  • Evaluate the tracker purpose - What's the business question that the tracker should address? This could be anything from "Are our competitors mentioned on late-stage deals" to "Are our reps using the messaging we recently trained on?".

  • Decide who you want to track - Based on the tracker purpose, who should the tracker apply to? This might be only your customers, your company employees, or anyone. You can also set the tracker to track according to the relevant call list (filter).

  • Take precision and recall into consideration - Precision can be seen as a measure of quality, and recall as a measure of quantity. Higher precision means that the tracker returns more relevant results than irrelevant ones, and high recall means that the tracker returns most of the relevant results (regardless of whether or not irrelevant ones are also returned).

  • Perform a reality check - Go to Conversations > Search and test your proposed tracker phrases by filtering the calls:

    1. Filter by any and all the relevant words and phrases that come to mind:

      1. In the Words or phrases search box, enter one of your phrases.

      2. In the Advanced search area, set the filter to match the tracker.

        For example, "said by" in your search should match "track when said by" in the tracker. If you checked "Include related word forms" in the tracker, select it when you perform the filter calls test.

    2. In the filter results, see how many times the phrase appears. We recommend that the phrase should appear in at least 5 calls to be worth including as part of your tracker.

    3. Check the context; read the call snippets and make sure they are relevant to your tracker.

    For example, if you perform a test and see that of the first 20 results, 17 snippets are accurate, then you could argue that your tracker will be ~85% accurate (or the precision will be ~85%).

  • Refine your phrase list - perform a health check on your tracker:

    • Think of words or phrases that can help focus the tracker, and add them.

    • Think of examples where a phrase isn't relevant to the tracker. Does this make sense, and how often would you expect to find such cases in calls?

    • Avoid using stopwords (common words that can be excluded from searches because they increase the work required to parse them while providing minimal benefit).

      For example: the, in, at, that, which, our, my,  his.

    • Repeat the previous step (reality check) with any new phrases.

    • Read the list of phrases, and see if you'd be able to deduce the tracker purpose based on the list.

  • Tracker review - once your phrase list is complete and you've created the tracker, perform another reality check (step 4). Review the call snippets of 3-5 call pages, noting:

    • Tracker volume - Does the % of calls that contain any phrase in the tracker what you would expect?

    • Relevance (context) - Are most snippets relevant to the tracker purpose and do they match the business question?

Common tracker traps

Here are some common issues that can make your trackers less effective, and suggestions on how to avoid them.

Using general phrases

Using phrases that are too general can lead to low precision. The tracker picks up the word completely out of context.

For example, the word quality in a "QA Process" tracker. The word quality is an integral part of the QA process. However, there are many other contexts it can appear in such as recording notification (“Your call is being recorded for quality and training…”), or as an attribute (“an improvement in product quality"), to name a few.

Therefore the phrase should be quality assurance and not quality by itself.

Using phrases that match other contexts

When you list the phrases of a tracker, consider whether they are likely to appear in a variety of contexts, and circumscribe them to the purpose of the specific tracker.

For example, having the phrase want to let you know in a "Recording notification" tracker - while very common in recording alerts, can also appear in many other contexts.

Using phrases that are too elaborate and long

Don’t use elaborate and long phrases; mostly likely they won't be repeated word for word.

For example, the following are some of the longest phrases we've seen in various trackers, all yielding zero results:

  • Objections - never what we intended

  • Remote work - working from home is definitely a challenge

  • Customer Feedback - maybe I should work for you guys, what pleasure has it been to work with you

  • Pain Funnel - what else did you try to resolve this issue

Not using the common denominator

When a word is repeated in several phrases, search for it in theSearch page to see if you’re missing some phrases, or whether you should eliminate redundancies and use the word as a stand-alone phrase.

For example, when your "Pricing" tracker has ballpark in two phrases - ballpark figure and ballpark pricing. Filtering calls by ballpark on its own, we found a couple of missed common phrases we'd missed, such as ballpark number and ballpark price. Just using the phrase ballpark would pick up all these phrases.

Using redundant phrases

Many stem phrases are included in other phrases and are therefore unnecessary.

For example, if we set the phrase our budget, we don't need to explicitly include out of our budget as well. That said, including the longer phrase does mean the entire phrase gets highlighted in snippets, which might be useful from an identification perspective.

Trackers vs the custom vocabulary

Alternative pronunciations should be added to the custom vocabulary and not to trackers.

For example, a tracker would include only the phrase words, whereas the custom vocabulary list would include pronunciation alternatives such as, SFDC, Ess Eff Dee Cee, etc.

Using overlapping phrases

Some trackers overlap and contain the same phrases. This is OK in a few phrases here and there, but if 2 of your trackers have over 50% of overlapping phrases (use the same phrases), the trackers could likely benefit from better definition and more specific phrases.

Creating trackers: examples

Here are some examples of the tracker creation process, using the checklist above.

Example 1: Legal tracker

Here's how we set up a legal tracker.

  1. Tracker purpose

    When we looked at the trackers in Gong, we saw that we didn't have a tracker that explicitly tracks words and phrases pertaining to the legal aspects of a deal.

    Tracker purpose: To track anything that has to do with any legal aspects of a deal

  2. Track when said by

    Any legal mention is relevant, whether said by the customer or the company, therefore the tracker should be set up to track anyone.

    Track when said by: Anyone

  3. Precision and recall considerations

    The nature of this tracker is such that we don't want to miss any legal mention, even at the expense of having a few snippets incorrectly marked as legal mentions.

    We thought about what words and phrases are relevant to the legal tracker, and came up with the following: legal team, legal review, legal departmentcontracts, general counsel, NDA, attorney.

    We can see that the word legal appears in several phrases. Since we want to capture any legal mention, we can remove the redundant phrases and use legal only.

  4. Reality check

    We verified our assumption regarding legal as part of the reality check: filtering for the word legal yielded many results (over 10% of all calls). Reviewing the snippets in several pages confirmed that legal can be a standalone phrase because most snippets were relevant to the tracker.

    We checked our other phrase candidates. When filtering for the phrase agreement and reviewing the matches in the calls page, it was clear that the word agreement alone can be used in many different contexts unrelated to legal, and therefore should be removed from the phrase list.

  5. Refinement

    Reading the filtered snippets during our reality checks brought to light some phrases we had missed initially, such as DPA and MSA.  They also served as a prompt for phrases we hadn't thought of , for example, signDocuSign, mutual NDAMNDA.

    By this stage, we had refined list of phrases to the following: legal, NDA, MNDA, DPA, non-disclosure, general counsel, MSA, addendum, amendment, red-lines, docusign

  6. Review

    Filtering the Searchpage by the tracker yielded a good volume of results, and a review of the snippets confirmed the tracker contains relevant phrases.


Example 2: Privacy concerns

Here's how we set up a tracker to highlight customer privacy concerns.

  1. Tracker purpose

    The tracker should surface any data, privacy, and legal questions, any issues and concerns arising from Gong recording, ingestion, and handling of sensitive customer data.

    Tracker purpose: To track anything that has to do with concerns over privacy and personal information

  2. Track when said by

    We are interested in concerns raised by customers about privacy and sensitive data, so the tracker should be set to capture customers only.

    Track when said by: People not at Gong

  3. Precision and recall considerations

    We set out to create a tracker that tracks concerns regarding privacy and data retention. Therefore, it is more important that the tracker identify calls correctly, even if it misses some relevant mentions, rather than identifying all calls with privacy issues and increasing false positives (identifying a call as having privacy concerns when it does not). Or simply put, we favor precision over recall.

    We came up with a list words and phrases are relevant to this tracker: privacy, data privacy, privacy concerns, privacy issues, legal issues, consent, sensitive data, sensitive information.

    The word privacy appears in several phrases, but since we want to capture privacy concerns rather than just any mention of privacy, we will review the calls page as part of the reality check to assess whether privacy alone can be included in the phrase list. 

  4. Reality check

    Filtering the Search page for the word privacy, and setting the advanced search to said by anyone not at Gong, (as we did in the tracker) yielded many results.

    Reviewing the snippets on several pages of results indicated that the word privacy on its own is too general, as it was mentioned in various contexts, such as account set-up calls, or as part of the sales process  (privacy team review), whereas we want to capture privacy issues and concerns.

    Filtering for the phrase consent yielded many results that had to do with the Consent page, and not necessarily with privacy concerns. Therefore we decided we won't include it in the tracker.

  5. Refinement

    Reading the snippets in the filtered Search page helps focus the eventual phrases we'd use in the tracker: privacy concerns, privacy issues, privacy laws, legal ramifications, legal implications, recording customers, without consent, sensitive information, confidential information, sensitive data.

  6. Review

    Filtering the Search page by the tracker yielded a good volume of results, and a review of the snippets confirmed the tracker contains relevant phrases.

Was this article helpful?


Eddy AI, a genAI helper, will scrub our help center to give you an answer that summarizes our content. Ask a question in plain language and let me do the rest.