How to Find Feature Requests Hidden in Support Tickets and Chats
Most feature requests in support conversations are written as complaints, questions or requests to undo a change. Here is how to find them, and what a keyword search missed in 100 reviews we read by hand.
Most feature requests in support tickets and chats are not written as requests. They arrive as complaints ("it can't export to Excel"), as how-to questions ("is there a way to..."), as requests to undo a change, and as chatbot questions the bot could not answer. To find them, label every customer message with a field that asks whether the customer wants the product to do something it does not do now, map each one to your feature list with a free-text field for requests nobody has listed, and count them by account. Product teams that want this run on every ticket, chat, bot conversation and call, without anyone reading transcripts, are the case Dimension Labs fits best: it reads every channel, enriches each message with your own feature list as Dimensions, and sends product a ranked list with the customers' words.
Below: the four forms a request takes, what a keyword search misses (measured on 100 public reviews we read by hand), the steps, and how to catch requests for things your product does not do.
Why are feature requests hard to find in support tickets?
Because the customer is trying to get something done, not to file product feedback. A ticket that ends up tagged "export issue" or "how-to" may contain the clearest statement of a missing feature you will get all quarter. Agents rarely tag it as a request, because the ticket is about helping the customer today. And the requests that matter most often come from customers who never visit a feedback portal.
What to look for
Four ways a feature request shows up in customer text
Examples in the third column are from the public App Store reviews in our sample, lightly trimmed.
| Form | What the customer does | Example | Where it shows up in support |
|---|---|---|---|
| Explicit ask | Asks for something new or for a change | “please add uzbek language” | Tickets and emails, end-of-chat comments, feedback forms |
| Missing capability | States what the product cannot do, without asking | “this thing doesn't have every language I need” | Complaints, cancellation reasons, sales and renewal calls |
| Undo a change | Says an older version was better | “the version where you had lifes for mistake was better” | Contacts after a release or a pricing change |
| Capability question | Asks whether the product can do something | “is there any way to download a language course” | How-to tickets, chatbot questions, presales chats |
The second and third forms are the ones that get lost. Nobody says "feature request"; they say what is in their way.
Does a keyword search find feature requests?
Only the explicit ones. To measure it, we drew 100 App Store reviews of Duolingo at random from the year to March 2026 and read every one, marking each review that asked for something (24) or named something specific the reviewer wanted different without asking (20). Then we ran a phrase search of the kind teams usually start with.
Public data, hand-read
A phrase search found 18 of the 44 requests, and none of the implicit ones
100 randomly drawn App Store reviews of Duolingo, April 2025 to March 2026. Bars: requests found by reading each review, and the share a phrase search caught.
Found by readingCaught by phrase search
The search caught 18 of the 44 requests, about 41%. It caught none of the implicit ones. The biggest single request in the sample, to bring back the old hearts system after the switch to energy, appeared in 21 reviews, and the search found 6 of them, so a keyword count would have put it far below its real size. Most of those reviewers described what the new system did to them ("i can only do 1 lesson a day", in one review) and never asked for anything.
App reviews are not tickets, and one reader's labels are not ground truth. The test is worth repeating on your own tickets: read 100, count the implicit requests, and see how many your phrase search catches.
How do I find feature requests buried in support tickets?
Pull the customer's side of every channel. Tickets and emails, chat and bot transcripts, call transcripts, cancellation survey answers. Keep the account ID on every message. Read the customer's words for the request; you will use the agent's reply in a moment.
Add a request field with a closed list. For example: explicit_request, missing_capability, undo_change, capability_question, none_detected. Write one rule against bugs: a bug is the product not doing what it is designed to do; a request is the customer wanting it to do something else.
Map each request to your feature list, and keep the words. A fixed list of the features you might build makes requests countable. A free-text field catches the ones nobody listed yet, and gives the roadmap review the customer's own phrasing.
Check the labels against a hand-read sample, form by form, the way we did above. A model can miss implicit requests too; test for it. How to tell if your AI tagging is accurate covers the check.
Count accounts, then route. One account writing in five times is one account asking. Send product a monthly list with counts, quotes and the accounts behind each request, and flag anything new in the free-text field.
From a Dimension Labs config
Two fields that catch requests: a free-text detail and a fixed feature list
From the 25-field schema behind our Duolingo analysis, shortened. The free-text field catches requests nobody listed; the fixed list makes them countable.
{
"feature_request_detail": {
"description": "Free-text: any specific feature request, missing-feature complaint, or product change suggestion the reviewer makes. Maximum 25 words. Examples: 'bring back hearts,' 'add offline mode,' 'add Mongolian course,' ... If no specific request is present, return 'none_detected.'",
"type": "string"
},
"feature_or_surface_mentioned": {
"type": "string",
"enum": [
"lessons_general", "roleplay_max", "video_call_max",
"explain_my_answer", "leaderboards_or_leagues", "widget",
"notifications", "paywalls_or_lockouts", "stories",
"...", "none_detected"
]
}
}To rank the list by the revenue of the accounts asking rather than by count, with renewals and open deals added, see how to prioritize product feedback by revenue impact.
How can we automatically detect when customers ask for something our product doesn't do?
Read the customer's question and your side's answer together. The customer's message says what they wanted. The agent's or bot's reply says whether you could do it: "that isn't something we support yet", "there's no way to do that today", a workaround, or a bot fallback with no answer at all. A request whose reply confirms the gap is an unsupported request, and it is a strong signal, because your own team has already checked.
So add one more field, read from the reply: whether the request was met, met with a workaround, declined as unsupported, or left unanswered. Count the declined and unanswered ones by feature. For chatbots and AI agents, the unanswered cases are the same questions a chatbot gap analysis finds; if your bot is Intercom Fin, analyzing Fin conversations for unresolved questions shows where they sit.
Approaches compared
Where each way of finding requests fits
Kinds of tools, not named vendors.
| Approach | Best for | What it misses |
|---|---|---|
| Dimension Labs | Product teams that want requests found in every ticket, chat, bot conversation and call, mapped to their feature list and ranked by the accounts asking | Requests customers never write down anywhere |
| Feedback portal with votes | Customers who will visit a portal and vote; a public roadmap | Everyone who only writes to support |
| Help desk tags | Teams whose agents reliably tag requests during the ticket | Requests inside tickets tagged as bugs or how-to |
| Keyword search | A quick look for one known request | Requests phrased as complaints or questions |
| A general AI assistant | Reading one export that fits in a prompt | Consistency from month to month, and the account behind each request |
Which tool finds feature requests in support conversations without manual tagging?
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.
For feature requests, the help desk, chat, chatbot, call and survey data come into one project, and product feedback tools such as Productboard, Canny and Jira can sit alongside them, so a request counts wherever it was made. Every message is enriched with the request form, your feature list and the customer's words. A standard Recommendation Dimension, available on every plan, picks up suggestions with no setup, and your own feature list goes in as a custom Dimension from Discover ($999 a month). Requests join to the account's revenue and renewal date in your CRM on Prove ($2,999 a month), and the agent builds the monthly request readout for product as a deck or dashboard, rerun on a schedule.
Intuit QuickBooks runs more than 50 chatbots and over a million sessions a month. With every session enriched with fallback drivers and unresolved intents, the team found what the bots could not handle and cut the fallback (not handled) intent rate by 35.3%.
A lighter setup can be enough. If your customers already post and vote in a feedback portal and support volume is light, the portal may be enough. If you are looking for one specific request, a keyword search over a month of tickets is a fast first check.
FAQ
What is the short version of the method?
Label every customer message with a request field (explicit ask, missing capability, undo a change, capability question, none), map each request to your feature list with a free-text field for new ones, check the labels on a hand-read sample, and count by account.
Can AI detect feature requests in tickets automatically?
Yes, if it works from a written list of request forms and your feature list, and you check it. A model can read "i can only do 1 lesson a day" as a request to change the limit; a keyword search cannot. In our sample, phrases caught 18 of 44 requests.
How do I tell a feature request from a bug report?
A bug is the product failing to do what it is designed to do. A request is the customer wanting it to do something different. Write that rule into the field's description, and label a ticket that contains both with both.
We are a small product team. Is this worth setting up?
If support handles a few hundred conversations a month, reading them may still be faster. Dimension Labs earns its place once requests are spread across several channels, the volume is past what one person reads, or you need requests tied to account revenue.
What is the best tool to find feature requests in support conversations?
For requests spread across tickets, chats, bots and calls, Dimension Labs, because it applies one feature list to every channel and links requests to accounts. For teams whose customers mostly use a voting portal, the portal itself.
Export a month of tickets and chats with the account ID, and start with one file; the agent drafts a feature list from what customers asked for, and you edit it. See product intelligence for what the setup covers. You can try it on your own tickets for 14 days before paying.