
TL;DR
Intent signals are evidence of what a person or account is likely to do next. Most guides cover third-party data that shows an account researching a topic.
The strongest first-party intent signal is what customers write and say to you on WhatsApp, chat, email and voice, because it is stated, specific and tied to one person.
Every inbound message is a buying signal, a support signal, or both. The mixed messages carry the most revenue and get misread most often.
Classify each message in real time against the customer's history, answer both parts in the same thread, and keep sales and support on one record.
Daniel Reyes has an auto loan with his bank. One Tuesday he sends a WhatsApp message saying his car payment was taken twice this month, and asking whether the bank refinances a second vehicle.
The service team opens a ticket for the double debit, refunds it in two days and closes the ticket. The refinancing question stays inside the closed ticket. Three weeks later Daniel finances his wife's car through the dealership.
That second sentence was one of the clearest intent signals the bank received all month, and its systems filed it as support.
Daniel's message held two signals at once. The double debit was a support signal about a product he already had. The refinancing question was a buying signal, naming a product, a vehicle and a need, from a customer the bank already knew.
Teams that read both halves of a message like that, while the thread is still open, turn service traffic into pipeline.
What are intent signals?
Ask a marketing team for its intent signals and you usually get a dashboard of accounts researching a topic. The term is broader than that. Anything a person or account does or says that hints at their next move qualifies, from a pricing-page visit to a new job posting to a question about loan eligibility.
Most of what ranks for the term comes from account-based marketing vendors. Their guides sort intent signals into a familiar set of types.
Third-party research signals. An account's employees read content about a topic across a network of publisher sites.
Engagement signals. Visits, downloads and email clicks on your own properties.
Firmographic and technographic signals. Hiring, funding, new tools in the stack.
Conversation signals. What a person writes or says to your team, usually listed in one line and left there.
The last type deserves far more than a line. For teams that sell to consumers or small businesses at volume, in banking, insurance, healthcare or car retail, it is often the richest signal they own.
Third-party intent data reads accounts, and conversations read people
Third-party intent data tells you that someone at an account is reading about a subject. It is anonymous, inferred and often some days old by the time it reaches a dashboard.
A customer message comes from a named person with a known history, at a known moment, in their own words. The two sources answer different questions, so the useful move is to know which one you are holding.
| Dimension | Third-party intent data | Intent signals in conversations |
|---|---|---|
| Who it describes | An account or company | One named person |
| How intent is known | Inferred from reading patterns | Stated by the customer |
| Timing | Often batched | The moment it is sent |
| What it tells you | A topic is being researched | What the person wants, by when, and why |
| Covers existing customers | Rarely | Yes, including expansion and churn cues |
Intent data still has a place in outbound targeting. Once a person starts talking to you, though, their own words outrank any score built from page views.
Why the customer's own words are the strongest first-party intent signal
Buyers now do most of their deciding alone. In a Gartner survey of 632 B2B buyers, run in 2024 and published in 2025, 61% said they prefer a buying experience without a sales rep. The intent-data vendor 6sense reported in its 2024 buyer study that buyers are nearly 70% through their purchase process before they engage a seller.
So the first message a customer sends carries a lot of decided intent. It also arrives on channels where people already spend their day. Meta said in 2023 that users and businesses interact more than 600 million times a day across its platforms.
A message like "can I add my daughter to my health plan before her surgery in March" names the person, the product, the timeline and the reason. No research-surge score gets that close to a decision.
Every inbound message is a buying signal, a support signal, or both
Sort any inbound message by two questions. Does it point toward a new purchase or a bigger one? Does it describe a problem with something the customer already has?
Buying signal. A prospect asks whether a clinic takes their insurance for a knee consultation next week.
Support signal. A policyholder asks where to download last year's premium certificate.
Both. A car owner reports a failed service booking and asks what a trade-in on the same model would look like.
Neither. A wrong number, spam or a thank-you.
The mixed messages carry the most revenue, and most systems can route a message to only one queue.
Ticketing tools file the "both" message under support because it contains a problem. Lead tools ignore it because it comes from an existing customer. That gap is where Daniel's second car went.
What do buying signals look like inside a conversation?
Buying signals in a conversation are statements that move a person closer to a decision. They show up in recognisable shapes across BFSI, healthcare and automotive.
A timeline. "We need the loan approved before the 15th."
A fit or eligibility question. "Does this cover pre-existing conditions?"
A comparison. "The other dealer offered a longer warranty on this model."
A scope change. "Could the policy cover my business vehicle too?"
A request for the next step. "What documents do you need from me?"
A trigger event. "We just had a baby" or "I'm moving cities in June."
Context decides how strong each one is. A price question from a new prospect is a buying signal. The same question from a customer holding a wrong invoice is a billing problem.
A buying signal is only as reliable as the context you read it in.
Support signals carry expansion and churn cues
Support signals describe friction with something the customer already owns. Most teams measure them as volume to reduce, which hides two kinds of revenue inside them.
Expansion cues
A customer who hits a limit is describing a bigger need. A patient asking whether the clinic also offers physiotherapy after surgery, or a business owner asking for more cards on a corporate account, is telling you what to offer next.
Gartner research from 2020 on service that enhances value found that among customers who rated an interaction highly on this measure, 82% stayed, 85% bought more and 97% recommended the supplier. The service moment is where those outcomes start.
Churn cues
Churn shows up in words before it shows up in the CRM. Watch for phrases like "close my account", "transfer my policy", "this is the third time" and any mention of a competitor during a complaint.
A repeated complaint across channels counts more than any single message. A customer who emails, then calls, then messages on WhatsApp about the same claim is telling you how close they are to leaving.
What does misreading an intent signal cost?
Misreading an intent signal costs revenue in two directions. Each one has a recognisable pattern.
A buying signal treated as a ticket
The question waits in a support queue built for resolution times, while the buyer's interest fades by the hour. A 2011 study published in Harvard Business Review found firms that tried to contact a lead within an hour were nearly seven times as likely to qualify it as firms that waited even one hour longer.
That study looked at web leads. A buying question sent in a chat is at least as time-sensitive, and it often gets slower treatment because it arrived on a service channel.
A support signal treated as a lead
The reverse error sends an upsell to someone whose claim is still unpaid. In the same Gartner survey of 632 buyers, 73% said they actively avoid suppliers who send irrelevant outreach, and nothing feels more irrelevant than a pitch in the middle of a complaint.
Customers also expect each team to know the history. In Zendesk's vendor-run 2023 CX Trends research, 70% of consumers said they expect anyone they interact with to have full context.
Both errors come from the same root, which is reading a message without the customer's history.
How to classify intent signals in real time
Classifying intent signals well takes five steps, and each one runs while the conversation is still open.
Read the whole message. Mixed messages often put the problem first and the buying question last.
Check who is asking. A prospect, an active customer and a customer with an open complaint need different readings of the same words.
Read against history. Earlier messages, open cases and recent purchases change what a question means.
Label buying, support or both. Keep both labels when both apply, and note urgency and any stated timeline.
Act in the same thread. Resolve the problem, answer the buying question, and pass the opportunity on with the full conversation attached.
Review a sample of labelled conversations each month. Look for buying questions that ended as closed tickets and churn phrases that never reached the retention team.
For the mechanics of how models extract intent and sentiment from text, see our guide to intent and sentiment extraction in conversational AI.
Why buying and support signals belong on one customer record
Sales tools and service tools each hold half of the conversation. The service desk sees Daniel's double debit, the CRM sees his loan, and neither sees the sentence that connects them.
One customer record that both teams read fixes the routing problem at its source. The timeline holds every message across channels, the labels attached to each one, and what happened next.
It also changes reporting. You can count how many buying questions arrived through service channels, how fast each was answered, and how many support threads carried an expansion or churn cue that someone acted on.
How Zigment reads intent signals in every conversation
Zigment is a Conversational Revenue Orchestration Platform for GTM teams. Its AI agents read every inbound message on WhatsApp, voice, email and webchat and decide whether it signals buying, support or both.
The Conversation Graph keeps one running timeline of what each customer said and did, so every agent reads the history before it replies. An agent can resolve Daniel's double debit, answer the refinancing question in the same thread, start the application and write both outcomes back to the CRM.
Escalation rules are configured once, for cases such as a regulated complaint, and the agent hands over with the full thread attached. Everything else runs on its own, inside the boundaries the team sets.
The next message already says what the customer wants
Pull last month's closed support tickets and search them for a timeline, a comparison or a question about another product. Count how many of those reached anyone who could sell.
That count is the gap between the intent signals you receive and the ones you act on.
See how the Conversation Graph keeps every signal on one record across the channels your customers already use.