How to sort Instagram comments and DMs by buying intent
Intent routing is the practice of labeling what each Instagram comment or DM actually asks for: pricing, purchase, booking, availability, trust, or an objection. It is also the step that makes Sales QA possible, because you cannot tell whether a conversation was handled well until you know what the person wanted from it.
A team that answers everything politely but never separates a pricing question from a booking question still loses deals, and nobody can say why. Intent is what turns a pile of messages into judgments you can check.
Why does intent come before the quality verdict?
Because “handled well” is defined by intent. A booking question that gets a friendly answer and no slot was handled badly. The same reply can be right for one intent and wrong for another.
So classification runs first. Reploom gives every conversation a 0-100 lead score, a buying-intent type, and a one-line reason, then asks: given what this person wanted, did the team do what that intent required?
This review happens after the conversation, not instead of it. Reploom does not write, draft, suggest, or send replies. It reviews Instagram, Facebook, and WhatsApp threads and reports the patterns it finds.
Which buying intents are worth separating?
Six labels cover almost every inbound social message that could turn into revenue: pricing, purchase, booking, availability, trust, and objection. Each owes the customer something different.
| Signal the customer sends | Intent type | What a good reply owes them |
|---|---|---|
| “Can you send the total price and what the plan includes?” | Pricing | A number, what it covers, and a follow-up within 24 to 48 hours if they go quiet. |
| “I want the twelve-session plan, how do I pay?” | Purchase | Payment details the same day, and a check that they arrived. |
| “Do you have anything near 8 AM this week?” | Booking | A specific slot, held, plus the one step that confirms it. |
| “Are you taking new clients in the morning block?” | Availability | A direct yes or no, and the nearest alternative if the answer is no. |
| “Do you work with people who have never trained before?” | Trust | One comparable case or proof point, then an invitation to start. |
| “That is more than I expected to spend.” | Objection | An answer to the actual concern, not a repeat of the price, and a smaller next step. |
That right-hand column is your standard. Sales QA checks, week after week, whether it was met.
What separates a lead from a non-lead?
A lead is someone whose message implies a purchase decision they have not made yet. A non-lead is everyone else: a compliment, a job application, a supplier pitch, a spam comment under a reel.
Enthusiasm is not the test. “This looks amazing” is a non-lead. “Is Tuesday possible” is a lead, because a decision sits behind it that your answer either advances or stalls. Buying signals rarely arrive dressed as inquiries, which is the argument in the leads hiding in your comments.
Scale is why this needs a system. The example report on the Reploom homepage shows a sample week of 420 conversations checked and 96 leads found, 31 high-intent. Rereading all 420 to find the 96 is where teams quietly give up.
Why non-leads must stay visible instead of being filtered away
Because a filtered category cannot be audited. If a scoring system deletes what it decides is noise, you cannot find the leads it misjudged, and those are the ones you need to see. Reploom keeps non-leads visible: everything is scored, nothing is dropped.
The denominator matters: “nine leads still waiting” means one thing against 420 conversations and something else against 40.
How a 0-100 score plus a one-line reason makes a verdict auditable
A number on its own is not auditable. A number with a stated reason is, because a human can read the reason and disagree with it in one sentence.
“Lead 92” tells a sales manager nothing they can act on. “Lead 92, pricing intent, asked for the total price and payment terms, then named a scheduling condition” can be confirmed or corrected at a glance. Every score should ship with the sentence that produced it, and the conversation underneath as evidence.
A label nobody can argue with is a label nobody can improve.
Two pricing conversations, two different failures
Both come from the example conversations on the homepage, at a fictional fitness studio. Same intent, two different leaks.
Noah Bennett, Instagram, lead score 92, pricing intent
He asked for the total price and what the plan includes, and got a clear answer: $720 for eight weeks, two coached sessions a week, payable in two installments of $360. He wrote back: “The split option would work. I just need to make sure Tuesday and Thursday mornings are possible.”
The thread ends there, twenty-six hours with no follow-up, one scheduling confirmation away from a sale. The failure is no follow-up after pricing.
Layla Perez, WhatsApp, lead score 81, pricing intent
She asked what private coaching costs for one session a week, was told $380 for a four-session monthly package, and asked what to do next. The reply: the package can begin at any point in the month. Her last message: “I am still interested, but I am not sure how to reserve the consultation.”
Also pricing intent, also a correct price. The different failure: answered, but no next step. No slot, no call, no question that moved her forward.
Without intent labels, both look like ordinary DMs where the customer went quiet. With them, they are two instances of one leak, covered in why pricing answers stall social leads. One stalled pricing thread is an incident. A repeated one is a process problem with a fix.
How to sort your own comments and DMs by intent
Five rules do the work, and they stay readable on purpose: anyone should be able to disagree with a route, change it, and see why a message landed where it did.
- Read the last customer message first. Intent settles at the end of a thread, not the start. Noah opened about price and closed about Tuesday mornings.
- Ask what the person cannot do until you answer. That is the intent, stated as a blocked action rather than a topic.
- Score the decision, not the tone. Warm language with no decision behind it is a non-lead. Flat language with a decision behind it is a lead.
- Write the reason in one line, in the customer’s words. If you must paraphrase heavily to justify the score, the score is probably wrong.
- Group by intent weekly. Per-message review finds mistakes. Weekly grouping finds patterns, and patterns are what you fix.
A few dozen conversations a week is a spreadsheet habit. Several hundred a month across three channels is not, which is where an automated Sales QA review earns its place.
How intent looks different on Instagram, Facebook, and WhatsApp
- Instagram: buying questions arrive in public comments and DM replies to stories, and they look casual. A comment asking whether something comes in another size is availability intent.
- Facebook: page comments and mentions are public, which raises the cost of a missed intent. An unanswered pricing question sits where the next buyer will read it.
- WhatsApp: monitoring starts at connection. Whether earlier threads can be reviewed depends on what message history is available, so connect the account before the period you want to review.
What should you measure once you route by intent?
Four numbers: the share of conversations that are leads, reply time by intent, the share of high-intent threads that got a concrete next step, and how often a human overrides a label.
The override rate is the most useful. If pricing questions keep getting scored as trust questions, the definition needs tightening. If the labels surface buyers the team used to scroll past, it is working.
From DM volume to a list of lost deals
Sorting by intent turns “we get a lot of DMs” into the headline on the example report on the homepage: “7 sales opportunities stalled this week,” with the threads attached. The first produces a meeting. The second produces a fix.
Reploom groups repeated leaks into a weekly Sales QA report, emailed Monday morning, and turns the biggest one into a 7-day tracked playbook. To see that on your own threads, book a demo.