There is a specific kind of dread that comes with opening four different apps at 9pm to check whether anyone replied to you. Instagram in one hand, LinkedIn on the laptop, a Facebook page that nobody has logged into since the last campaign, TikTok comments you only see because a customer reposted their own complaint. Every one of those surfaces holds a conversation that someone was waiting on, and none of them are connected to each other, so nobody on your team knows what has already been answered.

This is how brands go quiet without deciding to go quiet. Nobody switches off engagement on purpose. They just get busy, the replies slow from thirty minutes to four hours to two days, and one afternoon a customer posts a screenshot of your unanswered question from three weeks ago. At that point the problem is no longer etiquette. It is operational.

A unified social inbox is the fix, but the word "unified" is doing a lot of work. Pulling seven streams into one screen without changing anything else just gives you seven streams in one screen. The value comes from the decisions you layer on top: what counts as urgent, who owns it, how fast it must move, what a good reply looks like, and what happens when nobody is awake. This post walks through all of it — an audit of where messages actually arrive, a tag and folder design, triage rules, response templates, escalation, coverage across time zones, metrics, automation boundaries, and the practical setup.

An inbox is not a feature you switch on. It is a small operational system with owners, targets and a closing time.

What scattered replies actually cost you

Before building anything, it is worth being honest about the bill you are already paying. Four costs show up in almost every team we have audited.

• Lost customers at the decision moment. Questions asked in comments and DMs are purchase-intent questions. "Does this ship to Karachi?" asked under a product post is a sale sitting in a queue nobody opened. Industry studies consistently find that buyers who get an answer on social are far more likely to buy than those who do not, and a reply after forty-eight hours might as well be a reply after forty-eight days.
• Duplicate and contradictory answers. Two people answer the same complaint on two platforms with two different promises. The customer now has proof you do not know your own policy, and you have a screenshot problem.
• Silent burnout for whoever is checking everything. The person doing the 9pm sweep never finishes, never feels caught up, and eventually stops looking at the platforms where the loudest conversations happen. Volume without a system always lands on the most conscientious person on the team, which is the least fair place for it to land.
• No record when someone leaves. If reply history lives in personal browser tabs, a departing community manager takes your customer context with them. Conversations restart from zero and customers notice.

None of these are motivation problems. Every one of them is a queue-without-an-owner problem, which is good news, because queues are fixable.

Audit: where every message actually arrives

You cannot unify streams you have not listed. Spend one week doing this properly; it is the single highest-return hour in this whole process.

Run a seven-day capture

For seven consecutive days, log every inbound message anywhere. Not just comments — DMs, mention replies, story replies, tagged posts, review replies, YouTube comments, community group posts, and the odd places people message you that you forgot you set up. One row per channel is enough:

• Platform and surface (feed comment, DM, story reply, mention, review, group)
• Average daily volume in that week
• Who currently answers it, informally
• Average time to answer right now
• Whether it contains purchase, support or reputation risk

You will find two things. First, at least one channel is receiving real volume that nobody owns — usually Facebook page messages or TikTok comments. Second, the channel with the highest-risk messages is almost never the channel that gets checked first.

Count the actual load, not the feeling

Add up the daily totals. Most small and mid-size teams are shocked by the real number: a business posting five times a week across four networks will typically see 40 to 120 inbound messages a day, of which perhaps 10 to 30 need a written human reply. The rest are reactions, spam, emoji, and tags of friends.

That distinction matters. Of everything arriving, only a fraction requires a person with judgement. The rest requires correct routing. Design for both.

Map what people ask about

Group a sample of 100 messages into categories. A common split for an e-commerce or service brand:

• Pre-purchase questions: shipping, price, sizing, availability, compatibility — usually 35 to 50 percent
• Order and support issues: where is it, it arrived damaged, how do I get a refund — 20 to 30 percent
• Praise, tags and UGC — 15 to 25 percent
• Complaints and public criticism — 3 to 10 percent
• Spam, bots and irrelevant — 5 to 15 percent

The pre-purchase group is revenue. The complaint group is reputation. Both need a response time measured in hours. The praise group needs a lightweight habit — react, thank, maybe repost. The spam group needs a filter and a delete key.

Audit checklist

• Every inbound surface listed, including the ones you forgot about
• Seven days of volume logged rather than estimated
• Current owner named for each surface, even if the answer is "nobody"
• 100 messages categorised into five buckets
• Purchase-intent and complaint messages identified separately
• Current median time-to-reply measured on at least three channels

Design the queue: tags, folders and response targets

Now you are designing the system. Three layers: a tag vocabulary, a folder structure, and a response target per folder. Keep all three small. Most inbox implementations fail because somebody builds forty-two tags in week one and stops using them by week three.

A tag vocabulary that stays usable

Tag by intent, not by content. Content tags multiply forever; intent tags stay stable. A working starter set:

INTENT
  presale      — product, price, shipping, sizing, availability
  order        — status, delivery, damage, returns, refund
  praise       — thanks, fan mail, UGC, repost requests
  complaint    — public or private dissatisfaction
  lead         — partnership, bulk, press, collaboration
  abuse        — spam, scam, harassment, impersonation

PRIORITY
  p1           — public complaint, safety, legal, outage (answer < 1 hour)
  p2           — purchase question from an active thread (< 4 hours)
  p3           — everything else (< 24 hours)

OWNER
  [name]       — one assignee per open thread, never two

Nine tags. Anything more and you are cataloguing rather than triaging. If a genuinely new category appears more than three times in a month, add it then — never in advance.

Folders follow coverage, not platform

Do not build a folder per platform unless your team is split that way. Build folders around who can answer and when.

• All open — one screen, everything unowned
• Mine — your assigned threads
• Presale — answered by sales-aware staff, extended hours
• Support — needs order data or policy knowledge
• Escalated — p1 threads with a named decision-maker
• Resolved — closed within the last seven days, searchable

Notice that platform is a filter inside each folder, not the folder itself. The person answering a shipping question should not care whether it arrived on Instagram or Facebook.

Response targets that reflect risk

Set three, not twelve. Write them on the wall.

• P1: first response within 60 minutes during coverage hours. A public complaint that sits unanswered for two hours becomes the comment people screenshot.
• P2: within 4 hours. These are usually money questions in an active conversation.
• P3: within 24 hours. Praise, casual comments, non-urgent asks.

Two rules make these real. First, "coverage hours" must be written down — a target with no hours attached is a target that fails overnight and demoralises everyone. Second, measure first response, not resolution. First response is the number the customer feels; resolution can take a day and that is fine as long as the customer heard from you.

Response targets are promises to your team about how hard the job will be, as much as promises to customers about how fast you will answer.

Triage: deciding what gets answered first

With tags and targets in place, triage becomes a sequence instead of a judgement call. Train everyone on the same five questions.

1. 1Is it public and negative? Public criticism outranks everything. It is visible to everyone who scrolls, and your reply is read by lurkers, not just the complainant. Answer it first, calmly, without defensiveness.
2. 2Is someone about to spend money? Presale questions in an active thread go next. If a person asked three questions under your launch post and got none answered, they have moved on to a competitor's comment section.
3. 3Is it a support issue with an order? Route to the person who can check the order. Nothing destroys trust faster than a cheerful reply that does not answer the actual question.
4. 4Is it praise or UGC? React and acknowledge within the day. This is the cheapest goodwill you will ever buy.
5. 5Is it spam? Hide, delete, block. Do not deliberate.

The triage sequence in practice

Work top to bottom in one pass rather than picking favourites. Open the queue, filter to P1, clear it, then P2, then P3. The discipline of ordering prevents the common failure where a pleasant, easy-to-answer comment steals attention from an angry one simply because it appeared first.

Build a small rule set for the obvious automatable calls. Spam keyword lists, obvious bot patterns, and duplicate messages arriving from the same person within a few minutes can all be handled before a human looks. Everything else goes to a person with the intent tag already attached — that is what makes the queue fast rather than merely unified.

Handling the pile-up

Every inbox eventually hits a Monday morning with 200 unopened items. Do not reply to 200 items. Instead:

• Filter to P1 and P2 only, and clear those first — usually 30 to 50 messages
• Bulk-hide obvious spam rather than reading it
• Send one acknowledgement post or story reply for common questions, such as a shipping-times update, which often removes 20 to 40 percent of the queue at once
• Set a visible "we are behind, we are catching up" note if the backlog will take more than a day

Acknowledging a backlog honestly is almost always better than a silent scramble. Customers are far more forgiving of a queue than of being ignored.

Triage checklist

• Five triage questions memorised by everyone who replies
• One pass per session, P1 first, rather than opportunistic picking
• Spam rules running before human review
• Assignee set before reply, so nobody answers the same thread twice
• Backlog protocol written down for Monday-morning pile-ups
• Every reply tagged on close, so categories stay accurate for reporting

Response frameworks: templates that still sound human

Saved replies are the difference between a two-minute answer and a twenty-second one. They are also the fastest way to make your brand sound like a call centre, if you write them badly. The trick is to template the structure and leave the specifics untemplated.

Write templates for your top five

Go back to the categorisation from your audit. Your top five message types will usually cover more than 60 percent of everything that needs a human reply. Write one template each.

TEMPLATE: shipping times (presale)
Hi [name] — thanks for asking. [City] usually takes [n] working days,
[region] takes [n]. Tracking lands by email the moment it leaves us.
Anything specific about your order I can check?

TEMPLATE: item arrived damaged (order)
[Name], that is not the experience we want you to have. Send me your
order number and a photo of the damage here and I will sort a
replacement today. Sorry about this.

TEMPLATE: praise (praise)
Thank you — genuinely glad it is working for you. Would you mind if we
shared this on our page? We will tag you.

TEMPLATE: pricing or discount (presale)
[Name] — the [product] is [price] and there is no code needed. If you
are buying for a team of [n] or more, message me and I will sort a
volume rate.

TEMPLATE: feature request (lead)
Noted, and thank you — [specific thing] is on the list. I cannot give
you a date, but I have logged your account against it so you will hear
from us first when it ships.

Notice what is templated: the sequence of acknowledgement, information, and next question. What is not templated: the customer's name, their specific situation, and the detail that proves a person read it. Always leave one slot that forces a human to type something only they could know.

Tone rules for the awkward cases

Some messages need a script because the writer is nervous.

• Public complaint: acknowledge first, move to DM second, never argue in the thread. "You are right that we missed this — I am sorting it now, sending you a message" beats any defence of your process.
• Price objection: never apologise for the price. State what is included, offer the alternative tier, and give a real reason the price is what it is.
• Legal or safety language: do not improvise. Route it to the person authorised to speak, and reply only with "Thank you, we are looking into this and will come back to you today."
• Angry all-caps: do not match the volume. Short sentences, plain words, one clear next step.
• Misinformation about your brand: correct it once, politely, with a source, then stop replying. Feeding an argument trains the algorithm to show it to more people.

Platform quirks that change how you reply

The message is only half the context. The surface it arrived on changes what you are allowed to do and what the person expects.

• Instagram. Story replies and comment notifications sit in different places, and story replies are frequently missed because they never appear in the main comment stream. Comments can be moved into a private thread from the reply bar, which is ideal for order questions. The hidden-words filter runs before you see anything, so check it monthly — it silently eats legitimate questions containing words like "refund" or "cancelled".
• TikTok. Comments arrive fast, out of context, and often from people who have never seen your bio. Keyword filters in the comment tools are essential here more than anywhere, because volume spikes in hours. Direct messages are frequently closed by users, so do not build a process that depends on moving a TikTok commenter into DMs. Reply publicly or not at all.
• YouTube. Your held-for-review queue is the real inbox. Hearting a comment lifts it and tells the commenter you saw them, which is the cheapest acknowledgement available. Long, detailed comments deserve replies because they are read by everyone who scrolls, and YouTube weighs conversation in browse.
• LinkedIn. Messaging is constrained by connection state, so a company-page comment cannot always be moved to DM. Answer fully in the thread, then use DM only for people you are already connected to.
• Facebook. Page messages, comment replies and group activity all land in the same general area but carry very different expectations. Inside a group, drop the brand voice and answer like a member; outside it, keep the published coverage hours visible in the page's auto-reply.
• X. Replies are public, fast-moving and screenshot-friendly. Prioritise correcting facts here over being charming, and take anything with account details straight to DM requests — but note that many users never check that folder, so confirm the move with a public line first.

Build these differences into your triage rules rather than leaving them to individual judgement. A reply that is perfectly correct on LinkedIn can be invisible on TikTok.

Measure whether your templates help

If reply time drops but reply quality drops with it, the templates are too rigid. Sample ten replies a week and ask one question: would the customer know this was written for them? If more than two in ten would say no, loosen the template and add a required specific detail.

Escalation: when a comment needs someone above you

The most common inbox failure is not slow replies — it is confident wrong replies. A junior community manager promising a refund they cannot authorise, or denying a policy that changed last month. Escalation rules prevent this, and they protect your team as much as your customers.

Build a three-level escalation matrix

• Level 1 — community manager. Can answer questions, apply discounts within a published limit (say 10 percent), issue replacements up to a fixed value, and acknowledge complaints. Cannot discuss legal, safety, refunds above the limit, or partner disputes.
• Level 2 — team lead or support lead. Can approve refunds, handle serial complaints, speak about outages and delays, and make exceptions. Must respond to an escalation within two hours during coverage hours.
• Level 3 — founder, head of support or legal. Anything involving safety, money beyond a threshold, government or media attention, data privacy, or a threat of legal action. Must respond within one hour, twenty-four hours a day, for the duration of the incident.

Write the thresholds as numbers, not vibes. "Refunds above 5,000 PKR go to Level 2" is a rule. "Big refunds go to a senior person" is a guess.

Signals that should trigger escalation automatically

Train your team to recognise these, and never answer them in the thread without a check:

• Mentions of lawyers, regulators, journalists or police
• Any claim of injury, allergic reaction or safety failure
• Screenshots of internal messages, dashboards or employees
• Accusations involving a named individual rather than the company
• Viral velocity — a negative post being reshared far beyond your follower count
• A customer who has already been failed by your support team once

Post-escalation closure

Every escalated thread ends in three steps: the Level 2 or 3 owner replies to the customer, the original Level 1 person is told what happened, and the thread is tagged with the reason so the monthly review can spot repeat causes. Skipping the second step is how your front line stops escalating — they never see the outcome, so they assume escalation means silence.

Coverage: hours, handoffs and time zones

A queue without a rota is a queue that empties when the most eager person happens to be online. Coverage design answers four questions: when are we open, who is on, how do we hand over, and what happens when nobody is there.

Publish your coverage hours

Pick hours you can actually staff, not hours you wish you could. For most brands serving a national audience, a realistic starting point is 9am to 9pm local time, seven days a week, with a lighter weekend shift. If you genuinely cannot cover nights, say so in your bio and set an auto-reply in DMs that gives a next-day window. Unpredictability is worse than a stated limitation.

Design the handoff

Twice a day, at shift boundaries, run a two-minute handoff. It has a fixed shape:

1. 1Open thread count at start and now
2. 2Every P1 open, with a one-line status each
3. 3Anything promised to a customer with a deadline
4. 4Anything you started and deliberately left for the next person

Write it in the inbox, not in a chat app. If the handoff lives somewhere else, the next person will not read it before they open the queue.

Cover the gaps honestly

If you run a team spread across Karachi, London and Toronto — as many social teams now do — the handoff across time zones becomes your most important document. The receiving person should be able to open the queue cold and know, within thirty seconds, which three threads are live and what was promised.

For the hours nobody covers, use the two honest options: a published away window with a promised reply time, or scheduled posts that explicitly state the wait. Never leave a DM inbox open and silent.

Coverage checklist

• Coverage hours written publicly and internally
• Named on-duty person for every hour of coverage, not a shared "team" label
• Two-minute handoff format used at every shift change
• Escalation owner named for nights and weekends
• Away message configured with a real, specific reply window
• Monthly review of unanswered messages against published hours

Metrics that prove the inbox is working

Five numbers are enough. Resist the temptation to build a dashboard of thirty.

• First response time, median and 90th percentile. The median tells you the normal experience; the 90th tells you how bad the bad days get. If your median is 40 minutes and your 90th is 9 hours, your process has a hole — usually one day of the week.
• Reply rate on messages that need a reply. Not all inbound needs a human. Measure the percentage of actionable messages that received an answer within your target. Above 95 percent is healthy; below 85 percent you are visibly neglecting people.
• Backlog age. The age of the oldest unanswered actionable message. This single number catches everything else — if your oldest open thread is five days old, no other metric matters.
• Escalation volume and reasons. Rising escalations usually mean your templates or training are out of date, not that your customers got angrier.
• Comment sentiment or complaint share. Track the percentage of public comments that are negative, month over month. Slow-moving, but it is the number that shows whether the inbox is changing how people feel about you.

Reading the numbers without lying to yourself

Small queues produce noisy data, and three habits keep you honest.

• Compare weeks, not days. A single viral post can double your median response time for one day and tell you nothing about your process. Use rolling seven-day medians for everything except P1, where you look at every instance individually.
• Separate coverage hours from calendar hours. If you measure a 4-hour target across a night when nobody is on duty, you will spend months trying to fix a staffing decision disguised as a performance problem. Measure against the hours you promised, then judge whether the promised hours are enough.
• Watch the distribution, not just the middle. A healthy median with a catastrophic 90th percentile means the system works only when someone diligent is on shift. That is a rota problem, not a reply problem.
• Count what you did not answer. Every report should include the number of actionable messages that received nothing. It is the least flattering number in the set and the only one that reliably predicts a future complaint.

Report these five numbers in one short note each week, in plain sentences rather than charts. "P1 median fell from 74 to 38 minutes; backlog age rose to two days because Saturday had no cover" tells a team what to change. A dashboard tile does not.

Tie it to outcomes your business cares about

The inbox conversation that matters most to a founder is the one that produced revenue. Track two simple counts: how many presale threads ended in a click or a DM that became an order, and how many public complaints were resolved without turning into a reposted screenshot. Even manual attribution — a tag applied when a conversation converts — is enough to show that reply speed is not a cost centre.

In PostHub, the inbox view keeps response times and assignment history alongside the conversation itself, so you can pull these numbers without exporting anything. Teams that want the full picture across scheduling, inbox and analytics can see how the pieces connect on the PostHub overview.

If you cannot name the oldest unanswered message in your queue right now, you do not have an inbox. You have a pile.

Automation: what to automate and what never to automate

Automation belongs at the front of the queue and at the edges of it. It does not belong in the middle, where judgement lives.

Automate confidently

• Spam and keyword filtering, with a list reviewed monthly
• Intent tagging on arrival, so the queue is sorted before a human opens it
• Routing rules that send presale threads to sales staff and order issues to support
• Saved replies for your top five message types
• Out-of-hours acknowledgement that states a real reply window
• Auto-hide for comments matching a harassment or profanity list, with a human reviewing every hide
• Internal notifications for P1 tags, sent to the person on duty rather than a channel everyone mutes

Never automate the reply itself for these

• Complaints and refunds — the customer is already unhappy; a bot reply confirms they do not matter
• Anything about billing, cancellations or data privacy
• Safety, health or legal topics
• Grief, bereavement, medical or personal bad news in a comment
• Public disputes with another account
• Any thread where a person has already been waiting more than a day

A useful test: if the automated reply would look bad screenshotted next to the customer's message, a human must write it. If the automated reply simply saves a person three minutes of typing shipping times, automate it.

Review automation monthly

Every month, read twenty threads that hit an automated rule. Count how many were handled well. Below 90 percent accuracy, tighten the rule. Automation that is 70 percent right generates more work than no automation at all, because someone has to find and fix the errors.

Setting the whole system up in PostHub

Here is a practical rollout for a team of three to six people, sized to finish in about two weeks without disrupting the daily schedule.

1. 1Week one, days one to two. Connect every account you actually use. Disconnect the ones you do not — an unconnected account is still receiving messages, so either staff it or close it.
2. 2Week one, days three to four. Build the tag set, the three folders and the three response targets exactly as described. Keep it at nine tags.
3. 3Week one, day five. Assign roles: who is Level 1, who is Level 2, who is Level 3. Write the escalation numbers down.
4. 4Week two, days one to two. Write the five templates and the away message. Have everyone send each other a test complaint so the team feels how they land.
5. 5Week two, day three. Set automation rules for spam, routing and out-of-hours only. Nothing else yet.
6. 6Week two, days four to five. Run the system live with one shift change, then hold a thirty-minute review: what was confusing, which tag did nobody use, where did a reply get duplicated.
7. 7Day fourteen onward. Review metrics weekly, tags monthly, templates quarterly.

Most teams that follow this sequence are answering P1 threads inside an hour by the end of week two, not because they work harder, but because the queue finally tells them what to do first.

PostHub puts scheduling, approval and the shared inbox in one workspace, so the person who publishes the post and the person who answers its comments are looking at the same thread history. If you are trialling it with a team, start here and compare the plans and limits against how many accounts you actually need to connect.

The first thirty days: what to watch

• Days 1 to 7: expect the queue to look worse. You are now seeing messages you were previously missing.
• Days 8 to 14: backlog age should fall sharply. If it does not, your coverage hours are fiction — shorten them until they are true.
• Days 15 to 21: median first response time stabilises. Compare it against your three targets.
• Days 22 to 30: tag data becomes usable. Drop any tag with fewer than five uses, and check whether presale questions are actually being answered fastest, since they are the ones closest to revenue.

Key takeaways

• Scattered replies are an ownership problem, not a motivation problem — every inbound surface needs a named owner and a written response target.
• Audit for seven days and count real volume; most teams are missing an active channel nobody checks.
• Tag by intent, not content: nine tags stay usable, forty-two collapse within a month.
• Set three response targets tied to risk — P1 within 60 minutes of coverage, P2 within 4 hours, P3 within 24 hours — and measure first response, not resolution.
• Triage in a fixed order: public negativity, purchase intent, support, praise, spam.
• Template the structure of a reply and leave a slot that only a human could fill.
• Write escalation thresholds as numbers, and never automate replies about money, safety, privacy or grief.
• Publish your coverage hours and run a two-minute handoff at every shift change; measure backlog age weekly as the one number that cannot hide problems.

FAQ

How is a unified social inbox different from just keeping all the app notifications on my phone?

Notifications tell you something arrived; they do not tell you whether it was handled, who owns it, or how urgent it is. A unified inbox adds three things a notification cannot: a single list where nothing is hidden behind a login you do not have, an assignee on every open thread so two people never reply twice, and response targets that let you see which conversations are about to breach. The phone keeps you aware. The queue keeps the team accountable.

What response time should I actually aim for?

Set three targets instead of one. Sixty minutes for public complaints and anything with safety, legal or viral risk; four hours for purchase questions in an active conversation; twenty-four hours for everything else that needs a reply. Attach them to written coverage hours — a one-hour target you cannot staff overnight is worse than a stated eight-hour window, because it creates failures you will eventually stop even measuring.

How many tags and folders does a small team need?

Fewer than you think. Nine tags across three dimensions — intent, priority and owner — plus about six folders organised by coverage rather than by platform. If a tag has not been applied more than three times in a month, delete it. The purpose of tags is to sort a queue quickly and to produce a monthly report you can read; a vocabulary nobody applies does neither.

Should we reply to every comment?

No. Reply to every actionable message: questions, complaints, praise that deserves a thank you, and anything where a reply changes the next thing the person does. You can skip pure reactions, emoji, tags of friends, and obvious spam. If someone asks a question you have already answered publicly, a short "short answer: yes, 3 to 5 days — details above" is enough. Universal reply rates sound admirable and usually just burn the team out on no-value threads.

When should we move a conversation from comments to DMs?

Move to DMs as soon as the conversation needs an order number, an account detail, a refund, or anything the customer would not want public. Acknowledge in public first, then continue privately — jumping straight to DM makes it look like you are hiding. Conversely, keep it public when the answer is useful to everyone, such as shipping times or pricing, because those replies become free reassurance for the next hundred people who scroll past.

How do we keep quality up as we hire more reply people?

Three things: a written voice guide for replies with examples of good and bad, the five templates as a floor rather than a ceiling, and a weekly sample of ten replies reviewed by the team lead. Quality slips when new starters are given a queue and no standard. The review is not about catching mistakes — it is about sharing the phrasing that worked, so the team's language improves in public rather than in private.