Filter, answer, route: AI on WhatsApp Business

A customer sends a WhatsApp message at nine at night to the salesperson who helped them last month. It lands on her personal phone. She answers, quotes a delivery date from memory, and that’s where the conversation stays. It’s recorded nowhere anyone else can search, so nobody can check later what was promised, and when she’s away next week the follow-up sits unread on a device nobody else can open.

Imagine four different buyers, each at a version of that company. One wants AI to filter spam, answer simple questions on WhatsApp Business, and send complex requests to the right department. One is a retailer whose customers ask about products and care advice over messaging apps while sales staff and consultants pass them around. One is an owner who says staff handle chats separately and she has no oversight. One has sales teams split by product and by region, and needs an incoming lead to reach the right team and land in the right CRM pipeline.

The details differ, but all four are describing the same problem: the work is happening somewhere the company can’t see.

The scale is why this keeps coming up. Meta counts more than a billion active threads between people and businesses across its messaging apps every day. In India, WhatsApp’s biggest market, more than 500 million people use it, and in November 2025, 94% of its monthly users there opened it daily, according to Sensor Tower data reported by TechCrunch. For a large share of the world the customer relationship is a chat thread.

The problem all four buyers describe

The owner with no oversight puts it most plainly, and she’s describing three failures at once.

• No record. Nobody can answer what was promised, quoted or agreed, because the conversation is on somebody’s phone.
• No measurement. Nobody knows how many messages arrived, how many were answered, or how long the answers took.
• No continuity. When an employee leaves, is ill, or takes two weeks off, the customers she was handling go unanswered.

Panopto’s 2018 survey with YouGov, covering 1,001 US employees at organizations of 200 or more, found 42% of institutional knowledge is unique to one person and unavailable to coworkers. A customer chat history on a personal phone is knowledge of that kind: no colleague can reach it, and there’s no other copy anyone can read.

Fixing that needs no AI at all. It needs the conversation to arrive somewhere the company owns. The automation makes the move worth making, because an owner who can finally see every message now also has every message to handle.

Your account, your staff’s accounts, and your customer’s app

There are three flavors of WhatsApp here, and what you can automate depends on which one you’re talking about.

• The WhatsApp Business Platform is what a company connects. It works with a WhatsApp Business account and the phone number registered to it. Messages sent to that number can be read and answered by software, and replies go back out from the same number.
• Your staff’s personal accounts are ordinary WhatsApp on their own phones. The platform gives software no access to them, so we can’t read them, and that’s the source of the oversight problem.
• Your customer is on the ordinary app too. They see one continuous thread with your business, whether an employee or a workflow wrote the last message.

So the first step is to move customer conversations off personal numbers onto a business number the company owns. Filtering, answering and routing all depend on it.

What “filter the spam” means on a business number

Email spam gets caught by sender reputation, authentication records and volume patterns before anyone reads it. On a business number, every message comes from a phone number WhatsApp has already registered, so what gets through is a different mix: wrong numbers, resellers pitching you, people asking about a product you stopped selling two years ago, and customers who already emailed the same question an hour earlier.

Sorting that is classification, and AI is good at it. What matters is what happens to the rejects. An email filtered in error sits in a spam folder somebody can search. A message ignored in a chat thread looks to the customer like you’re ignoring them, because their app shows them it was delivered. Ask any vendor for a low-priority queue rather than a delete, and put reading it on someone’s weekly list.

Responding is easy, but answering correctly isn’t

Opening hours, the address, delivery times, the return window, whether a product suits a particular use, how to look after it once it arrives. AI answers all of those in seconds. It answers them wrong just as fast when the answer isn’t written down anywhere it can read.

That’s the part that’s easy to underestimate. Configuring the automation is quick. Writing down the answers your team currently holds in their heads takes longer, and it decides whether the replies are any good. AI with nothing to retrieve invents a plausible return window, and telling the customer the wrong return window leads to an argument about a refund.

Meta ships its own answer to this part. Meta Business Agent, announced on June 3, 2026, answers questions, recommends products, books appointments and qualifies leads inside the chat, and lets you decide when a team member steps in. Meta said more than a million businesses were already using one on WhatsApp and Messenger, and getting started is free, with paid subscriptions to follow. If answering the repeat questions on Meta’s apps is all you need, start there. When the same customer also emails, or the lead needs an owner on a task list your whole team works from, that part goes beyond what Meta announced.

Routing needs someone to own the table

Routing rules are easy to write and easy to get wrong, for reasons that have nothing to do with AI. “Send product questions to the product team” assumes a table of which teams exist, which products and regions each one covers, who takes over when an owner is away, and what happens to a message that matches two teams or none. Somebody has to own that table and keep it current.

Chat also gives the AI less to read. Email routing can use a To: address, a subject line, a signature and a sender domain. A chat message gives you a phone number and a sentence, so the AI decides from the message itself and the history with that number.

The buyer whose sales teams are split by product and region adds a harder step. A lead has to land in the right CRM pipeline, which means the software needs to write into your CRM rather than notify someone to do it. Ask which CRMs a vendor writes to, and whether creating a deal in the right pipeline is something you configure or something you request.

What still needs a person

Four categories are worth keeping manual on a customer channel, whatever the vendor demo shows.

• Money already paid. Refunds, a disputed charge, a discount somebody promised.
• A customer who is upset. An automated reply to an angry message reads as a brush-off even when the content is correct.
• Anything that commits the business. A delivery date, a custom quote, an exception to policy.
• Anything your written answers don’t cover. For a retailer giving care advice, the limit is what the manufacturer published. Past that, a consultant answers.

What makes this workable is an approval gate: the workflow drafts the reply and a person sends it, and once you’ve watched a pattern get it right often enough, you let that one go out on its own.

Where we fit

this+that reads Gmail, Outlook, Slack, Microsoft Teams, Google Chat, and WhatsApp Business as one stream, turns what arrives into tracked work, and runs the follow-through. Each channel keeps its own thread. What’s shared is the task list, the workflows and the knowledge behind them.

• Every connected channel in one app. Each keeps its own stream, and a WhatsApp Business message and an email about the same order get handled the same way by the same team.
• DoBox reads each conversation and turns what it asks for into a task on a board, with the thread attached and a person’s name on it.
• Workflows do the routine part. A workflow can branch on a condition, loop over a list, ask the AI for a judgment call mid-run, reach a connected tool over MCP to write the lead into your CRM, draft or send the reply, and hold anything that needs a person at an approval gate. On WhatsApp Business a new message starts the run, and the workflow’s first step does the sorting. Once the message is a task, DoBox events such as a comment or a resolution can start workflows too.
• The Brain grounds the answers, so a reply to a question your team has answered before comes from what the team decided rather than from an AI’s guess. The customer support page shows how that works for a support team.
• Contacts update themselves from WhatsApp Business messages with no configuration, reading contact details out of the conversations as they do on every other channel.

On security, we hold SOC 2 Type I as of July 31, 2026, with Type II underway, and CASA Tier 2. Your messages are never used to train AI models, ours or our providers’. The security page has the rest, and the report goes out under NDA.

Five questions to ask before you buy

1. Which channels does it read today? Ask for what’s connectable this week, listed separately from what’s planned.
2. What happens to a message the spam filter rejects? A queue somebody reviews beats a delete you’ll never find out about.
3. Where do the answers come from? If a vendor can’t show you which document produced a reply, the AI is guessing, and you’ll find out which sentences it invented from a customer.
4. Can it write into our CRM, and which pipeline? Creating the deal in the right place is a different capability from sending an alert about it.
5. What software do employees use to reply? If it’s still a chat app on their own phones, nothing about the oversight problem has changed.

Frequently asked questions

Can AI reply to customers on WhatsApp Business without anyone checking? Technically yes, but don’t start there. Run every draft through an approval gate at first, watch which categories it gets right, then release those one at a time. Opening hours and the return window are natural first candidates. Refunds stay with a person.

Do we need a WhatsApp Business account, or will our normal number work? You need a WhatsApp Business account and a number registered to it. That’s what the platform gives software access to. A personal account on an employee’s phone isn’t part of the platform, so we can’t connect it, which is also why the work on it is invisible to you.

Can a message create a lead in our CRM automatically? Yes, through a workflow you set up. A workflow can reach your CRM over MCP and create the record in a named pipeline. Contacts inside this+that keeps itself current from your connected email and chat accounts, WhatsApp Business included, but keeping Salesforce or HubSpot current is a workflow somebody builds, not a switch.

Is this+that a help desk? No. A workflow can label support messages, queue them as tasks and draft a first reply. Teams pick us when requests arrive across everybody’s channels and need routing, follow-through and a record.

If your customers message your staff’s personal numbers and your company can’t see what was said, the first move is getting those conversations somewhere shared. Get started with this+that, or write to sales@thisandthat.chat.