Messenger chatbots arrive: the first useful cases for business

At F8 in April, Facebook opened the Messenger Platform in beta: any Page can now connect a bot through the Send/Receive API, with templates for structured messages, buttons, image carousels and a persistent welcome screen. Messenger has about 900 million monthly users, and it sits on the phone of nearly every US customer you have. Alongside it came the Messenger Codes you can print on packaging, the "Send to Messenger" and "Message Us" plugins for websites, and m.me short links.

Early bots from CNN, 1-800-Flowers, Poncho and a handful of others set the tone: some useful, some embarrassing, most judged within a week. Kik launched its own bot store in April, Telegram has had a bot API for a year, and Slack bots are everywhere in offices. Microsoft talked about "conversation as a platform" at Build. The pattern is clear: messaging apps are the new app stores, and Messenger is the one where your customers already are.

Two limits shape everything. First, bots cannot message people who have not messaged first. Second, Facebook's review process and platform policy restrict promotional content, and it will not hesitate to block a Page that spams. A bot is an inbound channel with a memory, not a push list.

1. Frequently asked questions. A restaurant answering "are you open" and "do you take reservations", a clinic answering "do you take my insurance", a software company answering "how do I reset my password". If you can list the twenty questions your team answers most, a bot built as a menu of those questions handles 40 to 70 percent of inbound messages before a person is needed.

2. Simple structured requests. Order status, appointment availability, store hours by location, a quote for a defined service, a return request. These need a connection to your own data, which is where a developer enters. The bot gathers the inputs with buttons and quick replies, hits your system, and returns an answer. 1-800-Flowers lets you order a bouquet this way; a regional moving company can do the same for a quote.

3. Content delivery people opt into. A daily summary, a weather or price alert, a lesson sequence, a notification that a job is done. CNN's bot is the best-known example. For a business, "message me when my order ships" or "send me the weekly specials" is the honest version, and it beats email open rates by a wide margin right now because the channel is new and uncluttered.

What does not work yet: open-ended conversation. Natural language parsing from Wit.ai, which Facebook owns, or similar tools can catch intents like "track my order", but the moment a customer types a two-sentence complaint with a typo, a general-purpose bot fails and the customer blames you. Keep free text to narrow slots, and always offer a path to a human.

• Open with the menu. The welcome screen and first message should show three to five buttons that describe exactly what the bot can do. Do not pretend it can do more.
• Buttons over typing. Every step that can be a button or quick reply should be one. Free text is where bots break.
• Short messages, one idea each. Messenger is read on a phone, in a thumb-scroll.
• A human exit everywhere. "Talk to a person" should be one tap away, and someone should actually answer within business hours. State those hours.
• Confirm before acting. Repeat back the order, the appointment, the address. Then act.
• No surprise messages. Only send what the person opted into, within the window policy allows.
• A personality, lightly. A name and a tone are fine. Jokes in every message are not.
• Plan the failure message. "I did not understand that. Here is what I can help with:" followed by the menu, is the most important line in the whole bot.

Add ongoing attention. A bot is not finished at launch: someone has to read the conversation logs weekly, add the questions people really asked, and prune the ones they did not. Budget a few hours a week for the first three months.

Measure four things: share of conversations resolved without a human, response time when a human is needed, the number of people who return to the bot a second time, and the business outcome it exists for, such as bookings or reorders. Facebook's own Page analytics for Messenger are thin, so log events in your own analytics or a simple database from day one.

Spam and policy. The 24-hour messaging window and the restrictions on promotional content are enforced. Businesses that treat Messenger like an email list will be cut off.

Overpromising. The word "AI" in your launch post invites customers to test the bot with hard questions. Describe it as a quick way to get answers and place orders, and it will be judged on that.

Privacy. Do not collect payment card numbers in chat. For payments, send people to a web checkout. Collect only what the use case needs, and say how it is used.

Abandonment. Thousands of bots launched in May will be dead by autumn because nobody owns them. Assign an owner before you build.

We build bots as part of a connected system, which means the automation work and the growth, brand and web work talk to each other.

Discover. We read the Page inbox, support tickets and call notes, and count what people actually ask. The bot's scope comes from that list, not from a feature wish list.

Plan. A flow map with every button, every message and every failure path written out, plus the systems the bot must touch and the human handoff rules. Brand voice is set here, in a short style sheet.

Build. Either a no-code builder with light integration or a custom bot on the Messenger API, with event logging, a human handover into the tools the team already uses, and the website plugins and Messenger Codes to drive people in.

Improve. Weekly log review for the first quarter, adding and removing intents, tightening copy, and reporting resolution rate and the business outcome the bot was built for.

A representative engagement is our workflow transformation work, where a service business with a flooded inbox moved its most common requests into a structured, menu-driven flow connected to its scheduling and customer records, with a clean handover to staff for anything outside scope. The outcome that mattered was not the bot itself but the hours returned to the team and the faster replies customers noticed.