Mastering Marketing Chatbots: A Marketer's Playbook

A year after Facebook opened Messenger to bots, the novelty has worn off and the data has come in. Early open rates for bot messages looked spectacular next to email, then dropped off just as fast once the initial curiosity wore out. This is the playbook for building a chatbot in 2017 that survives past its first week.

Start by admitting what your bot is

Nearly every marketing chatbot in production right now is a decision tree with buttons, not artificial intelligence that understands language. Design it accordingly. Trying to make it feel conversational when it can’t parse open-ended input creates the exact frustration that tanked engagement for bots launched purely to chase the 2016 hype cycle. Be upfront in the interface about what the bot can and can’t do; users forgive a limited menu far more readily than they forgive a bot that pretends to understand and gets it wrong.

Pick one job, not five

The chatbots still working well a year in are the ones scoped to a single, specific task: tracking an order, booking an appointment, answering a fixed set of support questions. The ones that tried to be a general-purpose assistant for a brand, handling everything from customer service to content discovery to promotions, mostly collapsed into confusing menus that users abandoned. If you’re starting a bot project this year, write down the one job it needs to do well before you write a single line of dialogue.

Build the exit before you build the bot

Every flow needs an obvious, immediate path to a human. This isn’t a nice-to-have; it’s the single biggest predictor of whether a bot frustrates or helps. Facebook’s own platform updates through the year made handoff to a live agent easier to build, recognizing this was where most bot experiences were breaking down.

Use Messenger’s Discover tab and broadcast tools deliberately

Facebook expanded Messenger’s Discover tab this year, giving bots a dedicated place to be found outside of a direct link or ad. Businesses relying on this for discovery need content and utility worth finding, not just a bot that exists because competitors have one. Broadcast messaging to subscribed users, meanwhile, is powerful and easy to abuse; treat it with the same restraint you’d apply to email frequency, because Messenger users unsubscribe from an over-messaging bot even faster than they unsubscribe from an email list.

Measure the right thing

Open rate on the first message a user sends is close to meaningless if that’s where the relationship ends. Track completion rate for whatever the bot’s single job is: did the person get their order status, book the appointment, get their question answered without needing to escalate. That’s the number that tells you whether the bot is doing its job, not how many people happened to message it once.

Revisit scope every quarter

The bot that made sense in April may not make sense in October if your business has changed what customers ask about most. Review the actual conversation logs regularly, not just the metrics dashboard, and prune or expand the bot’s scope based on what people are trying to accomplish, not what seemed like a good idea when it launched.

The takeaway a year in

The businesses still getting value from chatbots in 2017 are the ones that treated the 2016 launch wave as a beta test rather than a finished product. Narrow the scope, make the human handoff effortless and measure task completion instead of vanity engagement. That’s the difference between a bot that quietly earns its keep and one that gets quietly abandoned by both the business and its customers.