A Practical Guide to Marketing Chatbots

Ask two marketers in 2017 what a chatbot could do and you’d get two different answers, depending entirely on which bot they’d built or used. That gap between the promise and the reality is the whole story of chatbots this year, and understanding it is what separates a tool that helps customers from one that quietly drives them away.

Where the sudden interest came from

Facebook opened its Messenger platform to bots in April 2016, and by early 2017 more than 100,000 businesses had built one, drawn by the promise of reaching people inside an app they already had open. Slack, Kik and WeChat pushed the same idea in their own ecosystems. Tools like Chatfuel, ManyChat and Drift made building a bot a weekend project rather than a development sprint, and that accessibility is exactly why the space filled up fast with bots that weren’t good.

What “AI chatbot” usually meant in practice

Most bots marketed as AI in 2017 were decision trees with a friendlier name. A user picked from button options or typed a keyword, and the bot matched it against a pre-built list of intents. Some tools layered in basic natural language processing, often through Facebook’s own Wit.ai or the newly Google-owned API.ai, which could recognize a handful of phrasings for a given intent. That’s a real step up from pure keyword matching, but it’s nowhere close to genuine understanding. Ask a 2017 bot something slightly off-script and you got the same result every time: “Sorry, I didn’t understand that.”

The distinction matters for anyone building one:

• Decision-tree bots follow pre-programmed flows. Predictable and easy to control, but brittle the moment a user goes off the happy path. Fine for narrow, repetitive jobs.
• NLP-assisted bots recognize a wider range of phrasing for the same intent, reducing dead ends without eliminating them. Better, but still limited to what you explicitly trained.

Where these bots earned their keep

Used for the right job, even a simple bot delivered real value in 2017:

• After-hours lead capture. A bot that greets a visitor, asks a few qualifying questions and books a follow-up beats a contact form that sits unanswered overnight.
• FAQ deflection. Store hours, shipping policy, order status: the questions that repeat endlessly are exactly what a scripted bot handles well.
• Messenger marketing. Businesses used bots to deliver content, promotions and reminders inside Messenger, where open rates ran well ahead of email at the time.
• Simple routing. Getting a visitor to the right page, department or human faster than a menu of links.

Where they fell apart

The same bots that helped also generated a wave of frustration, and the failure modes were consistent:

• No escape hatch. A bot that hits its limit and offers no path to a real person is the single most common complaint, and it’s entirely avoidable.
• Overpromising the “AI” label. Marketing a decision tree as an intelligent assistant sets expectations the bot can’t meet, so the letdown lands harder.
• Deflection as the only goal. Bots built purely to cut support costs, with no real attempt to resolve the issue, read as exactly what they are.
• No fallback for the unexpected. Real conversations wander. A bot with only three trained paths breaks constantly.

Building one that doesn’t annoy people

The businesses getting genuine value from bots in 2017 tended to follow a few consistent habits. Start with one narrow job rather than “handle everything,” whether that’s lead qualification or order status. Write the decision tree around the actual questions people ask, pulled from your support logs rather than guessed at a whiteboard. Always give the bot an obvious way to hand off to a human, and put that option somewhere the user can find it without hunting. Be upfront that it’s a bot; pretending otherwise erodes trust the moment someone notices, and someone always notices.

Measure resolution, not just conversation volume. A bot that handles ten thousand chats and resolves none of them isn’t working, no matter how the usage chart looks.

The honest state of things

Chatbots in 2017 are a useful tool for narrow, repetitive jobs, wrapped in marketing language that oversells what the underlying technology can do. The businesses getting real value aren’t the ones with the most sophisticated-sounding bot; they’re the ones who picked one job, built the decision tree around real customer questions and left the door open to a human for everything else. That’s a modest ambition, and it’s exactly why it works.