What I learned from 283,483 ManyChat messages

This fall, our team had the opportunity to work with a (confidential) visionary client who doubled-down on Facebook Messenger as a revenue channel for their business. Alongside the client’s incredible team we were able to achieve what I never thought might be available using ManyChat, an analytics-driven messenger bot that favored data and desire-path thinking over “gut feel” and anecdotes.

I’ll save our bot-funnel analytics journey for another post. Just know that the number of tools with “effective functionality” — for even advanced bot-builders — can be counted on one-hand. If your messenger bot is designed for conversions and will be running for more than a week, start looking now.

At the time of writing, our messaging volume was just over 283,000 messages. After analyzing the behavior of the thousands of people making their way through this ecommerce journey, we distilled our analysis approach into two types of funnel views.

One funnel view visualized the percentage of people who started and completed the journey. The other view visualized each step a person took in their journey from start to finish. We used the two step flow as an overall health-check. The multi-step flow identified the points of greatest “decay.”

This fall, our team had the opportunity to work with a (confidential) visionary client who doubled-down on Facebook Messenger as a revenue channel for their business. Alongside the client’s incredible team we were able to achieve what I never thought might be available using ManyChat, an analytics-driven messenger bot that favored data and desire-path thinking over “gut feel” and anecdotes.

I’ll save our bot-funnel analytics journey for another post. Just know that the number of tools with “effective functionality” — for even advanced bot-builders — can be counted on one-hand. If your messenger bot is designed for conversions and will be running for more than a week, start looking now.

At the time of writing, our messaging volume was just over 283,000 messages. After analyzing the behavior of the thousands of people making their way through this ecommerce journey, we distilled our analysis approach into two types of funnel views.

One funnel view visualized the percentage of people who started and completed the journey. The other view visualized each step a person took in their journey from start to finish. We used the two step flow as an overall health-check. The multi-step flow identified the points of greatest “decay.”

Absent data and outside of correcting broken flows, most messenger bot “fixes” are the result of “gut feelings,” anecdotes from one user’s feedback, input from “friends and family” who aren’t the target or insiders who might just have forgotten what it’s like to “be the audience.”

Using funnel analytics we separated the signal from the noise and focused on what the data told us. Our total conversion view demonstrated health. Our multi-step “journey decay” view indicated where we lost the most people. With that as our roadmap we had focus.

Which leads me to the second lesson in this post, human nature.

First let me say that our root-cause analysis was possible because we track and chart everything — including “no response” alerts from Janis.ai. Using our Janis-based reporting we concluded that people — instead of responding to one of two buttons — they typed in their answers. Crazy? Not really.

According to the Dashbot.io, people tap quick replies and buttons less than 40% of the time. According to the author, Deborah Kay, “..if the success of your bot is dependent [people] clicking on your buttons or quick replies, your bot will fail 60% of the time.”

There is a school of thought in landscape architecture called “desire paths.” An example would be landscapers waiting to see where “organic” footpaths develop across a park before paving a footpath.

That approach also applies to messenger bot authoring — especially in the light that it’s a platform people use for interpersonal communications. How often do you send an invitation to a friend via messenger with “yes’’ and “no” buttons attached?

Buttons and quick-replies make sense from a bot building perspective. However when the numbers show that human nature wins, we must ultimately bend to human behavior — these are your customers after all.

Because we were using Google’s Dialogflow natural language processing engine, we were able to build input opportunities to capture each person’s intent when they didn’t use the buttons and move them forward in the funnel, seamlessly. This eliminated 96% of our daily “no response” events, reducing our drop-off decay to normal levels.

The Takeaway

As we enter 2020 and survey the landscape of messenger marketing one thing is clear — businesses will have less of an opportunity to inundate users with broadcast messages. Each interaction is becoming more valuable. Wasting this opportunity without considering human nature or data would be a mistake

Send messages to which you’d like to respond. Build campaigns behind which you’d be willing to bet your own money.

The offerings of marketers and business who fail to embrace actionable metrics and desire-path thinking — especially natural language processing capabilities will be seen as no more than expensive gimmicks with limited, if any, potential for a return on investment.