By now, you have probably heard that Machine Learning is the process of predicting future behavior based on past experience, but how can those predictions help to increase your conversion rate?
There are several interesting use cases:
- Increase free to paying conversion rate by prediction which users are likely to convert (we will focus on this use case in the current article).
- Increase conversion rate of a landing page by predicting which version will perform best for each visitor.
- Increase revenue and purchase conversion by predicting which price will be the best for each user (dynamic pricing).
- Increase marketing efficiency by predicting which marketing campaign will be the most effective for different user segments.
- There are many more possible use cases; opportunities are endless…
In this article, we will focus on the first use case - using Machine Learning to convert more free users (trial users, free plan users or just leads) into paying customers.
Machine Learning can predict future behavior based on past experience, but a prediction by itself will not make more people convert.
We must do something based on the prediction.
The prediction should be actionable.
Fit score prediction
The fit score is answering a simple question:
Does the new user or lead look like a paying customer?
If he does (score “A” or “B”), the user has a good potential of becoming a paying customer, and we should do something about it.
If he doesn’t (score “C” or “D”), the user is unlikely to convert, and it probably doesn’t make sense to make any efforts trying to convert this user.
How is the prediction made?
To determine the score, our Machine Learning algorithms use hundreds of external data points such as: location, role, social network profiles, company industry, company size, technologies stack, open positions, news articles and much more.
All those data points are collected automatically by Kilometer.io(based on the lead’s/user’s email) and then used to perform accurate predictions.
A big advantage of Machine Learning algorithms is they are constantly improving as new data becomes available (for example, if a user got a “D” score and eventually converted, the algorithm will learn from the mistake and consider this future predictions).
We have the score, now what?
Let’s take an example based on one of our beta testers:
- He is getting 1000 trial users per month (about 35 per day).
- Historically, out of those trial users, 50 users will convert into paying customers (5% conversion rate).
Using our Machine Learning predictions he increased the conversion rate from 5% to 10%.
How did he do it?
Kilometer.io must “train” Machine Learning algorithms using historical data, so the first step was to upload a CSV file with few hundreds of converted and non-converted users (old users that signed up a while ago and it is already known if that did or didn’t convert).
Then, Kilometer.io can assign a prediction fit-score for every new user.
During the first month Kilometer.io analyzed 1000 users (numbers are rounded for simplicity) and assigned a score to every user; 150 users were marked as “A” score users.
According to past experience, we know that out of the 150 “A” users;
50 are going to convert anyway, even if no special actions are taken.
However, we also have another 100 users who have a very high potential of converting (“A” score users), but statistically we know they will eventually not convert.
What is preventing them from converting?
- Maybe they had a technical issue
- Maybe they didn’t understand the value of the product
- Maybe they need a little discount
- Maybe they have a few questions
Taking actions to help those high potential users convert is the real opportunity to increase conversion rate.
Kilometer.io is not guessing which users are going to convert; instead, it identifies users who have a high potential of converting.
Taking actions based on predictions
Our user decided to use in-app notification to offer a personal demo to “A” users (it would be highly ineffective to make this offer to all the 1,000 trial users, so the offer is made to a small group with very high conversion potential instead).
In addition, Kilometer.io can integrate with CRMs and marketing automation platforms and feed the prediction scores into those tools.
Our user have used the fit-scores in order to trigger webinar invitation emails to “A” users using his marketing automation platform.
“A” users who didn’t respond to the demo/webinar invitations were personally approached by a sales person, either by email or phone.
Becuase of all the actions taken based on the fit-score provided by Kilometer.io the company almost doubled its conversion rate.
Kilometer.io is a Plug & Play solution, which means it doesn’t requires no heavy integration or code changes. Thus, it is an affordable and flexible tool making the Machine Learning technology accessible to companies of all sizes.
Want to give it a try? Signup to our beta.
