HeadsUp’s Machine Learning engine allows you to define any conversion objective you want to drive. Common campaign types created by customers include:
- Free Trial to Paid Subscription
- Starter Tier to Premium Tier
- Self-Serve to Enterprise
- Cross-sell from Product A to A+B
How it works
Once you define the start and end milestones and connect your various data sources, HeadsUp automatically generates hundreds of combinations of different usage metric types and applies time series transformations.
For example, if one of your key usage events was a user making an API call, our model generates a range of metrics such as total number of API calls in the last 30, 60 and 90 days, week-over-week change in API calls, average API calls over the past 8 weeks, etc.
Our ML algorithms then examine all past conversions from the start to end state you’ve defined to identify which usage metrics are the most reliable indicators of conversion. These predictors are applied to live leads to score accounts (or users) for probability of moving to your desired objectives.
Unlike linear regression models used by some analytics teams, our ML approach has been engineered to handle a large number of different variables, control for outliers, and work even with a moderate number of data points.
How accurate are your models?
With our early customers, we’ve seen significant improvements in lead quality and conversion rates after implementing HeadsUp, within as little as 4 weeks.
For example with Contentful, HeadsUp’s ML models successfully eliminated 50% of false positives (”bad” leads which would never convert), while maintaining 99% of high potential opportunities, resulting in a 100% improvement in conversion rate.
In a different case, our ML models surfaced 44% more high potential leads than the customer’s sales team previously were receiving from an internal rule-based lead qualification model.
Perhaps most importantly, as part of our implementation process, we backtest models against your historical data to determine their accuracy before running it on live leads. That way, you can be fully assured of the lift before you start sending PQLs from HeadsUp into your marketing or sales workflow.
Impact you can expect
The superior accuracy of PQLs translates directly into a step-change in revenue outcomes.
For instance, for a sales rep with a 1m ARR annual quota, an increase in conversion rate of 40% will translate to 400k more per year. This gain comes from reps saving time previous wasted on low quality leads, and getting more high-intent leads that close more consistently.
Multiply these figures by the size of your sales team and the ROI from investing in an accurate Machine Learning PQL model will add up easily.
What you need for this to work well
For HeadsUp to effectively help you identify PQLs or other high potential opportunities, we need the following kinds of customer data:
- Product usage data
- Firmographic and demographic data (we buy 3rd party data and can augment your leads if you don’t already have this)
- Marketing engagement data (website visits, event sign-ups, etc.)
- Billing data
- Conversion data (historical closed/won or lost, typically stored in your CRM)
The most important factor influencing the efficacy of our models is your number of historical conversions (for each goal you’ve defined).
- 100 historical conversions will allow a highly accurate model.
- 30-100 historical conversions will allow us to build a viable predictive model, although iterations may be needed to improve accuracy over time.
- 10-30 historical conversions will allow us to produce simple heuristics, for example, “what are the 5 behaviors that tend to lead to conversion?”