How Faraday’s predictions saved a national home services brand $41,000 per month on affiliate leads

This customer story has been anonymized to respect the company's competitive strategy.

The problem: predicting lead quality at point of purchase

This national home services brand relies heavily on affiliates to acquire new customers. However, they encountered a recurring issue: a subset of the leads they purchased were actually unqualified. In other words, they were wasting money on the wrong leads—and missing opportunities to connect with the right ones.

The team tried several methods to improve lead quality. They put a process in place to avoid the purchase of duplicate leads. They instituted an identity verification layer that verified a lead's contact info prior to purchase. Nonetheless, the problem with purchasing unqualified leads persisted across all of their affiliate partners.

Their team knew that they needed to generate additional data points to enable effective real time decisioning at the point of lead purchase. The challenge was what data points would be available AND most effective at this point of purchase. They saw 3rd party data and AI as a potential tool to enable this type of decisioning, but did have the in-house resourcing to do this, nor the budget.

The obstacle: building AI into existing processes

For a team with no dedicated data science resources, building predictive models themselves would be a nearly impossible task. To do so would require overcoming several hurdles:

Licensing consumer data

The first step would be getting the right data. AI models benefit from having a lot of data to train on. The marketing team knew it had plenty of leads and customers, but it lacked data about these leads and customers that was correlated with lead quality. One way to broaden its training set would be to license 3rd party data like household income and property details to help with model training, but they knew it would also need this same data for the leads they were considering purchasing, if they want to use this data at the point of lead purchase.  Licensing this much data, let alone using it to create a predictive model, could easily run into the six figures per year.

Building a data science team

Even with this data in hand, the company would have needed to make a significant investment in a data science practice. Traditionally, hiring the team to create, iterate and generate accurate predictions requires an estimated 18-month timeline. Not only would the company need to build a team from the ground up, but they would also need to maintain those resources to constantly iterate and refine their models, which would easily result in a high six to seven figure investment.

Integrating predictions into existing systems

A notorious issue in data science is that the vast majority of projects fail. One of the key reasons for this is failing to build a scalable solution that works within existing workflows. The team used LeadConduit in its lead purchase workflow. Any data point used in lead purchase decisioning had to be integrated into LeadConduit in order to be effective.

Why Faraday?

Built-in consumer data for predicting customer behavior

Employing the Faraday identity graph (FIG),  with its 1500+ attributes on nearly every U.S. adult, the affiliate team instantly had enough data to create models on their leads and customers. With only an email address, the affiliate team was able to match contacts to their property, lifestyle, demographic, and psychographic information within Faraday.

These data points gave a much deeper picture than the 1st-party data they had been offering previously and provided the deep well of data necessary for predictive models to train and spot patterns associated with leads that converted to sales and those that did not.

Easy integrations means easy deployment

The IT team was able to use Faraday's API to receive lead scores in Lead Conduit in real time. This allowed lead purchase decisioning to be made in real time, based on the probability of the lead converting to a sale.

With Faraday, the affiliate team now scores contacts the moment they are presented for purchase ("pinged") and they use that score to determine if they will bid to purchase the lead ("post"). Any leads in the bottom percentile of propensity scores are instantly rejected, and the remaining scores are then pushed into HubSpot for use by their sales team.

Testing and iteration with turn-key prediction

This home services organization was able to skip the typical years-long creation of an AI program and, using Faraday, add predictions to their stack in two weeks.

Rather than risking a heavy upfront investment into building an untested prediction program, the affiliate team could perform tests on holdout groups of their data over a test period to get a sense of whether they could find a predictive signal on the leads they were buying.

The results

Instant ROI

Immediately, the organization saw the benefits of applying a predictive layer to their lead purchasing program. Starting with rejecting the bottom 2% in the first two weeks, the program had grown to rejecting 9% three weeks later.

Over a 30-day period, Faraday helped this brand save $41,976 by rejecting 1,072 leads who were predicted to never convert.

Areas of expansion

Predicting lead behavior was such a roaring success that the marketing team has begun rolling out predictions to other departments.

  • The affiliate team is using Faraday to predict the optimal number of touchpoints before abandoning follow-up on a lead.
  • The email team is using predictions to reduce spam complaints and improve deliverability.
  • The marketing team is using Faraday to understand emerging markets and inform their direct mail strategy.