Work

Case Studies

Attribution Model

Going from 0 analytics to realtime analytics

2022

Client

Overview

I was hired by Byram Healthcare to analyze and report on all digital marketing campaigns to understand the monetary impact. Byram Healthcare is a healthcare company and all conversions required speaking to a customer service representative to verify insurance information and to place their order. The challenge with transactions requiring customer service was that when a lead submitted their information via a form, or called us after clicking on an online advertisement, there was no method to show the customer service rep where the lead originated from. Thus, when a representative submitted an order for a customer, they didn't attribute or assign credit to a digital marketing campaign leaving us in the dark as to what value was being generated. The goal of the project was twofold, to create a new tracking plan to understand which campaigns were generating leads, and to build a system to merge online and offline data so it could be reported on accurately in realtime.

Part 1: Upgrading Our Tracking

First Step: Tracking Digital Marketing Lead Conversions Through the Forms

Overview

The first challenge we tackled was understanding how many conversions originated from leads who submitted a form after clicking a digital advertisement. We knew we needed to identify the customer and for that identifier to remain with the customer throughout their entire onboarding process. The real question was what is the best way to accomplish this given our time and resources.

Discovery: How to Create a Tracking Framework for Online Leads

At the outset of the project we knew we could send a referral number from each form. The problem with that is if we wanted to create a new campaign we needed a new form created and most likely a new landing page. That presented a problem — what happens if we want to create, update, or change sets of campaigns in a given week, month, or period?

Option 1 — Create a unique landing page with a unique form for each campaign so that you could send a hardcoded referral number through a form field.
Option 2 — Dynamically pass a referral or campaign ID through a URL in the online advertisement directly into the form via cookies.

After discussing with the digital marketing team about the flexibility they needed as far as campaign development was concerned, we forecasted the time we would save tracking fluctuating campaign sets more than made up for the focus and development time needed to set up the dynamic process and chose option 2.

The end result was a profile of a lead in the CRM platform which included a referral number next to the demographic and contact information submitted. The customer service teams would have access to the referral number data in the event a lead confirmed they would like to order supplies and they are eligible for them. This is also true for option 2. The difference between one and two was the method of sending the referral number.

In Comes the Customer Data Platform

At this point we began seriously discussing the need for a customer data platform.

The major problems were thus —

While our existing financial reports could now track leads who submitted a form and turned into a paying customer there were still some leads that we couldn't track which left us blind as to what the total value of our advertising was.

For example, in some cases a patient could submit a form, not convert over the phone, and then go see their doctor and ask for Byram Healthcare to fulfill that order. The patient would not be counted under the total orders attributed to digital advertising. We noticed thousands of leads not converting which left us wondering if these touchpoints were valuable to the business.

After seeing the benefits of realtime tracking of online leads, a few other use cases came to light:

—Large scale top-of-funnel CRM campaigns. Meaning the collected data (phone numbers and emails), if not tied with a converted customer, could be placed automatically in top-of-the-funnel campaigns to discover what stage the customer is at and to encourage the completion of their initial inquiry.
—Predictive models which could use the data we collect on user behavior (form submissions, orders, phone calls, web visits) to predict the health of the user and help plan for tailored follow-up to improve retention and customer satisfaction.

The Groundwork

Before we began, the company had started investing in cloud infrastructure and were interested in using modern tools to see if it could generate alpha. We had access to a modern data pipelining, stitching, and analytics tool which was able to pipe our online web, app, survey, and call center data into a cloud database which hosted our demographic and ordering data.

Rudderstack

Snowflake

Google Ads

The next step was to use the built-in stitching tool to tie together the identifiers from each respective database to generate a universal ID — an umbrella number which was our golden record tying all of our data together.

This was accomplished in a little over a year with a three-person team including myself, a data engineer, and a data scientist. The key to our success was frequent and detailed communication: recording videos explaining tickets, data products, and data validation processes. We had weekly sprint planning meetings with an instant message group chat which hosted daily conversations between the three of us.