Prioritizing Sales Outreach with Account Scoring
The Problem
Sales outbound is one of the key revenue generators at Brex. Our sales reps communicate with hundreds of prospects every day to convert them to actual Brex users.
Today, there are over 1M prospects in our account pool. The size of the account pool keeps increasing as we continue to enrich it by leveraging various data sources. Without a single indicator of customer value, sales reps may spend a considerable amount of time searching through the account pool rather than outreaching prospects, and they may still struggle to identify the highest potential customers.
To help the sales team improve the outreach process, we developed a multi-faceted account-level scoring system to prioritize prospects based on a holistic view of their attributes.
What is a “Good” Customer?
We determine account quality based on 2 main approaches.
(1) The Ideal Customer Profile Propensity Model
We target customers with a high lifetime value that have grown with Brex and require less time to close the sales process. We begin our account scoring approach by assigning a propensity score based on how well a prospect fits our Ideal Customer Profile (ICP).
We use a data-driven method to identify the characteristics shared by the best customers and a sales insights-based approach based on interviews with the sales team. Model features are derived from firmographic information (e.g., company size, industry, business model), and signals on business growth (e.g., having received external funding and website popularity changes). Features that indicate preferable characteristics are assigned higher weights.
(2) The Revenue Model
Business revenue is a direct measure of the total amount a customer will spend with Brex. Therefore, in addition to the propensity model, we introduce the revenue model.
Revenue estimation is a challenging task because one method that works for certain types of businesses might not work for others. To address this problem, we gradually roll out different versions of revenue scoring specified for different business models. For instance, businesses that heavily rely on online channels such as ecommerce are estimated based on online traffic, conversion rate, and SKU price. For medium to large and well-established companies, the revenue scoring approach is based on more conventional data sources such as credit history and public records.
To further improve the model accuracy, we incorporate firmographic information and calibrate the estimated revenue to a probability that indicates how likely the customer’s revenue is in our preferred range using classification models.
By leveraging firmographic information, we can detect anomalies in the revenue data that erroneously skew our conclusions. For instance, if a company has an employee count of 1 but has an estimated monthly revenue of $1M, it is more likely that the revenue estimation is inaccurate. The revenue to probability conversion enables us to assign a lower score to the accounts with anomalies.
Balancing the Team’s Efforts
The Sales Engagement Model
On top of prioritizing accounts using the above approaches, we also want to equally distribute the sales team’s efforts across “good” accounts, so as to avoid duplicate work on the same customer or underserving accounts who deserve more attention. To address this, we developed the sales engagement model. This model outputs an engagement score indicating the level of prospecting effort made by the sales team.
The engagement score is calculated by aggregating sales engagement history and the recency of sales activity. All sales activities, including meetings, phone calls, emails etc., are captured in real time through the integration with the CRM platform. The engagement score is updated every few hours to reflect the latest activities. A higher score indicates a higher level of previous involvement. Sales reps are encouraged to prioritize “good” accounts with low engagement scores.
As the next step, we are working on researching the engagement needed to convert sales leads to qualified sales opportunities.
Conclusion
We use both data-driven and sales insights-driven approaches to create a sales account scoring system that optimizes the sales outreach strategy and ensures that the team’s efforts are spent on high-value customers.
As Brex grows and the market environment shifts, the model components, methodologies, and ICP approach may evolve over time. We collaborate closely with the GTM organization to continuously monitor model performance and iterate models to reach perfection.
Special thanks to Seung Ham, Jay Chua and Greg Keiser for your valuable feedback on this blog post.
