Improving sales performance with Razor Sprint
Here is the scene: You have a sales team. You have customers. Some are new and the majority existing. Some of the sales team get more orders than others. Some lose money on some orders, some make a bit more than you expected.
This range is then amplified by growth. It is a manual or gut-based process, initially, by a few very skilled people. When scaled out to a diverse cohort, they all have very different levels of experience, insight and knowledge. As the business scales, the successful aspects get amplified, and so do the not so successful.
The challenge; can technology help improve the overall sales team performance, generate more sales and continue to be sensibly profitable? Can we use historical data to provide the sales team with a suggested sale price that keeps the business profitable and wins business in a consistent way?
With a clear understanding of the problem, the next step was to explore the data that the business had. ERP and CRM systems are a great source of information, especially when configured to capture the relevant information in a structured way. This is exactly where we started to look for useful data.
The challenge here, above the cleanliness and understanding of the data, was the scale. Big data to some, isn’t what we would call big data. Millions of rows might sound a lot, but to create an accurate machine learning model, it isn’t.
The problem wasn’t so much the scale of the data, but how the data was split up and categorised. Customers and products are all different and the pricing prediction has to have an understanding of this, causing each prediction to have a much smaller subset of data to learn from.
With a limited data constraint, a machine learning model was created. The limitations forced the data engineers to be creative with the post-process to ensure that the predictions were valid and within tolerances.
The next piece in the puzzle was how to get this tool into the sales process. Can we inject it into existing tools like the CRM? Think of this approach like the assisted sentence filling tools that are in Gmail, or pre-populating input boxes that can be overwritten. The key to success here is not to force people to change or to replace them, it is to enhance them.
To introduce the new prediction tool, it was initially delivered via a standalone simple and engaging interface. The salesperson has the power to use it at their discretion. They can use it to generate a suggested sale price and either accept it or alter it and feedback into the system.
Over time, the model is then updated using what it has learned, knowing when it was right or wrong. It also knew when the salesperson was right or wrong, as it knew from the ERP system which quotes were converted into orders and which ones weren’t.
Adoption and trust of anything like this is key to its success. Forcing something onto people or making a big change without people accepting it can cause an initiative to fail. Starting out small and making it optional to use gave the solution the opportunity to learn. As a result, it gained trust and was then rolled out as an integrated tool within existing systems. This is the way to achieve long term adoption and success.
This is a perfect example of finding an area of marginal gains that extrapolate into big wins, which, over time, unlock the next level of growth and profitability.