Let’s examine how to convert lost deals into sales using Machine Learning. CRMs (Salesforce, Microsoft Dynamics, Hubspot or any from top 10) today provide the flexibility to configure your sales pipeline with multiple active stages and multiple closing stages (WON, LOST, UnQUALIFIED) to capture the outcomes of deals in your pipeline. When you win a deal, it feels great all around, but when you lose the deal not many realize that it is valuable data that can tell you a lot more about your sales team, process and something about the product.
Philosophically, this only makes sense – “learn from your failures”, but lets take a practical data-driven view of the same. Lets look at a typical sales funnel below, with some active stages and closing stages. As seen from the table view of the pipeline, for every 1000 deals that enter this pipeline we close about 50. This means, we have only 50 data points to learn from. However, if you look at the remaining buckets together, you have the 950 data points (1000 – 50) that couldn’t make it to the finish line. Statistically speaking, insights that you glean from lost data probably have a higher chance of being significant, when compared to insights from won deals. Think of that again !
All of this starts with configuring your CRM with the right closing stages and let me add an additional closing stage (DROP OFF, which we shall discuss below) and the right tools and workflows to capture the lost reason. Unless your CRM is configured to capture this knowledge, it typically sits in the heads of the reps or at best as unstructured data (notes, emails, text field, texts etc) and almost always never looked at again.
Understanding the four buckets
Let us take a closer look at each of the buckets and what we can learn from them.
Won: When you win a deal, a whole lot of things went in its favor for the deal to have gone through the finish line. Most good CRMs out there today, give you a comprehensive set of reports that tell you more about the sale cycles, total sale values, rep performances and other ratios that tell you efficiencies in your sales process.
DropOffs/NoResponse: It is important to separate out the unqualified deals from the not responded ones (either in lead stage or in any specific stage of the pipeline). The non-responders or drop-offs could pollute the data and throw stage conversion ratios off, clouding your analysis of the sales funnel. Also, as any good sales person would say – no response only means, try again later !
Unqualified: The unqualified deals should typically go into a separate bucket. These are deals where the sales reps got a sales qualified lead (SAL) and worked on it for a while before deciding it does not qualify to move forward. More often than not, it is due to a product-market mismatch and it is important to document and circle the feedback to both marketing for better MQL process and to engineering/product for better roadmap prioritization.
Lost: When a deal is lost, it is important to understand why a prospect chose to look in the other direction after engaging and investing their time. Is it the features, pricing, discount, competition, bad closing process, constraints on their end, expectation mis-match? All lost deals would go into this bucket and the representative should tag the deal with an appropriate and objective reason for the loss.
[Tweet “True cost of a lost deal = value of the lost deal + time of the rep + company resources “]
Similar analysis can be done for lead qualification funnel and that can tell you a lot along the lines above, with a focus on marketing effectiveness and sales alignment in the company.
Challenges in analysis of lost deals
- Sales Process : Does your sales process as implemented by your CRM supports a wide variety of hierarchical loss reasons and the rep workflows are built to trigger the data creation?
- Data Availability: Can you trust the sales reps to enter the data at the right point into the system, after analyzing the correct outcome and loss reasons?
- Data Quality: Are the sales reps sufficiently trained to understand the various loss-reasons in your CRM and their applicability to different scenarios?
Algorithms to the rescue
At Predera, we have built algorithms to work with low-quality sales data by normalization and projection in a statistically sound manner. We have also built a suite of text processing libraries which in the absence of human entered data, can exploit unstructured data sources like email, notes, description fields associated with the deal to extract the loss reasons.
If you wish to learn how machine learning can help sales and need help with your sales analytics and find ways to optimize your sales process? – drop us a note at firstname.lastname@example.org