The key benefit of predictive deal scoring over traditional deal scoring is that it brings objectivity to your data. All predictions are based on machine learning algorithms. Obviously, you need data to get started. The good news is that you probably already have it in place.
In this blog post, we address the top 5 questions we get asked about predictive deal scoring.
- What kind of data can I use for predictive deal scoring?
- Do I have enough sales data?
- Which data points have an impact on the deal score?
- How many data variables should I include?
- My data includes a lot of inaccuracies, what can I do?
What kind of data can I use for Predictive Deal Scoring?
Our Predictive Deal Scoring model can analyze any textual and numerical data that is in a structured form. An example of structured data is an Excel sheet in which information is organized in columns and rows. Examples of unstructured data that our predictive scoring model finds challenging to analyze are PDF documents and photos.
The model can analyze data from almost any database. This way all relevant marketing and sales data from your CRM system, Excel, and other data sources can be included in the analysis.
Do I have enough sales data?
The more data, the more accurate predictions we can achieve. However, we often advise our customers to start small and then grow the model. Our predictive scoring model is self-learning and automatically includes all new data points in the analysis. This way there is a positive spiral of time. As data accumulates, the prediction accuracy increases.
As a starting point, an optimal number of sales cases is more than 100 sales cases with an outcome. The outcome does not have to be binary, such as a lost or won sales case, and can also state that the deal was “postponed” or the customer “bought two products out of five”.
Even 100 sales cases can result in good predictions and around 90% accuracy.
In an ideal situation, your data includes information about
- Pipeline turnover. How fast do your sales cases move from one pipeline stage to another? How long sales cases stay in the most critical pipeline stages?
- Engagement. How many times have the contacts associated with a deal shown interest in your company and products? Activities such as a high number of meetings, opened emails, and sales calls impact the deal scores.
- Company information. Company information, such as company size, industry, company form, and financial metrics like Quick Ratio and Ebitda are used to predict the likelihood of closing.
- Purchase history. Has your company made deal with a certain company before? How much time has passed since the deal was made? Is the deal value higher or lower than last time? These factors have an impact on the deal score.
- Deal source. Are your deals based on social media activities, advertising, or were the contacts manually added to your CRM system?
Based on previous cases, these data points have shown the most significant impact on the deal scores.
Even though the more data the better, often the basic information companies have in their CRM systems is enough. To ensure the suitability of our model for your business, we offer new customers a free test period that states in black and white what level of accuracy we can achieve with your data.
What if I only have information about the deal value and some sales activities?
That is okay. We will use a company data provider and open data sources to get information about the companies in your sales pipeline. To do this, we only need a small clue, such as a list of company names.
Enriching your data is valuable only if you don’t have any information about the companies in your pipeline. If you have information like company names, their industries, and revenue, enrichment is usually not needed.
How many data variables should I include?
This is something you should not worry about. Our data scientists will help you to define which data sets are relevant in relation to Predictive Deal Scoring to achieve the best possible prediction accuracy. We also make sure that the data has good granularity for the scoring to function effortlessly.
Usually, the data we receive from our customers includes hundreds of variables. From here, our data scientists select the most relevant combination from the perspective of Predictive Deal Scoring and the company’s business model and field.
The number of variables used in the model depends on the data and selected algorithm. In some cases, only five variables will do the trick. More often, 30 to 35 variables are included in the predictive scoring model.
My sales team documents information inconsistently, and there is a lot of missing data. What can I do?
This is a common problem. Our data scientists always strive to use all available information and to make sure the data used for Predictive Deal Scoring is clean and consistent. In other words, that there is no missing data, and all information is in the same format.
Sales data derived from CRM systems often include inaccuracies. One salesperson can mark ten thousand as 10M and another as 10 000. Sales personnel can also use different languages and punctuations, such as a dot instead of a comma.
We’ve got this covered. We use a machine learning-based data cleaning process. This model fulfills missing information and corrects outliers. This way we make automated corrections to inconsistent data and ensure our customers will reach the best possible prediction accuracy.
Your sales team will get more motivated to document information congruently when they experience the benefits of Predictive Deal Scoring. This is also a perfect time to set up common rules about what information is necessary for your sales team to document. You can also estimate if some parts of the documentation could be automated or even be completely left out.
Ready to get started?