In the previous blog, I talked about enhancing Sales Enablement with machine learning. In this blog, I will go into some examples of using machine learning to improve sales operations.

Machine Learning in Sales Operations

Sales operations, despite its name, has more focus on business intelligence and data analytics. Back to the analogy about military operations: sales operations is closer to the intelligence analysts watching satellite images and sensor data to relay progress on a battlefield to generals. In the sales team’s case, the generals would be the SVP of Sales or CFO, who constantly need to adjust sales strategies, reallocate resources and change priorities to meet shareholder expectations.

So what does it take to generate intelligence on the health of sales activities? A well-maintained, clean Customer Relationship Management (CRM) system. Depending on the size of your firm, you may have interacted with products such as Salesforce, Hubspot, SAP, etc. The user experience may be quite different across these products, but the core functionality of recording sales activities is universal.

Ideally, sales reps and marketing folks would follow set-in-stone processes to add and update leads. We all know, however, when things get hectic, this is not going to happen. We’ve found, on average, 23% of leads and 5% of existing customer records are duplicates. Let’s have this sink in for a second.

What this translates to is a 23% over-estimation of the annual revenue growth, or 5% over-estimation of recurring revenue. We have also seen the duplication lead to different sales reps fighting for the same client and cause internal disharmony.

There are a number of solutions out there that aim to merge duplicates. For example, if you search “leads deduplication” on the Salesforce App Exchange, you may notice hundreds of search results. We examined some of the solutions, which provide cohesive integration with Salesforce and see them work very well with a portion of CRM data. If you have a CRM with simple objects (meaning a small number of fields, few exception rules), these solutions work very well for you.

However, what if you have a lot of exceptions and your leads may not have exact name matching?

For example,

Richard T Thomson vs Bill Timothy Thomson vs Bill T. vs R Thomson

When companies acquire leads from a variety of sources, this type of behavior is bound to occur. To understand why this occurs, let’s go one step back in the chain. Having worked with some of the US’s largest leads provider, we had a chance to witness how they work. They typically buy contact data from scrapers and vendors from countries with relatively cheap labor cost. Then, they merge the contact info together by using keywords matching and finally use robot call/email pinging to verify the phone number and email server. The quality of the data is as good as their subcontractors. These subcontractors often produce misspelled texts, which some of them directly copying text and pasting from search engines. This is why you are can see the same individual spelled differently.

Applying Machine Learning in De-duplicating Complex CRM Systems?

Instead of using simple textual matching, we leverage a field of machine learning technologies called “Identity Resolution” a.k.a. “Entity Resolution”. These types of machine learning models predict whether two objects are essentially the same entity, either an individual or an organization. By studying a dataset of linked profiles, the models discover the underlying patterns. For example, in our past work, our model has discovered the profile image, the writing style, location, overlap of social networks all attributed to the linkage.

For our most recent CRM de-duplication work, our machine-learning-powered solutions had effectively removed over 98% of duplicates, which equated to 34% of the client’s CRM filled with millions of leads, even though the names were spelled differently with different contact info.

Our solution can also be called real-time or scheduled. In the recent client’s case, it’s scheduled to de-duplicate the CRM every night. So sales reps always have a clean CRM the next morning. They also don’t need to change any processes as our algorithms automatically took care of it.

Applying Machine Learning in Churn Analysis?

Now that we have clean CRM, the revenue growth forecast is accurate. But what about recurring revenues? That’s where our second solution for sales operations come into play: Churn Analysis.

Good account managers typically would be able to determine with ease which accounts are at risk. But by the time the account managers find out about it, it is often already too late. There are also many other signals that could help sales operations forecast the churn far ahead of time. For example, whether the client has job openings that are competing with the offering; the client recently raised a round of funding, providing the resources to acquire more expensive, higher-quality offerings, etc. There are many channels where the pieces of information can be acquired and put together. Nevertheless, it is extremely time-consuming to monitor several hundred information channels and consolidate the data together while serving your portfolio of customers at the same time.

The key to accurately predict churn is taking into account as much information as possible. Let me give you an example of one of the solutions we built for a client. Our bots are set up to monitor over 30 indicators of the clients’ customers including:

Are customers growing in terms of the number of employees?
Did the customers’ websites change in search engine ranking?
Are there positive/negative news about the clients in the news in the last 5 business days?
Has there been news about the customers’ fund-raising?
Have the customers posted X and Y job openings in the last 5 business days?
How often the customers are using the products?
What’s the sentiment of the customer communication?
Is the negative/positive sentiment of the communication progressively better or worse?
…..

By studying the historical data, our machine learning model is trained and can predict with high precision on when customers will churn in a few months of time. The prediction has several benefits to this particular client of ours:

It helps them adjust the forecast better by taking into account of “expected value” of recurring revenue
It also helps account managers reach out to customers earlier to either stop the churn in time or relay information back to the product engineering team to develop features that can meet the demands of clients.

We have not seen an effective one-size-fits-all solution due to the nature of different sales methods and sales processes. That’s why we witness custom-built solutions producing much better results than $79.99/month subscription-based analytical tools.

I highly encourage you to think about your sales process and sales objectives, then discover what would be an appropriate and effective solution for it. I am always happy to learn from you what has worked and what hasn’t. More importantly, I’d be more than happy to provide any insight I have from what we have learned from our clients to help you greatly improve your sales enablement and sales operations. On that note, please feel free to drop me a message at jia[at sign ]softmaxdata.com .

Good luck and I look forward to hearing from you!