Alert! Not all sales machine learning models are created equally

Machine Learning — the buzziest of buzzwords in the sales world. But it’s not all hype; there’s a reason so many companies are offering machine learning (ML), natural language processing (NLP) or artificial intelligence (AI) solutions to augment sales efficiency and effectiveness. Turns out, using a model to leverage the mounds of data you are already collecting works and, when done correctly, can help your team rise above the competition. But there’s one caveat.

You need to know why it works.

Transparency is incredibly important. And that’s where the most significant difference between offerings lies. Many companies obfuscate the inner workings of their models, never letting you see what’s actually happening when you pump in your data. It’s kind of like a microwave: you throw in your leftovers (raw data) and out comes a hot meal (useful learnings). You’re fairly certain something is happening inside but it’s not readily apparent.

If your machine learning solution is a black box, how can you trust the insights it produces? If the how is a mystery, then what can you point to or change if the model fails to surface anything meaningful.

To safeguard you from purchasing what one notable data scientist referred to as “snake oil,” here are the crucial capabilities your machine learning offering needs to have.

What’s in the box?

As mentioned above, you need to know how the technology works, i.e. you should be able to understand how the model got the results it is presenting. If you dump all of your historical sales data into the model and out comes the insight that your best customers are mid-market companies who use marketing automation, you should be able to see how the model arrived at that finding. I get it — the inner workings of most ML offerings are closely held secrets for many reasons, but why should you trust your livelihood with something if you don’t know how it got to a certain conclusion.

Furthermore, if you can’t get a good look at the model, how do you know what levers to pull to achieve a desired outcome? Short answer: you can’t.

Look for “clear boxes” when researching ML tools. Solutions that “show their work” allow you to not only make better decisions, but be more confident in those decisions. You have more control, more visibility and you can safeguard against seemingly oddball correlations. With enough data, you tie anything together — that doesn’t make it right or useful. Being able to sanity check insights and predictions is key.

Where’s the lift?

So you’ve fed the model a ton of data on your sales efforts, including information on prospects, customers, and accounts that have churned. Lo and behold — some patterns have surfaced!

Now what?

Good question. Does your ML solution prescribe any action? If so, is there measurable lift attributed to that action? While I’m sure with some effort you can connect the dots for yourself, but you shouldn’t have to. A good ML tool will not only crunch the numbers and surface patterns, it will prescribe action that will deliver lift for your business.

The reliability of those prescriptions will hinge on the quality of data you put into the model, but they should be there nonetheless and they need to actually improve how your business operates.

How hard is it to implement?

Most companies turn to a machine learning provider because they don’t have the resources to hire a data scientist, let alone a team of them. That means someone on the sales team — whether that’s sales ops or the Vice President — will be tasked with implementing the solution.

If you need a developer or engineer to implement the model, consider it a red flag. Ultimately, the patterns and recommendations the model produces should be immediately actionable and, frankly, understandable for the average salesperson. Usability is important, so look for a solution that prioritizes accessibility.

Bottomline, find something that you can trust to provide measurable lift to your company. There are tons of data scientists out there making ML models, but there’s more to it than shoving data in one end and watching patterns come out the other. Insight without recommendations just trades a bunch of data for slightly less data. The goal is to make better, data-driven decisions that will positively affect your business; to mine your existing reservoir of information to find new opportunities. And the best machine learning solutions do that by providing transparency, lift, and ease of use.