Reading Time: 4 minutes

I love this question. It’s central to all the early conversations I have with would-be clients. And perhaps surprisingly, my answer isn’t always, “of course!”

In this post, I’ll cover:

  1. What lift does data science generally provide?
  2. How do you know when you’re ready for data science?
  3. If you’re not ready, what sorts of “smart things” can you do to achieve product gains without machine learning?

What lift does data science generally provide?

This one is too broad as it stands. Generally, if you have a well-defined product and collecting data relevant to your main KPI’s, doing something smart with the data can show a tens of percent lift. But if your product isn’t as well defined, sometimes a small fundamental update to the product will create a huge lift without data science. Then layering on predictive models can really increase profits.

For example, if you’re selling products and you don’t offer an up-sell during checkout, just adding the up-sell can provide a big lift in sales. But what do you provide there? The most popular item that isn’t already in the cart? That’s a good start, but it may not always be relevant, nor will it generally be personalized to the user. Bringing in machine learning and using something simple like the FP growth algorithm to provide a smarter solution will generally add a small tens of percent to the already large lift to revenue.

Data science works in addition to the great products you already have, it’s not usually a product in and of itself. If I had a dollar for every time I was approached and asked to “make our products 10x better using all that data you guys have”!!

How do you know when you’re ready for data science?

This one comes back to how I like to define what Bennett Data Science does.”Companies build products, and in the process generate tons of data. We take a product-first approach at using that data to help companies generate more revenue.”

That’s fairly high level, and there’s a lot more to it. You might be ready for data science at these stages:

  1. You’re early in your company development, and you know you’re going to need data science, but you still don’t have a product. In that case, it’s important to have an initial consultation with a data scientist. Here’s why: when you explain your product or business to a data scientist, they’ll be able to make sure you’re setting yourself up for success later on by answering these sorts of questions:
    1. Are you collecting the right data? – you can’t predict car color from shoe size. Make sure you have the right data!
    2. How long should I collect data before I can leverage AI or machine learning? – there are rules of thumb here for most use cases.
    3. What’s the lead time between an initial data science engagement and working production predictive models? – a lot goes into these answers, such as reach, type of product, and how much dev support your team intends on providing
  2. You’re still considered a startup, but you have products and have been collecting data for a while but don’t use machine learning and know you’re leaving revenue on the table. There are generally big wins to be had through simple constructs. More on this below.
  3. You’re a mature company and either not leveraging data science yet or have a small or underperforming team. These cases can be more challenging and generally take longer to incorporate working solutions. Established teams might be hesitant to adopt new tech, and current teams can be threatened by outsiders. We’ve worked in these situations several times, and the results were transformational in all cases. I’ll leave the company names out, but it’s not just small companies that have these types of issues!

If you’re not ready for data science:

This can be a fun place to find yourself. Let’s stick with e-commerce companies for a moment. Let’s say you have 1,000 products and 100,000 customers (at that point, you really should be using machine learning!) and you want to build a simple product recommender. Here are a few smart things you can do:

This can be a fun place to find yourself. Let’s stick with e-commerce companies for a moment. Let’s say you have 1,000 products and 100,000 customers (at that point, you really should be using machine learning!) and you want to build a simple product recommender. Here are a few smart things you can do:

  1. Count the number of each product sold and recommend the most popular products to your audience. You might be surprised just how good this is.
  2. Want to make it better? Try limiting the counts to only the past three months or weeks or even days, if you have enough purchase data. That will remove some of the seasonality effects you’ll encounter when you, for example, look back to what was popular during spring and it’s already summer.
  3. Segment your users. How about counting most popular product by segment? Say you have US and European customers. Try building two lists of popular items and showing them to the respective lists. How about segmenting by age?

We call these heuristic models and they can be very very performant! Here are some of the huge benefits of heuristic models:

  1. The force you to build a data pipeline to train the model, then surface it to your users. When you have a strong pipeline in place, it can be trivial to swap out the predictive model for something smarter. The value of a good pipeline cannot be overstated!
  2. These “models” are simple to understand, track and troubleshoot. There’s no complicated math involved.
  3. The performance of such heuristics is generally very good, especially when constructed with the help of a domain expert who understands how customers tend to behave.

In most cases, I certainly recommend contacting us for advice before hiring your own data scientist(s). Here’s an Inc article that dives into some of the details. Data science teams are expensive to hire, tricky to manage and require a lot more infrastructure than most companies expect.

Well, is data science worth it for your team?

  • Are you working on a killer product, collecting a wealth of data and pushing to be more personalized with your offerings?
  • Do you believe in, or have you started, building rock-solid pipelines and started using heuristics to better serve your customers?
  • Are you sure you are collecting the right data for your future goals? Do you have products or a team that is underperforming?

Zank Bennett is CEO of Bennett Data Science, a group that works with companies from early-stage startups to the Fortune 500. BDS specializes in working with large volumes of data to solve complex business problems, finding novel ways for companies to grow their products and revenue using data, and maximizing the effectiveness of existing data science personnel. bennettdatascience.com