Market Your Data Science Like a Product

A 7-Step ‘Go-to-Market’ Plan for Your Next Data Product

Why do internal tools need marketing?

Have you ever developed a great solution that never gets used? Accuracy, statistical significance, model type: none of these matter if your data product is not put into action. Positively impacting your organization as a data scientist means developing high quality data products and successfully launching those data products.

As a product scientist at Indeed (product science is a team in data science ) , I think about launching both business products and internal data products. This has helped me see that marketing techniques for launching goods and services can also be applied to launching data products internally. With this perspective, I’ve helped the tools I developed become among the top 10% most used at Indeed.

I have broken down what I do into seven steps:

  1. Naming/branding
  2. Documentation
  3. Champion identification
  4. Timing
  5. Outreach
  6. Demoing
  7. Tracking

1. Get an MBA name

Your product needs a name that’s MBA: Memorable, Brandable, and Available.

Indeed runs over 500 IPython notebook web applications for internal reporting each day. We’ve developed and deployed over 12,000 IPython notebook web applications. In this rich reporting environment, data products need a way to distinguish themselves from one another. It’s hard to summarize the months you have spent exploring data, developing a model, and validating output into just a few words, but it also can shortchange your work to go with “The model” or “The revenue/ job seeker behavior/ sales thing I have been making!”

Identify your high-quality data products in ways that signal your past and future investment in the work.

Memorable

Apple and Starbucks are two of the most valuable brands in the world. Still, only 20% of people in a study by Signs.com could draw the Apple logo perfectly and only 6% for Starbucks. This points to the power of the name. People do not need to remember exactly how a logo or your data product looks and works, but they need to be able to recall it by name.

Memorable names are often:

Pronounceable. They start with a sharp sound and roll off the tongue. Research on English speakers suggests names with initial plosive consonants (p, t, k) are more memorable, but also see research on word symbolism.

Plain. They frequently repurpose common words (e.g., Apple or Indeed), which help you combine rich mental images to your product. Be aware that discoverability through search may be limited when using common words. Slightly modifying the word can help overcome this (Lyft) as long as it’s memorable.

Produced. They can even be entirely new. Making up a new word is also a strategy (Google, Intel, Sony, or Garmin), but this requires substantially more initial seeding to establish the name. This may not be in line with the audience and timeframe of an internal data product launch.

Brandable

You want your name to consistently represent the identity of the data product and reflect an overall positive attitude towards it. This way it can be incorporated seamlessly into the tool and documentation.

Available

Make sure no one else has called their data product the same thing!

Once you have picked the name, you can dress it up with a logo. The logo can simply be your MBA name that’s been stylized following the same MBA principles. A shortcut like Font Meme Text Generator can quickly create a sufficient design.

For example,

a logo reading "marketing for data science" in a stylized font

2. Document the product

You know what your code does. But what if you’re not around to answer questions, or give a demo when the CEO or a curious new intern ponder to themselves, “What does this thing do?”

Documentation is not only good practice as a data scientist/developer, it is also an opportunity for your work to be found. When one business wants to know if another business has the products and services it needs, 71% start with a simple Google search. Similarly, in addition to being valuable for your user group, wiki documentation and code comments create searchable content that helps your work get discovered.

When writing your documentation, identify:

  • the main problem your data product is solving
  • key features and how they solve the problem
  • key definitions
  • key technical aspects that need to be explained

Documenting your product’s journey can also help build trust in the product. Use consistent messaging by including your MBA name and logo within the documentation to further establish your brand.

3. Identify champions

Who else “gets” the problem you are trying to solve and how the data product delivers a solution?

Seek out people who are affected by that problem, and share your work with them. Also, look to your own team members who have participated in the build or know your work. These champions can recommend your work to others who would also appreciate the solution.

Identifying champions is analogous to customer advocacy in consumer business. Word-of-mouth is a leading influencer across continents and generations for ~83% of consumers (according to a study by Nielsen) when making a purchase decision. Your data product champions will be your top sales reps, lending credibility to the tool and answering questions when you are not around.

4. Timing is everything

Before each launch, consider the current business environment, and time your launch accordingly. The moment you have finished working on your data product is not necessarily the best time to launch it. For example, a product team may be in the middle of fixing a major bug and not ready for a new idea. Conversely, an upcoming related communication activity (e.g., blog post) could be an opportune time for a release with cross promotion.

Look at other recent data products: When were they released and how were they received? Stakeholders can feel inundated with too many new dashboards and models and this may even contribute to “analysis paralysis.”

5. Know your audience

If your champions are not happy, your product can lose its luster in a Snap. Developing positive working relationships with your champions and users is important for the early and long-term success of your data product.

Identify and reach your audience — those who will be using what you’ve made and can benefit from it. With this target audience in mind, comment on tickets, post on Slack, chat, send emails to relevant groups, or go directly to talk to your audience.

Use your audience’s preferred channels to communicate development progress, releases, and feedback. Establishing this communication will build early confidence in your data product. As iteration requests come in, you will have the opportunity to build this confidence with thoughtful acknowledgement of requests.

In 2017, Indeed’s Data Science Platform team — software engineers who built a machine learning deployment framework — went on a roadshow to Indeed’s multiple tech offices to share the data science platform framework. This was a great example of engaging with an audience across offices.

6. Go live!

Only you can see the picture in your mind of how something works. Demoing is a powerful way to communicate what your new data product does. A great way to do this is by getting a minimum viable data product, a prototype, out early to your champions.

Examples include creating a working application with minimal data, sketching a mockup of a dashboard, or taking screenshots. Forbes has more examples of consumer products. As a demo to explain a sales lead qualification machine learning model to the Sales organization, the product science team built a simple interactive web app that returned the model results when a user changed the value of the model features with sliders.

7. Own the results

It’s not that I’m so smart, it’s just that I stay with problems longer.” — Albert Einstein

You may love the theoretical foundation and implementation of your data product, but ultimately the success of a data product comes down to the user. Long term marketing and retaining users depends on how much you can ensure reliability. Reliability is key to building your data product’s brand, your reputation and your technical credibility. This affects the marketing for your other current and future data products as well. It’s worth noting that this doesn’t mean perfection — it often just means dealing with problems quickly, fully and transparently.

Monitor key metrics of your data product to see how it’s working and what its impact is. Actively seek and be responsive to feedback. Evaluate if your data product is achieving its intended objectives and determine if features can be improved to better suit your audience.

If you are not achieving impact or the tool is not being used, revisit your initial assumptions about the problem you thought you were solving. Then, talk to your users (and non-users) about what might not be working. Be willing to destroy and start again, and create something even better with a new perspective. The initiative to iterate and improve your data product tools requires persistence but will raise the quality of your data products and enhance the rest of your marketing efforts.

Final thoughts

Teams outside the analytics community depend on your marketing efforts to learn about your data products that can make them and the company more effective. You don’t have to wait until the product is finished to start letting other teams know about the product. The marketing can start with documentation, champion identification, and outreach as soon as initial requirements are being gathered.

That being said, creating a data product of quality is a priority over marketing for data science, so choose what you market. A data scientist’s credibility is essential for people to trust your data-driven recommendations and act on them. Ensure that you’re investing it wisely.

If you are passionate about both developing great data products and making sure your data products have impact, check out product science and data science at Indeed!