What's your “data story”?
Presenting Data and Machine Learning product concepts to stakeholders and management
Everyone loves data. Everyone loves talking like they understand data. Everyone loves using ML and AI buzzwords — but many times they are just talking, hoping they are using them in the correct context, but still usually just throwing them out there into the open. Which of course is also important. That’s how you sell your product. That’s how one shows that their company is the top in their field, that they have high-end abilities, and of course, that they are the best at what they do.
What I have learned along the way, is that really learning to share and explain a “data story” is hard work.
As part of my previous position as Data and ML Product manager, one of my tasks was a quarterly presentation to the global chairman, who was also our main investor. These presentations were particularly difficult for all PMs in the company to build, since we were jumping from high level concepts and quarterly planned roadmaps, all the way down to specific features that would be delivered in the upcoming weeks. On one hand we would try to explain our day-to-day framework, and on the other hand, present this coherently, explaining how it fits into the bigger picture and the company’s business strategy. Even though presentation repeated itself regularly, I believe the insights learned from preparing these, can be relevant for any of us when presenting to investors and stakeholders, where they do not know always know the technology, but have a generalized understanding of the product.
So in addition to these complexities, as the only Data ML PM in the company, I felt my challenge presenting was even bigger.
And why is that?
Good data features are invisible. They are trivial to an outsider and a user. The only time a user will actually see an algorithm is when something is wrong, when something is off and the user is receiving incorrect output. But the challenge is not only that. The data world does not excel in explaining itself with nice user-facing UI, which is how stakeholders best understand what you are referring to. There is lots and lots of text, of code, and of course loads of charts and diagrams. In a B2C world, translating this information into something that can explain the bigger picture and also show the visual impact, is quite challenging. How can you show what you’re working on, when you’re presenting to a group that sees only the tip of the iceberg?
Here’s how I chose to tackle the problem.
One of the first things I did was a bit non-standard for our presentations. Instead of jumping right in, I chose to add extended explanations in one of my first presentations. The idea was to detail the process and basics of Machine Learning, and how our team implements these algorithms.
For subsequent presentations, I would start with a bit of an overview, going through this same process and explaining which area(s) we would be focusing on during the given presentation.
I think this concept is even relevant for one-time presentations, building a framework that makes the whole presentation more coherent and understandable.
This is one of the posts I found relevant and helped me build the framework and overview to guide my presentations: https://hackernoon.com/the-ai-hierarchy-of-needs-18f111fcc007
The feedback I received regarding these few guidance slides (i.e., the presentation framework) was amazingly positive. People finally were able to understand the structure of our data, the abilities we have as part of our platform, and last but not least, the process of building new abilities — how we define hypotheses for new features and algorithms, and then move towards prioritizing, testing and implementing.
In the long run, I found that not only did these guidance slides help me for these presentations to the chairman, but also for others requiring explanation of our teams work.
Aside from these guidance slides, I built new presentations using the following additional guidelines, which helped me focus on the task at hand:
1- Don’t fall in love with your mathematical design
Data is usually shown in flowcharts, schemes, tables, and such, but there is always (at least, should always be) a feature waiting down the line to use this new ability. That is what we want to demonstrate. What gives business value? What brings value to our customers? In a B2C world we show visual UI components. This of course might not be the only algorithm populating the UI component/feature. But when showing it in-context, it can be explained as part of the same flow, explaining why each part of an initial diagram is needed.
2 — Explain from the user’s perspective
Remember that each and every data table or logic is supposed to enable a feature, which in turn enables the user to convert better. Not always is this understandable when explaining the back-end algorithm of a feature. However, if we focus on explaining the user impact, this can be the key to building specific slides — or even the entire presentation, gaining our stakeholders’ understanding.
3 — Using visual diagrams (and as little text as possible)
There is no escape from explaining the new amazing ML algorithm you are working on and how the system works — but even these should be presented in a visual fashion. Remember, these slides are the ones which the stakeholders are probably going to show off to all their friends, bragging that they have the “better AI” — so they have to have the right buzzwords, but not be too complex. You don’t want to lose them while they feel they have to read every word, on every box, in every diagram.
4 — Data, Data, Data
We’ve explained the flow and showed it in visual diagrams and flowcharts. But since dealing with data is a big part of our work, it should also be a part of our presentation. Availability of quality data is a key component of bringing value. That’s why the “cost” and investment in these areas are large, but nonetheless crucial. Unfortunately, sometimes this is not obvious to everyone, with such presentations representing missed opportunities in getting everyone on board.
To summarize, until ML becomes the-way-of-the-world (and that day will come :) ), until it’s obvious that we are as atomized and optimized as can be, describing data, ML and AI elements will remain complex. Therefore, Data ML PMs remain critical for the bigger picture. They supply the ability to fully understand and analyze the users’ needs, define lower level data features and algorithms, and “translate” and simplify, in order to get the whole team, on the same page.
In this Mozzarella podcast you can hear me speak a bit about this topic and the Data & ML Product Manager position in general (recorded in hebrew) — https://soundcloud.com/productpodcast/51-data-and-machine-learning-product-management-featuring-sarit-naiman?fbclid=IwAR0MJeGSAYcmecLPtt9Ngkd5cSTnoMGVWkKX-__l2gT6OJF97dmvVncK7_k
Originally published at towardsdatascience.com on February 25, 2019.
