Office Hours
Unsheathe your Slides to Kill a Dragon and Save the Business
Five tips to sharpen your storytelling and demonstrate the best value of your data science experiments.
I know… Slides. Most data scientists hate building them. It is frustrating because you spend loads of time designing them and, often, the results are not as good as you expect. But don't be sad, I am here to help you make your slide-building funnier.
As you may have already noticed, slides are the most popular tool to exchange knowledge within companies (big or small ones). Without them, few people pay attention to what you have to say. Presentations are so popular because human-beings are visual learners. A 2014 study from MIT measured that our brain can process images in about 13 milliseconds. That's immensely faster than reading even a tiny excerpt of text.
If you want to bring attention to your data science results, you should master your Powerpoint fu; or your Google slides fu.
To make slide-building more entertaining, I turn it into a world-building quest; a challenge to create a small world and tell a story. And, as in every narrative, you firstly need to consider two main questions:
- What is the story about? (you need to answer that with a single phrase)
- What is the audience of the story?
The first question will tell you what you need to present. Does your answer sound interesting? The second question should give you ideas about how you should present it.
Once you have answered them, you can organise your plot structure and decide what elements you should use. Below, I give five tips that may also help you on your journey.
1. Create a Lure
Some famous storytellers say you should begin your story with a hook. Alternatively, the well-known novelist Jeff VanderMeer, on his amazing Wonderbook, teaches something different:
"The fact is, anything at all can be made interesting, so you should think […] more about where in a particular situation or scene the interest lies, and from what perspective. A hook canoot be just a hook, either — it must be a lure and alluring, and it must also be an anchor. You are invinting the reader to some sort of enjoyment or challenge…"
You can use this strategy in your presentations. What is the most valuable piece of information for your audience?
For example, consider you built an ML model to predict sales for the forthcoming months. Therefore, instead of starting the presentation explaining all the techniques that you used to prepare the data and build your model, begin with interesting numbers obtained with your prediction.
Moreover, create the lure from the right perspective. For instance, the sales team are probably interested in how much money is involved in the predicted transactions. In contrast, the marketing team may prefer the number of items sold per customer segment and demographics.
Use your lure as an anchor. As you progress in the presentation — giving details about your findings, drilling down your results — the audience can easily get lost. Thus, use your lure to keep them connected to what is most important to them.
Coming back to our example, you may have started showing how much money is involved in the predicted transactions, but then you can explain how that money can be translated to market share. You jump from your anchor (money) to a new shore (market share data). If you keep doing it, the audience feels that everything is properly connected to what is most important to them. It's easier to follow.
Also, help the audience understand that the lure is just an invitation to something grandiose ahead. If you spoil all your insights in the lure, the audience will soon become bored and believe that they already got everything from you. If they understand you have more valuable information to share, they will remain connected.
2. Describe your Characters Properly
Just like in a story, we should also think of our results as characters.
Who is the hero of your story?
Maybe it's your novel algorithm or some insights from your data mining experiments. As scientists, we are passionate about describing everything in detail. However, this may be a terrible strategy to engage your audience because too much information can easily become too hard or too boring.
In a book, novel writers typically don't give you all the characteristics of their heroes. Usually, they describe what is more important or the most striking traces. Our brain imagines the missing pieces based on known archetypes and intuition. Try to use the same strategy with your results. Describe only the most important aspects of your research. You can always leave a link to detailed documentation at the bottom of the slides for those who want to understand all the nuts and bolts.
Additionally, abstraction is key for good descriptions. Use easy-to-understand diagrams, charts, metaphors, anything that guide the intuition and the understanding of the audience even if they don’t have a technical background. It's like pointing shortcuts to their minds.
3. Make Good Use of a Language
Did you watch the 2016 sci-fi movie Arrival? If you didn’t, watch it after reading this post.
Presenting data science results can be as hard as chatting with aliens. And imagine they need to understand you or they may destroy your world. To avoid that, be like Louise Banks and try the following:
- Find out or establish some “common language” between you and your audience.
- Communicate the message developing the use of that language.
You can use symbols, colours, keywords as the building blocks of your language. Anything that can make it easier for your audience to follow your story and keep them connected.
Once you define your language, be consistent. If you choose a colour to represent something (a variable, a class, or a strategy), always use the same colour throughout the presentation. If you need variance within that tone, change transparency but keep the hue. Also, keep consistency among the variable names, charts, categories… everything.
4. The Plot Thickens!
After data science experiments, we usually have several charts and diagrams to present. The order in which you present your results can help you tell a better story. Keep in mind that, just like in books, chronological order is not necessarily the best choice to make the story interesting.
Unfortunately, I don't believe there is a golden rule when it comes to result ordering. Two aspects I consider are:
- The complexity of the results;
- Their impact.
Firstly, build complex results from simpler pieces. Start with elementary premises and keep adding elements until the bigger picture is complete. This is a simple strategy to let your audience follow your reasoning. It's like puzzle-building, and you should let everyone play, one piece at a time.
Secondly, the most impactful results should be closer to the end. This way, the audience is more likely to remember them. However, the impact is something hard to measure because it depends on perspective. You can try to guess it and, over time, as you gain experience, your choices will be better.
5. No Open Endings
Science is open-ended. Your slides don't need to be. I believe conclusion slides work better if they are concrete and objective. Often, I replace the conclusion slide by takeaways. I list the most valuable pieces that my audience should keep in their minds.
Depending on the context, the audience may expect to hear about next steps (or future work). In this case, there is room for an open ending, but don’t forget to close your current story very well first.
Some presentation guides tell you to do a summary of your work in the conclusions. Personally, I find it boring, and the audience gets distracted, willing to go away. If you delivered an engaging talk, people would remember your story. You better highlight your impactful insights and end like a hero.
Do you have your own tricks to make your storytelling more interesting? Please let me know in the comments! Moreover, if you want to keep in touch and receive some data science tips + geek culture in your e-mail, please subscribe to my soon-to-start monthly newsletter.
