The Guide
You've just done an incredible analysis on your tool of choice (when we explored this dataset ourselves we used a Jupyter notebook and open source Python libraries, but of course it could be anything), full of figures, code—maybe including multiple dead ends—to support your conclusions. Now you are asked to present your work. You start showing your notebook, and explaining all the details to your audience. Except you notice that after the first few minutes everyone is distracted and you seem to have lost your audience. What went wrong? And how can you keep your audience engaged?
Lead with the Solution
Attention—both when reading and listening—is highest at the beginning and at the end, that’s why you should lead with your main message. Put the bottom line up front (BLUF) to maximize the chances of your audience retaining the main message.
To do this well, you must fine tune your presentation to the audience that is listening. There are many ways to think about this, but we believe the most useful dimensions to consider are a) the audience’s context for the problem and b) their appetite and ability to take on future actions. To give a concrete example, let’s consider three primary archetypes of people to present to: Data Science Peers, the Cross-Functional team and Sponsors.
If you have to remember only one thing about this, it should be the following table, where we have summarized our suggestions.
Peers: Data scientists who are working on the same team as you
Cross-Functional team: Other types of professionals you are collaborating with
Sponsors: The people who commissioned this work
Let’s start in descending order of context: we’ll start with the Peers group—other data scientists—then talk about the Cross-Functional team of stakeholders. Finally we’ll talk about how to present your work to the busiest (and most influential) group: the Sponsors.
Lead with the Solution
Attention—both when reading and listening—is highest at the beginning and at the end, that’s why you should lead with your main message. Put the bottom line up front (BLUF) to maximize the chances of your audience retaining the main message.
To do this well, you must fine tune your presentation to the audience that is listening. There are many ways to think about this, but we believe the most useful dimensions to consider are a) the audience’s context for the problem and b) their appetite and ability to take on future actions. To give a concrete example, let’s consider three primary archetypes of people to present to: Data Science Peers, the Cross-Functional team and Sponsors.
If you have to remember only one thing about this, it should be the following table, where we have summarized our suggestions.
Peers: Data scientists who are working on the same team as you
- Existing Context: High
- Appetite for Action Items: Low
Cross-Functional team: Other types of professionals you are collaborating with
- Existing Context: Medium to High
- Appetite for Action Items: Medium
Sponsors: The people who commissioned this work
- Existing Context: Low
- Appetite for Action Items: High
Let’s start in descending order of context: we’ll start with the Peers group—other data scientists—then talk about the Cross-Functional team of stakeholders. Finally we’ll talk about how to present your work to the busiest (and most influential) group: the Sponsors.
DALL-E prompt: Cyberpunk dogs doing data science presentations
Peers
A group comprised of other data scientists, who are mostly interested in the nitty-gritty detail of the problem. Their Existing Context is high, and when you present to them, get ready to answer specific questions about regularization, model selection, metrics and why things were done in that way. Your peers usually don't have decision making power, just intellectual curiosity that may help you sharpen your analysis, which means their appetite for Action Items is low. Both in our scenario and in a tech company, these are data scientists or analysts on your team or who work closely with the same or similar Cross-Functional team.
Peers, especially in academic settings, tend to poke at every assumption to really understand how your analysis supports the argument you're making. Do not take any of it personally! Suggesting a different path, or offering a critique to an idea, is it not an attack on you. Remain calm and answer to the best of your knowledge. It’s totally ok to say “I don’t know” or “I haven’t looked into this yet”. Adopt a growth mindset, and be thankful that your colleagues are making your results stronger.
Hopefully by the end of this presentation, you will have pressure tested your work and assumptions, or got ideas for some directions you haven’t thought of. Do keep in mind that this is your presentation at the end of the day, so you should feel free to cut off any line of questioning that seems like it’s getting too in the weeds and return to what you were saying before. In an ideal world—one with infinite time and resources—you would explore all the different ideas and ramifications, but in the world we are stuck in, it’s important to be mindful of the time and effort. It’s pointless to have the perfect analysis if no one can benefit from an analysis that will never be finished. It’s your duty to decide what paths are worth exploring and leave the rest as suggestions for future directions (which might just remain suggestions).
Cross-Functional team
The group that is mostly affected by the decisions. In our framework, their Existing Context is medium. They understand the problem space generally, but because they don’t know or care about the specific data science questions you’re working on, they will be more concerned about how your analysis applies to the real world. They may be interested in your findings generally, but they also want actionable insights, e.g. what resources do they need to deploy and how that would change the status quo. This means their appetite for Action Items is medium.
In a tech company, the Cross-functional team might include all kinds of stakeholders groups: e.g. Product Managers, Engineering Managers, Software Engineers, Researchers, Designers.
In our example case, this group would have representatives of residents of the BIDs, members of the business owners associations, sanitation workers, shop owners, local police, etc.
When presenting to the Cross-Functional team, focus your attention on what will change for their customers (people their work is in service of) and what the tradeoffs are. In most cases the Cross-Functional team will need to make tradeoffs, so it’s important to outline those. While it’s great for you to have an opinion on which tradeoffs are worth making, you should still present pros and cons of each option.
There is no need to go into much detail about the methodology or the techniques used, but you should however make sure to emphasize the limitations of your analysis. For example, if we are only tracking positive feedback, does an increase in those imply an improvement across the board? Maybe one business improvement district is getting great service (thus increasing the positive feedback) while the rest are in shambles. If we don’t take into account those limitations and assumptions made in creating the dataset, we can get fooled by our own data.
Sponsors
Sponsors are the people who have the most decision making power and are typically involved in many different things at once, which means their appetite for Action Items is high and their Existing Context is low. In our example the Head of Sanitation, the Mayor of NYC and the Head of the BID program would be in this group. At a large organization, Sponsors typically spend their entire day moving between presentations, with many people telling them what’s happening and asking them for support. Sponsors are mostly focused on the big picture and may forget or be unaware of context and specific details. Their goal is to collect as much information they can, connect you with others and make things happen.
When you present to them you should remember their low Existing Context and high appetite for Action Items by doing two things: remind them what the problem is (i.e. why you are presenting to them) and emphasize the results of the analysis, especially actionable parts. They may not have a lot of time to deeply consider your work, so presentations to them are most effective when you have clear requests for them. If you can make it easy to make a decision while giving them a broad enough context, you will be successful.
In most cases you should skim over or exclude technical details of the analysis (e.g. what kind of models you used and why). One exception of the above rule would happen when one of the execs is a former data scientist themselves, and thus might ask some specific questions about something they have expertise on or to reconnect with their past.
Note that having time with the Sponsor can be a blessing and a curse: While it’s always a great opportunity to be able to present to them (and might have career benefits down the line), they are very often busy and will frequently arrive late. This can make it feel like you’re already at a disadvantage and can make it legitimately stressful to tell them what to do.
Exposure to Sponsors is nearly always a little bit of a trust building exercise, as well. Not everyone will get to do this regularly, or it may not be a part of your job at all. It’s fine if that’s the case, but it’s always worth remembering that presentations for Sponsors require the most polish, so if you have to spend time fine tuning a presentation, this is it.
A group comprised of other data scientists, who are mostly interested in the nitty-gritty detail of the problem. Their Existing Context is high, and when you present to them, get ready to answer specific questions about regularization, model selection, metrics and why things were done in that way. Your peers usually don't have decision making power, just intellectual curiosity that may help you sharpen your analysis, which means their appetite for Action Items is low. Both in our scenario and in a tech company, these are data scientists or analysts on your team or who work closely with the same or similar Cross-Functional team.
Peers, especially in academic settings, tend to poke at every assumption to really understand how your analysis supports the argument you're making. Do not take any of it personally! Suggesting a different path, or offering a critique to an idea, is it not an attack on you. Remain calm and answer to the best of your knowledge. It’s totally ok to say “I don’t know” or “I haven’t looked into this yet”. Adopt a growth mindset, and be thankful that your colleagues are making your results stronger.
Hopefully by the end of this presentation, you will have pressure tested your work and assumptions, or got ideas for some directions you haven’t thought of. Do keep in mind that this is your presentation at the end of the day, so you should feel free to cut off any line of questioning that seems like it’s getting too in the weeds and return to what you were saying before. In an ideal world—one with infinite time and resources—you would explore all the different ideas and ramifications, but in the world we are stuck in, it’s important to be mindful of the time and effort. It’s pointless to have the perfect analysis if no one can benefit from an analysis that will never be finished. It’s your duty to decide what paths are worth exploring and leave the rest as suggestions for future directions (which might just remain suggestions).
Cross-Functional team
The group that is mostly affected by the decisions. In our framework, their Existing Context is medium. They understand the problem space generally, but because they don’t know or care about the specific data science questions you’re working on, they will be more concerned about how your analysis applies to the real world. They may be interested in your findings generally, but they also want actionable insights, e.g. what resources do they need to deploy and how that would change the status quo. This means their appetite for Action Items is medium.
In a tech company, the Cross-functional team might include all kinds of stakeholders groups: e.g. Product Managers, Engineering Managers, Software Engineers, Researchers, Designers.
In our example case, this group would have representatives of residents of the BIDs, members of the business owners associations, sanitation workers, shop owners, local police, etc.
When presenting to the Cross-Functional team, focus your attention on what will change for their customers (people their work is in service of) and what the tradeoffs are. In most cases the Cross-Functional team will need to make tradeoffs, so it’s important to outline those. While it’s great for you to have an opinion on which tradeoffs are worth making, you should still present pros and cons of each option.
There is no need to go into much detail about the methodology or the techniques used, but you should however make sure to emphasize the limitations of your analysis. For example, if we are only tracking positive feedback, does an increase in those imply an improvement across the board? Maybe one business improvement district is getting great service (thus increasing the positive feedback) while the rest are in shambles. If we don’t take into account those limitations and assumptions made in creating the dataset, we can get fooled by our own data.
Sponsors
Sponsors are the people who have the most decision making power and are typically involved in many different things at once, which means their appetite for Action Items is high and their Existing Context is low. In our example the Head of Sanitation, the Mayor of NYC and the Head of the BID program would be in this group. At a large organization, Sponsors typically spend their entire day moving between presentations, with many people telling them what’s happening and asking them for support. Sponsors are mostly focused on the big picture and may forget or be unaware of context and specific details. Their goal is to collect as much information they can, connect you with others and make things happen.
When you present to them you should remember their low Existing Context and high appetite for Action Items by doing two things: remind them what the problem is (i.e. why you are presenting to them) and emphasize the results of the analysis, especially actionable parts. They may not have a lot of time to deeply consider your work, so presentations to them are most effective when you have clear requests for them. If you can make it easy to make a decision while giving them a broad enough context, you will be successful.
In most cases you should skim over or exclude technical details of the analysis (e.g. what kind of models you used and why). One exception of the above rule would happen when one of the execs is a former data scientist themselves, and thus might ask some specific questions about something they have expertise on or to reconnect with their past.
Note that having time with the Sponsor can be a blessing and a curse: While it’s always a great opportunity to be able to present to them (and might have career benefits down the line), they are very often busy and will frequently arrive late. This can make it feel like you’re already at a disadvantage and can make it legitimately stressful to tell them what to do.
Exposure to Sponsors is nearly always a little bit of a trust building exercise, as well. Not everyone will get to do this regularly, or it may not be a part of your job at all. It’s fine if that’s the case, but it’s always worth remembering that presentations for Sponsors require the most polish, so if you have to spend time fine tuning a presentation, this is it.
DALL-E prompt: oil painting of cyberpunk dogs doing data science presentations with dramatic lighting in the style of caravaggio
Data Science Problem
While the Solution section is the most impactful, the Data Science Problem section is still a crucial one. It is the section most likely to vary in length (the shortest being for Sponsors and the longest being for Data Science Peers), because here you discuss the specifics of your analysis and how you translated the real world problem into a problem that can be solved with Data Science methods. For the purposes of this guide, we are not debating the specific DS choices, we are just advising on presenting your work.
This section will be the most detailed, but it should still have an easy-to-follow structure. Don’t describe the entire process you went through to go from the beginning of your analysis to the end. Instead, think about what information supports the claims you just made in the Solution section prior, and throw out any part of the Data Science Problem section that distracts from the point you are trying to make.
This is also the section where you’ll want to talk about the assumptions behind your analysis, as well as the limitations of your methodology. It might seem intimidating to talk about limitations, especially to a group of skeptical Peers or to a Sponsor who you only have a small amount of time to persuade, but it’s important to remember that every kind of analysis has its limitations. It might feel like weakness to admit to not knowing something, or like you are weakening your argument, but being transparent about this builds trust with the audience.
Beyond assumptions and limitations, you’ll want to think carefully about the goal of the presentation and how that relates to the audience’s appetite for Action Items. For example, if presenting to Data Science Peers, you’re probably seeking technical or methodological feedback, so you should focus this section on what you want their perspective on, and not spend time on things you have high confidence in. For a medium appetite group of Cross-Functional Stakeholders, you’ll probably be better served by talking about the ways in which the analysis represents or doesn’t cover their area of expertise, and testing how well the assumptions of your analysis match with their lived experiences. For a Sponsor, you probably only want to stick to communicating your assumptions and limitations. They likely won’t have time to digest much else!
While the Solution section is the most impactful, the Data Science Problem section is still a crucial one. It is the section most likely to vary in length (the shortest being for Sponsors and the longest being for Data Science Peers), because here you discuss the specifics of your analysis and how you translated the real world problem into a problem that can be solved with Data Science methods. For the purposes of this guide, we are not debating the specific DS choices, we are just advising on presenting your work.
This section will be the most detailed, but it should still have an easy-to-follow structure. Don’t describe the entire process you went through to go from the beginning of your analysis to the end. Instead, think about what information supports the claims you just made in the Solution section prior, and throw out any part of the Data Science Problem section that distracts from the point you are trying to make.
This is also the section where you’ll want to talk about the assumptions behind your analysis, as well as the limitations of your methodology. It might seem intimidating to talk about limitations, especially to a group of skeptical Peers or to a Sponsor who you only have a small amount of time to persuade, but it’s important to remember that every kind of analysis has its limitations. It might feel like weakness to admit to not knowing something, or like you are weakening your argument, but being transparent about this builds trust with the audience.
Beyond assumptions and limitations, you’ll want to think carefully about the goal of the presentation and how that relates to the audience’s appetite for Action Items. For example, if presenting to Data Science Peers, you’re probably seeking technical or methodological feedback, so you should focus this section on what you want their perspective on, and not spend time on things you have high confidence in. For a medium appetite group of Cross-Functional Stakeholders, you’ll probably be better served by talking about the ways in which the analysis represents or doesn’t cover their area of expertise, and testing how well the assumptions of your analysis match with their lived experiences. For a Sponsor, you probably only want to stick to communicating your assumptions and limitations. They likely won’t have time to digest much else!
DALL-E prompt: data scientist doing a presentation Bauhaus illustration
Conclusion
Once you have reached the Conclusion section, your point should be clearly made and well-supported by the details you included in the Data Science Problem section. Now all you have left to do is stick the landing!
Remember: attention is highest at the beginning and end of a presentation, so one of the most impactful things you can do is to restate the point you made at the beginning a second time, just before you finish speaking. It might feel redundant to do this, but after spending a portion of your presentation talking about specifics and details, most audiences will find it genuinely helpful for the presenter to zoom back out and remind them of the reason why they’re listening to you in the first place.
Summarize the most important points that support the main argument of your presentation, while still tailoring it to your audience’s Existing Context and keeping it brief. A conclusion should not take more than a few minutes, so details should be single bullet points in a list or simply recycling the headings from your slides. You can expand verbally on them if you really need to, but it probably won't be necessary. Your audience will have just listened to them described in more detail, so less is more here.
Once you’ve restated your main argument and supporting points, you should close with a call-to-action.
For an audience of Peers (whose appetite for Action Items is low), this might simply be asking for detailed feedback on your methodological approach in the Data Science Problem section. For example: “My original analysis did not incorporate which borough a BID is in as a factor that might drive sanitation expenses. Do you think that might be a useful variable, and if so how should I include it in a regression?”
For an audience of Cross-Functional Stakeholders who have medium appetite for Action Items, you might have two requests: asking them if you missed anything obvious from the perspective of their discipline, as well as giving them recommendations on what they should do with the information you just gave them. For example, “My findings showed that removing graffiti is a huge driver of sanitation expenses. For those of you who are shop owners, how much do you believe the presence of graffiti impacts whether people perceive a space as clean or not?”
For Sponsors, who have a high appetite for Action Items, your call-to-action should be recommendations about what you think they should do next. Do not deliver these as demands or orders, simply state what you believe to be the best next step based on the data and your analysis, making sure to remind them how the action you recommend is likely to support outcomes they are interested in driving. For example: “According to my analysis, the number of trash bags collected is a big driver of sanitation costs in a BID. I don’t think we should set a goal of collecting fewer trash bags since that might hurt overall cleanliness, but I do believe there are opportunities for more efficient trash can collection. Two options I’d suggest further research on are increasing trash can size, which would allow more trash to be collected using the same trash collection routes, and adjusting the placement of existing trash cans, which would alter collection routes but potentially make them more efficient and less costly to support.”
Once you have restated your main argument, summarized important details, and delivered your call-to-action, all you have left to do is thank your audience for their time and attention, and let them know how they can get in touch with you if they’d like to discuss the conversation further. You may have time for a Questions & Answers (Q&A) session, and it’s always a nice touch to send out a summary in writing later if you have the relevant contact information. And in either case, you should yield the floor and feel proud of a presentation well-given.
What's next?
If you use this guide to prepare your next presentation or generally find it helpful, please let us know by reaching via our contact info at the beginning or message us. We’d love to hear about how the presentation went, and what other content like this you would find useful in your data science journey!
And, if you feel so inclined, can buy us a coffee here.
Once you have reached the Conclusion section, your point should be clearly made and well-supported by the details you included in the Data Science Problem section. Now all you have left to do is stick the landing!
Remember: attention is highest at the beginning and end of a presentation, so one of the most impactful things you can do is to restate the point you made at the beginning a second time, just before you finish speaking. It might feel redundant to do this, but after spending a portion of your presentation talking about specifics and details, most audiences will find it genuinely helpful for the presenter to zoom back out and remind them of the reason why they’re listening to you in the first place.
Summarize the most important points that support the main argument of your presentation, while still tailoring it to your audience’s Existing Context and keeping it brief. A conclusion should not take more than a few minutes, so details should be single bullet points in a list or simply recycling the headings from your slides. You can expand verbally on them if you really need to, but it probably won't be necessary. Your audience will have just listened to them described in more detail, so less is more here.
Once you’ve restated your main argument and supporting points, you should close with a call-to-action.
For an audience of Peers (whose appetite for Action Items is low), this might simply be asking for detailed feedback on your methodological approach in the Data Science Problem section. For example: “My original analysis did not incorporate which borough a BID is in as a factor that might drive sanitation expenses. Do you think that might be a useful variable, and if so how should I include it in a regression?”
For an audience of Cross-Functional Stakeholders who have medium appetite for Action Items, you might have two requests: asking them if you missed anything obvious from the perspective of their discipline, as well as giving them recommendations on what they should do with the information you just gave them. For example, “My findings showed that removing graffiti is a huge driver of sanitation expenses. For those of you who are shop owners, how much do you believe the presence of graffiti impacts whether people perceive a space as clean or not?”
For Sponsors, who have a high appetite for Action Items, your call-to-action should be recommendations about what you think they should do next. Do not deliver these as demands or orders, simply state what you believe to be the best next step based on the data and your analysis, making sure to remind them how the action you recommend is likely to support outcomes they are interested in driving. For example: “According to my analysis, the number of trash bags collected is a big driver of sanitation costs in a BID. I don’t think we should set a goal of collecting fewer trash bags since that might hurt overall cleanliness, but I do believe there are opportunities for more efficient trash can collection. Two options I’d suggest further research on are increasing trash can size, which would allow more trash to be collected using the same trash collection routes, and adjusting the placement of existing trash cans, which would alter collection routes but potentially make them more efficient and less costly to support.”
Once you have restated your main argument, summarized important details, and delivered your call-to-action, all you have left to do is thank your audience for their time and attention, and let them know how they can get in touch with you if they’d like to discuss the conversation further. You may have time for a Questions & Answers (Q&A) session, and it’s always a nice touch to send out a summary in writing later if you have the relevant contact information. And in either case, you should yield the floor and feel proud of a presentation well-given.
What's next?
If you use this guide to prepare your next presentation or generally find it helpful, please let us know by reaching via our contact info at the beginning or message us. We’d love to hear about how the presentation went, and what other content like this you would find useful in your data science journey!
And, if you feel so inclined, can buy us a coffee here.