In 2007, two roommates struggling to make rent in San Francisco had a bright idea. A major design conference was coming up and hotel prices were soaring. What if, the roommates thought, they rented out air mattresses on the floor of their apartment to cash-strapped attendees — a kind of designers’ (air)bed & breakfast? It took off, and the pair realized they were onto something. Within four years, their Airbnb site was a global success, with one million bookings made across 89 countries.
Here’s another way to explain Airbnb: a sharing-economy, hotel-alternative concept headquartered in San Francisco, with a daily private room rental rate of $80.67 in the US, and annual US advertising spend of $23.5 million.
Which of these will stick in your mind? The first one, obviously. That’s because our brains are hardwired to remember stories far more than stats and figures, no matter how illuminating these might be.
It’s easy to forget this when you’re trying to get your voice heard by the decision-makers or to get your C-suite excited about a machine learning project in the works. You might assume that they’re looking for cold, hard facts. In reality, what they need to hear are machine learning stories that makes sense to them, one in which they turn out to be the hero.
Let’s take a look at some effective ways to do just that.
It’s far easier to spin a convincing narrative about things that have already happened than things that could happen. Take a look at case studies of great machine learning projects from your domain and industry, paying close attention to the concrete business benefits these generated.
People love talking about their successes, so you should be able to dig up some interesting and impressive examples from trade magazines, news sites, academic journals or even the blogs of companies and individual data scientists.
It’s great to have specific machine learning stories like these in your back pocket to share with your colleagues in order to lend credence to your proposed machine learning projects. However, you can also go a step further by treating these success stories as valuable data in their own right.
That means extracting key details from each story, like the algorithm, software and analytics model used by the team; the types of benefits they derived; and the costs associated with the project. From here, you can identify patterns for success to replicate in your own projects, whether that be based on supervised learning or unsupervised learning.
If you want all your stakeholders to be pleased with the outcome of your machine learning project, it’s absolutely vital to involve them in every step of the data science project life cycle. This includes the very early stages, as you figure out which datasets to use and in which direction to take the project.
That’s because people are far more receptive to an idea when they feel as if they came up with it in the first place. The more you get your non-technical colleagues to contribute to the design of your project, the more they will feel as if this is their story, and the more invested they will feel in the results.
One effective way to approach this is to draw out some simple observations from the data early on and discuss their implications with your business colleagues, long before you start to build a complex machine learning model.
For example, let’s say your company is interested in exploring ways to reduce staff absences and improve productivity. When you start reviewing the datasets made available to you, you notice a correlation between the employees who work from home the most and the employees who take sick days the least.
Pointing this out to business colleagues and asking what they make of it will likely spark far more interesting questions and precise lines of investigation, as well as highlighting what kinds of external data would help you make sense of this further. It’s a great way to harness their perspective and domain knowledge, and to make them part of the story you’re developing.
To get on board, your colleagues want to know if this story has a happy ending. Typically, that means getting a decent return on their investment in your machine learning project.
This, again, is why it’s so important to engage all stakeholders through the design of your project, establishing their key questions and concerns, and working out how these will inform the data science project lifecycle. You need to know what success for the rest of the team looks like so you can give them a convincing reason to throw resources behind your work. You need to understand how best to frame your machine learning stories around their interests and needs.
For example, is the business keen to reduce waste or inefficiencies? To improve its churn rate? To identify new ways to monetize its data? To bring down process failures or maintenance callouts? To find more accurate ways to assess whether a customer qualifies for credit or a loan? To increase revenue per customer?
Whatever your organization’s priorities are, it’s your job to figure out how your machine learning project will help them identify ways to achieve their goals.
The flipside of this, of course, is that you need to be keenly aware of the costs and risks involved in the project. That way, you can convincingly present a range of possible scenarios — storylines, essentially — and the varying ROI your colleagues could expect from each.
Sometimes it’s simply too difficult to get your colleagues to picture what your algorithm can offer them. Or to explain why the machine learning model you’re building is better than the system they have in place now. At this stage, you may have to jump in and build a prototype that helps you prove your point and sell your narrative to decision-makers in the business.
This approach can be very convincing, because you not only bring the concept to life for your colleagues, you may also be in a position to present a clear side-by-side comparison — a champion-challenger or split-test that demonstrates clearly why your new approach does the job better.
That said, this is a very expensive, resource-intensive strategy. You need a lot of clean, complete, ready-to-use data at your fingertips to build a decent prototype. Unless you’re confident that you can build something that does the bigger project justice, you could do more harm than good. Think very carefully about whether you’re in a position to prove what’s possible with what you currently have.
Telling a story in data science also means highlighting which part of the story you need help telling. What gaps can your colleagues fill in with their expertise and insight? Which types of data can you draw on from inside the company, and what do you need to bring in from outside? And, of course, what data science platform can help you bring all these strands together, helping ensure the data you use is clean, well-organized and fit for purpose?
It’s important to grapple with all these questions early on. That will help you get your machine learning stories straight — and make sure that you and all your stakeholders remain on the same page.