What "Dream Big" Meant for Data Science Innovation at LinkedIn
"I feel stuck maintaining dashboards and analyzing A/B tests" is a sentiment that I heard from many data scientists at LinkedIn and elsewhere.
Those activities are essential for running the Business. However, the data scientists also dream big of innovating with ML and AI for making a bigger business impact that also helps their career growth.
The good news is that businesses increasingly want data teams to step up to drive more innovation, especially as AI becomes more accessible.
In Wavestone's "2024 Data and AI Leadership Executive Survey" results, we see 77.6% of organizations say they're "driving business innovation from data" – up from 59.5% the prior year. However, the principal challenge to becoming data-driven is "culture/people/process/organization" according to 78% of the companies, rather than the technology.
These statistics should help data scientists feel more confident in stepping up to lead business innovation beyond maintaining dashboards and analyzing A/B tests.
As a Data Science leader formerly from LinkedIn, I'll share with you my process of driving business innovation with our "Dream Big" culture.
"Dream Big" Days At LinkedIn
When I led data science teams at LinkedIn, we would periodically host a "Dream Big" day or equivalent event. On this day, we ran brainstorming sessions for innovation, alongside fun team-building activities.
Beyond our data scientists, we also invited a diverse range of people from product development, finance, sales, marketing, operations, engineering, and other departments. Together, we would prioritize business challenges and brainstorm possible solutions as an extended team.
The collaboration was vital for understanding customer challenges and then building new solutions together. For example, we combined our unique understanding of the customer experience:
- Sales reps brought poignant customer stories.
- User experience researchers brought structured customer interviews.
- Data scientists brought customer usage patterns and trends data.
The diversity of ideas and viewpoints fostered a wide range of creative ideas for innovation. From the ideas, we aimed for "big bet" revolutionary leaps rather than "sustaining" evolutionary increments. We aimed for a moonshot.
It was also more fun with others.
The events often included other fun team-building activities like playing games together, singing karaoke, going on a hike, and even learning yoga together. The purpose of these events was to inspire the extended team to dream big and build new data solutions together.
These fun "Dream Big" joint sessions, along with other events like hackathons, fostered ongoing innovations and collaborations as part of our business culture.
Brainstorming Techniques From Stanford
Where did I first learn creativity, innovation, and collaboration practices to apply to the data science culture at LinkedIn? I'm grateful to have learned many of these "product development" techniques earlier from school.
While I was an engineering student at Stanford University, I took a required course on "Ambidextrous Thinking." Design and art professors taught this engineering course to spark creativity among engineers. They taught us to use our "right brain" for creativity and intuition, in addition to our usual "left brain" for logic and analysis. The intent of combining left brain and right brain thinking was to produce more innovations in the world.
I remembered the professor assigned us homework the first day: "Your first homework assignment is to get a sketchpad, sit outside, and draw your ideas without words."
- Week 1: Sketch out 300 unsolved problems in the world.
- Week 2: Pick one of those problems with a partner. Then draw 300 proposed solutions.
- Remaining weeks: Pick one idea to build a prototype, test, and iterate.
These exercises taught us to go fast, let go of perfection, and stretch for out-of-the-box creativity while working with a partner.
Why 300 ideas? The first 100 ideas tended to be conventional and frankly boring. The 300th idea, however, ended up being very unconventional, unique, and possibly worthy of a patent. True innovations came from those rare gems that were unearthed only after churning out a massive quantity of initial ideas.
- Example problem: Too laborious for supermarkets to manually retrieve shopping carts back from the parking lot.
- Example solutions: Self-driving shopping carts, aerial drones to cars in the parking lot, conveyor belts from supermarket to cars, direct delivery to consumer's homes via conveyor belts, etc.
Each of us did these design activities with a partner. My partner and I laughed at many of the extreme ideas in our sketchbooks, which looked like comic books. Brainstorming is more fun and creative with other people.
I'm grateful to have applied these same product development principles of brainstorming and collaboration later for data science.
Phases of Innovation From Dream Big to Deliver Big
Brainstorming is only one part of a culture of innovation. The ultimate goal is always to turn those "Dream Big" ideas into reality to make the world better. With limited resources, we need to be systematic about which big ideas to advocate and to invest.
Author Jim Collins wrote about the concept to "fire bullets, then cannonballs." What this means is to first try lots of quick easy experiments, like firing small bullets, until we hit our target. Once our aim is proven, then we earn the trust for the big investment to scale the winning solution, like with our one cannonball.
Fire bullets, then cannonballs. — Jim Collins
That one big cannonball for a data-driven solution might require us to request millions of dollars of funding from a business executive. There's a reputational cost for everyone too. We'd want to make sure that we're able to demonstrate that our aim is true with the smaller prototype tests first.
Here's how I organized collaborative data science AI projects to turn those multiple "dream big" ideas from initial trials into a breakout winner. The three phases of data innovation:
- Exploration: quickly evaluate different data ideas to narrow down which ones to spend more effort on
- Experimentation: build pilot prototypes for people to use
- Expansion: scale the data solution and generalize it
Phase 1: Exploration
In the first exploration phase, we're brainstorming a variety of different ideas, then quickly validating the best ideas to proceed to the next phase.
For example, from our "Project H" brainstorming sessions to help customers, our data scientists and business partners collected over 100 hypotheses on what B2B business clients did to get successful results with ads on LinkedIn. The group narrowed down the ideas to the top 10 hypotheses for further analysis.
Our senior data scientist analyzed each of those hypotheses using peer-reviewed observational causal inference techniques. These observational studies were relatively quick and easy to do, compared to A/B tests. A/B test experiments could have disrupted customers, risked losing business revenue, and added extra workload to teams like customer service. Observational studies were sufficient for this early stage of exploration.
The observational studies showed everyone for the first time using real data which hypotheses had merit. Some of the unexpected results challenged long-held business assumptions. Stakeholders were happy to see these findings from Project H, and started using the data insights right away as recommendations for clients.
However, we didn't stop here with the insights. This exploration milestone was only the first step toward bigger goals.
Phase 2: Experimentation
After delivering the initial data insights, our next phase for experimentation was to pilot a repeatable data solution with real users based on those insights.
A repeatable data solution could be in any of these forms:
- Dashboard or other business intelligence report
- Machine learning model, like recommendation systems
- Custom app or other UI for user interactivity
The point here is to take the one-time insights from Phase 1 to build a data product for pilot users to start using on a regular basis in Phase 2.
This phase is also high collaborative.
Over an informal water cooler chat, a sales director and I explored wishful thoughts about Chrome web browser extensions. What if we deliver the customer recommendations from our "Project H" recommendation system via a custom Chrome extension for LinkedIn.com?
- After a few hours, we built a working proof-of-concept custom Chrome extension which we demonstrated to the working group.
- Two months later, we trained our sales and customer service pilot users, and launched our internal prototype.
- A year later, we attributed over $75 million incremental revenue from the Project H prototype.
The innovation process spanning from the initial brainstorming to launching the pilot required close collaboration between data scientists and other business partners.
In Phase 2 experimentation, we proved out the innovation idea via the successful prototype with pilot users. As per Jim Collins' "bullets" analogy, we were able to hit our business target with our aim, and were now ready for Phase 3 expansion.
Phase 3: Expansion
Now in this final phase of innovation, we get to expand the pilot solution to broad audiences. Again, we need to collaborate with more people to scale the solution.
For Project H, here's what we did to scale our solution:
- We shared the incremental business impact and user testimonials with broad internal audiences. More sales and customer success teams were onboarded to use our prototype solution.
- The company C-suite executives heard about our innovation and invited us to meet them to talk about our pilot project successes and collaboration.
- Two years later, product development and engineering finished integrating the functionality from our initial hacky Chrome extension prototype into the main LinkedIn internet-facing website. The recommendation system was scaled into a generalized system too. Customers benefited and our company benefited.
Here are ideas that you can use for expanding your data solution:
- Generalize the solution: Encourage multiple stakeholders to use the same back-end dataset or model outputs. Create an API for AI-as-a-Service.
- Celebrate the wins: Measure the business impact. Collect and share user testimonials. Be sure the wins connect back to the same business KPIs that you hear at Company All Hands meetings.
- Spread the word: Tell others about the data solution, internally and possibly externally. Blog about it, announce it at group events, and tell people one-on-one.
With these steps you can drive "Dream Big" innovation with data science AI solutions.
Final Thoughts on Dream Big With Others
If you want to go fast go alone. If you want to go far go together. — African Proverb
In my experience, data science teams can do a lot more than just build more dashboards. It just takes some Leadership abilities to proactively engage with business partners to dream big together and deliver amazing results together.
My Project H example above started off as a small KPI metric dashboard request from one product manager. However, through proactively collaborating with additional business functions, we were able to reframe the original project into a bigger multi-phase AI/ML project that helped more customers to be more successful, rather than just make the measurements.
In summary, the "Dream Big" culture allowed us to brainstorm and deliver bigger data science projects. Together with the "relationships matter" culture, the end results helped customers, the company, and us individually as data scientists. I now coach data science leaders to apply these creativity, collaboration, and leadership techniques to drive bigger projects in their business.
You can do do this too!
Enjoyed Reading This Story?
Get notified when I publish a new story: subscribe for free
