Zillow Was Just The Beginning. Companies Are Feeling The Pain Of Failed Data Science Initiatives.
Zillow's Data Science failures were apparent. Most businesses struggling with Data Science are not. Zillow took the bold step of throwing their Data Science team straight under the bus giving us more transparency than we typically get.
When they did, Zillow shined a light on a growing problem for business and our field. Data Science initiative failures are rampant even in mature companies with capable teams. Losses that happen in the dark have the same level of cost and consequence as Zillow's. The threat to Data Science teams is also the same.
Zillow decided to take the hit all at once. They disclosed the loss and dropped the business unit altogether. Better to be done with it than let it drag out over a year. The same events are playing out in other companies more slowly and privately.
Several companies have been taking hits from failed Data Science initiatives over the last three months. I will explain what a Data Science failure looks like to an outsider. I'll talk about what's going on inside the company. I’ll use counterexamples from businesses that did execute and their outcomes.
‘AI Strategy and business transformation are necessary to succeed with Machine Learning,’ is the underlying theme but this post focuses on tangible examples. If you want to dive into the solutions, I teach a class about AI Strategy and wrote a post about a critical role that businesses need to drive AI Strategy and transformation.
Failure To Successfully Deploy A Data Science Platform
Zillow wasn't the only recent platform failure. IBM missed out on the chance to be a dominant player in the AI space. They launched Watson years before any other Machine Learning based platform. They were so well positioned that their missteps have been painful to watch over the last seven years.
This month, the sale of Watson Health brought an end to the most significant Data Science cautionary tale to date. Everyone who deployed an AI platform faced early challenges. Amazon, Microsoft, and Google all overcame what IBM could not.
IBM oversold Watson's capabilities from the earliest days. Its appearance on Jeopardy was spun into promises of applications across industries. Those never came to pass because Watson could not deliver.
Microsoft faced a similar challenge in Azure, their cloud offering. It was an ugly failure, and they knew they had to make impactful changes. IBM worked on their issues from a technology perspective. Microsoft realized the problems started at the business level.
Microsoft changed leadership and the business around Azure and built a strategy-first approach. The changes and reorganization were expensive and painful. It took buy-in from the top and discipline to work on a long-term journey.
Azure's subsequent success allowed Microsoft to make its next move from cloud to AI. Again, this required organizational change. Again, they have achieved success.
IBM looked for quick technical fixes. Then they followed Zillow's playbook. They blamed the Data Science team and the technology, just not publicly. They pushed Watson Health into hospitals before it worked as well as they promised it did. Within two years, those early customers uninstalled Watson. Those were ugly public breakups.
AI requires strategy and organizational changes to be successful. The business alignment is just as important as the technology and Data Science team's capabilities. IBM and Microsoft had utterly different outcomes because one realized that while the other did not.
Failure To Apply Data Science For Revenue Growth
Netflix and Peloton have taken a beating this month due to failed Data Science initiatives. Netflix is seeing a slowdown in subscriber growth, and its stock has been dropping in response.
Amazon and Facebook realized subscriber growth has an upper limit. They wouldn't always be able to rely on new subscribers for growing revenues. From Sage Maker to Alexa, Amazon has monetized Machine Learning to drive growth when subscriber numbers stalled. Facebook leverages Machine Learning to keep users on-platform longer. Revenue growth came from increasing the revenue per user, so revenue growth did not decrease when user growth slowed.
There must be a monetization strategy. The quiet Data Science initiative failures at Netflix are a lack of paid features supported by the technology. They haven't found ways to open new business models or improve their monetization of existing customers.
Peloton isn't aware of its subscribers' power to drive revenue. They were firmly focused on selling bikes. They made a significant investment to increase their manufacturing capacity. This month they idled most of that manufacturing capacity due to plummeting demand.
They have an excellent Data Science team capable of building demand forecasting models. Outside the company, investors asked questions about post-pandemic behaviors and their potential impacts on demand. Peloton's first Data Science initiative failure was not seeing this coming sooner.
The second failure was not finding the opportunity to grow revenue through their subscriber base as new bike sales fell. They have everything in place to evaluate new opportunities but did not execute. Again, investors talked about increasing revenues from subscribers. There was plenty of data to act on, and again, it seemed like everyone else knew more about Peloton's business than they did.
Like Netflix, Peloton has opportunities to monetize its subscriber base with Machine Learning based products and features. This is another missed opportunity because Peloton is married to a legacy business model.
These are not apparent Data Science failures until you compare companies struggling to companies succeeding. The difference in behaviors and outcomes exposes Zillow-type failures across many businesses.
Failure To Apply Data Science To Optimize The Business
Supply chain disruption has been a recurring theme for almost 18 months. Companies like Apple have used Data Science to make their supply chain more resilient. Intel and GE have not.
Apple saw this coming just like everyone else did. They felt the impacts last September and October, taking a $6 billion hit to growth from supply chain constraints. However, they used their Machine Learning capabilities to optimize their supply chain and reduce future impacts.
Tim Cook said these issues are impossible to resolve completely, but their team did everything possible to put Apple in the best position and prevent further impacts to growth. This most recent quarter's earnings beat expectations and showed a decreasing impact from supply chain issues.
When it comes to Data Science capabilities, Intel and GE are comparable. However, neither one has been able to deploy those capabilities to improve their supply chains in the way Apple has. I am not making a 1 to 1 comparison because Apple has very different products and needs. However, Intel and GE have not achieved Apple's progress in reducing the impacts.
Intel's main competitor, AMD, is about to roll out several new chip lines while facing the same challenges Intel is. AMD has seen double-digit growth over the last two years while Intel is challenged to execute.
Intel has known about this problem since 2019 and still hasn't made the same improvements as competitors. There are many points of failure, but Data Science initiatives are undeniably falling short of their potential.
Underestimating Costs And Complexity
Amazon, Tesla, and Google have all hit this challenge. Elon Musk's quote about how autonomous driving is "actually a really hard problem" is a case in point. Ford's early push then retreat on autonomous is a significant Machine Learning initiative failure. That failure allows Tesla, who is figuring out how to meet the challenge, to take market share from them.
Repeatedly, companies have jumped into complex Machine Learning challenges without listening to their experts' advice about cost and complexity.
Amazon is another example. One of their most significant headwinds is labor costs. Amazon is an intelligent business that saw this coming. They have been automating much of the physical labor Amazon Prime is built on.
However, it's a tough challenge. Delivery drones, autonomous trucks, and robotic warehouse automation have all taken longer than expected. Prime's massive growth rate has been resting on an equally massive labor pool.
It's unsustainable given the current rising labor costs and talent shortage. Amazon is doing Herculean work to keep its capabilities stable. Growth will eventually be limited if they cannot move their automation initiatives forward.
Google walked into the robotics challenge by purchasing Boston Dynamics and building out a robotics arm. That closed with the sale of Boston Dynamics to Softbank then Hyundai. The problem is more extensive than Google expected. They did not have the business model to support all the investments required to succeed.
Expect remarkable advances to be coming out of Hyundai over the next two years. Their background in manufacturing and robotics puts them in a position to succeed. They knew the cost and complexity. They have the right business and operating models to monetize advanced robotics capabilities.
However, they are vulnerable to the same risk of failure. They need to implement the strategy and execute or they will be the next in a growing list of initiative failures.
A Multiverse Of Metaverse Wannabes
I am calling my shot with this final section. Look at every business talking about building a metaverse or growing their business in the metaverse. If they have never successfully deployed a large-scale Machine Learning platform, it's all smoke.
This is the next field of failures because metaverse success requires advanced Machine Learning capabilities. Machine Learning is the platform the metaverse is built on. The businesses that fail to execute do so because of several failed Data Science initiatives.
The companies I've mentioned have best-in-class Data Science teams and Machine Learning capabilities. Failures from Zillow, IBM, Netflix, Intel, and Peloton are eye-opening because they show how even capable businesses can fail. It is not enough to simply hire Data Scientists and invest in the technology.
There is a strategic component to success. Companies must build their business and operating models to monetize the technology.
Organizations must be built to deliver solutions to complex challenges. Data Science fails on an island. Capability, product, and functional silos must be broken down.
Businesses either prepare users for Machine Learning based products, or they will not adopt them. Their value will be wasted. Customers must also be prepared, or they will not buy.
The outwardly visible consequences of Data Science initiative failures are not obviously connected. They look like past failures, and it is easy to assume the root causes are the same now as they were before.
There will be an increased focus on AI strategy and Data Science initiative execution. As investors begin to connect the dots, the C Suite will become much more engaged. All signs point to increasing accountability for our field.
Where does the data science strategist fit in this? In terms of the monetization failures of revenue growth and business optimization, the business isn't taking advantage of the opportunities that exist with DS/ML. Do you think it's due to the data strategist/CDO not being a strategic partner and advocating for the capabilities or the business not thinking about how to best use these capabilities? Or is it execution failures of having the strategies to do it but not being able to actually implement their strategies?