A new take on data-driven manufacturing
Our investment in Fero Labs & how ML is making manufacturing more sustainable and efficient
We’re seeing a wave of machine learning companies tackling optimization problems old and new — everything from delivery routing and dynamic production scheduling to plant-based dairy formulation. These companies are taking advantage of new modeling techniques, increased data sources, compute accessibility, and their customers’ continued migration to the cloud. We get excited when we see emerging technologies like this transforming large, real-world industries. Fero Labs is just that: cutting-edge machine learning for optimizing complex manufacturing processes.
Manufacturing presents a huge opportunity for applied machine learning. It’s a massive space — north of $5 trillion in the US alone — and increasingly data-rich. Over the past few decades, manufacturers have invested in new technologies, from process management software to industrial automation and robotics. More recently, we’ve seen increased investment in data collection: sensors that generate enormous amounts of production data and the data historians for organizing the resulting time-series data. As manufacturers invest in their data assets, they need a way to make sense of that data. By layering in intelligence, they’re able to optimize processes at scale and increase automation.
Process optimization is certainly not new to the manufacturing space. There’s a long legacy of process improvement using both Six Sigma and Lean Manufacturing. While different, both methodologies strive to minimize waste, reduce defects, and increase efficiency. Both also rely on data to identify areas for improvement through understanding process flows and variation. However, the methods for establishing new processes, resolving issues, or generating ideas have still been relatively manual in nature. For example, in talking to process engineers, we heard that they will pull time-series production and output data for a period in question and investigate to understand what happened in order to resolve the issue. Both the diagnosis and remedy are reached through first principles and heuristic-based reasoning. While these methods can prove valuable, the results are likely not optimal and the remediation process itself consumes significant time and resources.
Fero Labs offers tooling to superpower process, industrial, and chemical engineers. There is a large expert workforce in manufacturing, made up of hundreds of thousands of engineers who are deeply knowledgeable about their products and processes, but often not about data science and machine learning. Fero aims to equip them with new data-driven capabilities. Given the complexity of a six-stage process with hundreds of data points coming off of the line per second, and an interwoven set of correlations between each step, there is no way the human brain could effectively optimize a process like steel formulation or predict shampoo quality. Fero enables manufacturers to leverage their teams’ expertise while leaving the quantitative and predictive heavy-lifting to Fero.
One aspect of Fero’s technology that is critical for these industrial users is its “explainability.” Many machine learning products are “black box,” meaning that you can’t interpret the rationale behind a prediction or recommendation. In manufacturing, being able to understand what’s behind a recommendation is critical for both gaining the trust of users and for leveraging their expertise in problem framing and solving. A core advantage of Fero Labs is how they’re productizing “white box” machine learning to meet customer needs.
Fero’s co-founders, Berk Birand (CEO) and Alp Kucukelbir (CSO), are incredibly well-suited to the problem they’re tackling. With PhDs from Columbia and Yale, respectively, both are experts in machine learning and deeply understand the challenges and opportunities of implementing applied machine learning in the industrial sector. In particular, Alp is a leading researcher in Bayesian inference and probabilistic programming, which provides the technical backbone for the company’s explainable ML products.
Lastly, we are spending more and more of our time — personally, as a society, and as a firm — thinking about climate change and decarbonization strategies. We see Fero Labs as a critical player in the decarbonization of industrial sectors. While Fero’s product can of course help with throughput maximization and cost minimization, Fero also helps manufacturers use machine learning to make their processes more sustainable. The team is already helping to decarbonize steel production through their work with Gerdau, where they are increasing recycled inputs to decrease costs and amounts of virgin materials used. They’re also helping Volvo Trucks minimize paint waste. We see a future where industries are incentivized and/or mandated to decarbonize and believe Fero will be an important part of helping companies achieve their targets.
As investors in Fero’s seed round in 2020, we had front row seats to the team’s progress in building product, generating early commercial traction, and driving results for customers. We are thrilled to deepen our partnership with Fero Labs by leading their Series A, and we are eager to see the team transform manufacturing with root cause analysis, better predictions, and greener processes.
