Canada: Flexport’s next AI-first office? 

Flexport, a software-powered freight forwarding company, needs to build an AI-first office in Canada in order to improve its product development and global operations.

Flexport (https://www.flexport.com/) is a freight forwarding company that uses software to help businesses optimize its supply chain and ship cargo around the world. Based in San Francisco, Flexport has raised over $300M in financing from Google Ventures (GV), First Round Capital, and other prominent investors.

Logistics is 12% of global GDP, and companies spend $1.1 trillion dollars on freight forwarding each year. The massive scale and complex network of people involved in transporting goods – from freight forwarders to retailers – make supply chain visibility a challenge for all stakeholders involved [1].

 

Machine Learning’s importance to Flexport’s product development

Machine Learning (ML) is a component of Artificial Intelligence (AI) that uses an algorithm to learn from prior data in order to produce a behavior or predict an outcome. As a software company that processes shipping data across air, ocean, rail, and truck for over 15,000 global businesses every day, Flexport is uniquely positioned to leverage that data to create new products that help businesses optimize their supply chains.

Given the complexity of the supply chain, the biggest data problem facing Flexport is that forwarders in Flexport’s networks provide unstructured data that are not labelled adequately for machine learning. Oftentimes, Flexport employees would need to call the forwarder, ask detailed questions on the data provided, and then manually update Flexport’s platform [2]. Without clean and accurate data to train machine learning algorithms, predictions for shipping events, shipment times, and shipping demands become practically impossible.

At the same time, legacy technology companies are starting to enter into the freight-forwarding space. Most notably, IBM launched Watson Supply Chain last year to create supply chain visibility in shipping [3]. IBM’s focus with this new platform is to leverage data from external sources (e.g., weather forecasts) and the forwarder’s internal data sources to predict supply chain disruptions. Given IBM Watson’s strong foothold in the machine learning space and its parent company’s financial resources, this new supply chain platform creates pressure for Flexport.

 

How Flexport plans to address the data challenges

In the next two years, Flexport’s goal is to provide better end-to-end services for companies, while including more inland services [4]. More importantly, Flexport will create easy-to-use interfaces for the forwarder to input unstructured data into a format that is accurate and labelled into the Flexport platform. The main focus here is to collect clean data throughout the entire system in order to train a machine learning algorithm.

The company is doubling down on efforts that will help factories coordinate bookings, logistics teams collaborating on imports, and supply chain managers relaying pickup details to truckers [5]. In order to succeed in the next two years, Flexport will continue to focus on operations and deliver a better customer experience at a relatively low cost.

Over the next ten years, the CEO plans to open offices in the top 20 exporting countries in the world and continue to build out its network of warehouses so that more data is being captured and fed into the company’s algorithm [6].

 

The case for building Flexport’s next AI-first office in Canada

My recommendation is for Flexport to build an AI-first (instead of an operations-first) office in Toronto, Canada.

Currently, Flexport has offices in the US (San Francisco, New York), Asia (Hong Kong, Shenzhen), and Europe (Amsterdam, Hamburg) that’s focused strictly on operations. All of these offices are strictly focused on operations and not on the research & development of machine learning expertise. Without world-class machine learning capabilities, Flexport will not be able to build the operating system for global trade.

Canada is the best place to build a world-class machine learning company. In addition to having two pioneers of machine learning in Geoffrey Hinton and Yoshua Bengio [7], Toronto is home of the Vector Institute for Artificial Intelligence – a $120 million initiative that exists to further generate and retain world-class AI talent.

The political landscape in the US reinforces Canada’s AI sector’s strength [8]. More than 50% of Toronto companies are seeing more US applicants in 2017 than in 2016, signalling that there is a potential “reverse brain drain” of top talent flowing from the US to Canada.

Additionally, Flexport building an office in Toronto will provide immediate access to top machine learning experts at a fraction of the cost. Not many Silicon Valley startups have adopted this strategy yet, which provides Flexport a first-mover advantage in nurturing machine learning talent.

 

Open-ended questions

  1. Does machine learning expertise need to be developed in-house in order to build a competitive advantage for product development?
  2. Is it better to solicit help from “AI-as-a-Service” vendors to help Flexport apply machine learning using data that Flexport collects so that Flexport can focus only on operations?

 

[Word count: 780]

 

References:

[1] “Retail’s Adapt-Or-Die Moment: How Artificial Intelligence Is Reshaping Commerce”. CB Insights, https://app-cbinsights-com.prd2.ezproxy-prod.hbs.edu/research/artificial-intelligence-reshaping-commerce/#supply

[2] “Ryan Petersen on Building Flexport, a Modern Freight Forwarder”. Y Combinator Blog, https://blog.ycombinator.com/ryan-petersen-on-building-flexport-a-modern-freight-forwarder/

[3] Vyas, Nick. “Four compass points for global supply chain management….revisited”. Supply Chain Management Review 22, ProQuest, http://search.proquest.com.ezp-prod1.hul.harvard.edu/docview/2111104217?accountid=11311

[4] Bauml, Manuel. “Air-freight forwarders move forward into a digital future”. McKinsey & Company, https://www.mckinsey.com/industries/travel-transport-and-logistics/our-insights/air-freight-forwarders-move-forward-into-a-digital-future

[5] Petersen, Ryan. “Announcing Flexport’s $110M Series C”. Flexport News, https://www.flexport.com/blog/announcing-flexports-110m-series-c/

[6] Nicholas Shields. “Top 5 startups to watch: Digital freight”. Business Insider Intelligence, https://intelligence.businessinsider.com/post/top-5-startups-to-watch-digital-freight-2018-7

[7] Irvine, Steve. “Canada can be an AI leader. I left Silicon Valley to prove it”. The Globe and Mail, https://www.theglobeandmail.com/report-on-business/rob-commentary/canada-can-be-an-ai-leader-i-left-facebook-to-prove-it/article33727882/

[8] Briggs, Paul. “Artificial Intelligence Incubators: Why Canada’s Hubs for AI Research Are Attracting Global Tech”. eMarketer PRO, https://content-na1.emarketer.com/artificial-intelligence-incubators

Previous:

Bioprinting – The Future of the Healthcare Industry

Next:

Major League Baseball and Machine Learning: Can the San Francisco Giants Compete in the Analytics Arms Race?

Student comments on Canada: Flexport’s next AI-first office? 

  1. You have identified a very interesting problem. I appreciate that you decoupled the need for machine learning here from the current business goals (operations) of Flexport. Combining the two could have proven a very difficult, and potentially fatal, pivot for Flexport.
    While I agree that Flexport should bring in outside vendors to help create the machine learning solution here, I think Flexport should play a more active role in shaping the product to address the bottleneck to its use that you identified in your article. By employing machine learning, the product could use various variables (who is inputting the data, seasonality, size of shipment, destination, etc.) to suggest values for the fields that need to be entered by the shipper. That way, the effort needed is lower, and Flexport can have better data to create the supply chain optimizations it needs to.

  2. Very interesting read. Especially knowing that logistic costs are usually the biggest cost item for companies, Flexport could provide immediate benefits. I like your suggestion for Canada as the next hub but shouldn’t Flexport leverage some countries with more talent when it comes to engineering (like India) that have also cheaper labor? I think this could also help expand in Asia too with half the cost probably.

  3. Great post! This is an interesting problem tied to a huge addressable market. To the questions you posed: I think that with the continued development of open source machine learning libraries, the value lies more and more in owning the data rather than the ability to implement to implement a ML algorithm. In that sense, if Flexport can develop an interface for customers that allows them to streamline the data collection / labeling process, they should be able to maintain a competitive advantage. To the extent that they can own this from end to end, I would argue that they should.

  4. Really cool to understand how machine learning can be applied to an industry as traditional as shipping and transportation! I agree with you that the team should focus on building out their machine learning capabilities, although it is interesting to think about them leveraging the “AI-as-a-service” model. With the knowledge that IBM (and potentially other tech companies) are entering the space, my main fear that is they would be able to quickly utilize their own machine learning capabilities to replicate Flexport’s product and provide better insights to their end users. I imagine that developing ML capabilities in-house would allow them to (1) quickly identify insights that could serve their end users, (2) improve their ability to make the software as pain-free as possible for the data entry process, and (3) raise more funding to compete with the larger tech companies, given that ML tends to attract investors. I don’t think this necessarily requires them to choose between operations and machine learning. They could use their machine learning capabilities to improve their operations, attract talent, and potentially expand their capabilities for their customers.

  5. When I think of the meccas of shipping, Toronto does not immediately come to mind, but rather large port cities like Amsterdam, Shanghai, Busan etc where there already exist operational offices for Flexport. On the operation side, one could also consider the largest shipments by air, which in the USA come out of Memphis. However, when we consider that we want to bring machine learning capabilities to multiple offices around the world, it seems prudent to select a single office for this venture as opposed to opening additional ML capabilities everywhere.

    Toronto is quickly becoming well known for the AI industry with many companies opening up AI centers there, such as Samsung in 2018. The idea of separating an AI lab from the general operation of the core business is a strategy that is being followed by Samsung because that is where they feel the best talent is to be found. If Flexport believes in the long term benefit of using AI to supplement their business, it seems reasonable to consider the AI hub cities such as Toronto for a foothold.

    When it comes to core competency for Flexport, the operations need to be there first, and I would suggest Flexport consider additional operational offices where the largest air shipments are conducted and in tandem find the most reasonable location to for AI talent that can communicate with all the regional operational offices around the world.

  6. Very interesting article — and I do beleive that machine learning/ big data analysis’s role in logistic industries has become more and more important.

    Overall, I completely agree that data collection is the key challenge for leveraging machine learning in optimizing operation efficiency. Therefore, I’m a little bit confused about your suggestion on setting up and AI office. While I agree that it is critical for them to develop their own algorithm if they want to use machine learning to build up their core competence, I’m a little bit lost about how this initiative can help them solve the issue of getting clean data? I feel like focusing on operation and helping the partners in the network to upgrade their data collecting system might be a more straight-forward way to solve the challenge.

    And back to your first question, I do believe that machine learning expertise should be built in house because in the future the competition is based on who has better forecast capability. And you wouldn’t want to miss this critical skill.

  7. Very interesting read on AI application in shipping & cargo space. I also appreciate the thought on potentially looking for outsourcing model rather than trying to develop in-house solution. Would also be curious how does competitors place in this space and case example of a successful application of leveraging AI for shipping cost tracking & optimization.

Leave a comment