Scale AI, a startup providing the infrastructure for AI applications headquartered in San Francisco, recently announced it had raised $325 million in a Series E funding round co-led by Dragoneer, Greenoaks Capital and Tiger Global. The company has raised more than $600 million since 2016 and is now valued at $7.3 billion.

The company also announced a partnership with Flexport, a leading global trade platform, to simplify logistics paperwork processing through machine learning and AI.

Freightwaves interviewed Scale AI’s general manager of document AI, Melisa Tokmak, to discuss the importance of machine learning within the supply chain industry and how the company is improving Flexport’s processes.

This question-and-answer interview was edited for clarity and length.

FREIGHTWAVES: Why is shipment data important and how is it used to create machine learning models?

TOKMAK: “Data is powerful. You can use it to accelerate your shipment processing to increase output, reduce delays at customs, improve cash flow, and build new products. Better data is crucial for getting a detailed understanding of what happened in the past, know what is happening now, and be better prepared for the future. 

“Logistics companies are sitting on top of a treasure trove of data buried in millions of different documents, but cannot use it to its fullest extent. If you can unlock that data with high quality document processing, that’s the beginning of everything you can do with AI. 

“Think about AI as a pyramid. The bottom layer of the pyramid is all about data. Without structured data, you can’t leverage AI or improve business workflows. At Scale, we help logistics companies unlock this data trapped in documents by taking a machine learning approach.” 

FREIGHTWAVES: How have you helped Flexport structure their document data?

TOKMAK: “In Flexport’s case, their mission is to modernize global trade, and one of the challenges they had to deal with is extracting information from all of their logistics paperwork. How do we extract the data, but also do it accurately and quickly to ultimately impact downstream business metrics? 

“So what Scale AI has done is build machine learning models that are adjusted for Flexport’s own use cases so that the team can process all of their documents at much faster and at higher quality, minimizing the need for manual efforts. Even better, the Flexport team does not need to create any templates for every new document type. Scale’s approach is robust to variability and can process previously unseen document layouts. 

FREIGHTWAVES: There are companies that can help with documentation management; what problems does Scale AI fix that others are not?

TOKMAK: “There’s a lot of different document types, including commercial invoices, bills of lading, air waybills, arrival notices, etc. The problem is that many of our customers weren’t getting the results they were looking for because solutions in the market today are designed for perfect documents and that just doesn’t exist in the real world.

“Sometimes information is crammed. Sometimes it’s very hastily submitted by a scan or photo. Sometimes it’s a very complicated document because you need to extract descriptions, hard-to-read fields, individual unit prices, etc. So every time I talk to people, I always hear that everything they have tried did not work. On top of that, their teams had to spend all this time trying to manually create templates that became outdated when their vendors or document layouts changed.”

FREIGHTWAVES: That is interesting because if a document is not perfect, theoretically it would lead to more transit issues, correct?

TOKMAK: “Yes, it causes a lot of issues that can impact transit. One of the most important things you care about as a logistics company is to prevent unnecessary delays at customs that can impact downstream steps in moving the goods and decrease customer satisfaction. For example, unnecessary container openings.

“That is why it’s so important to have that end-to-end document processing workflow. When we set up our machine learning solutions, the system is built to handle constantly changing document formats and layouts. Every time we identify something that is not right or a human fixes something, that information goes back and re-trains the models to continuously get better over time. You want to be able to learn from the bad examples.”

FREIGHTWAVES: What is problematic about having humans helping with these processes?

TOKMAK: “Efficiency. People are very valuable, because there’s so many things in your business that the people need to do, whether it’s dispute resolution or direct customer communication. So if you can take away a big bulk of this tedious manual work off of their plates, they can focus on areas that need to be resolved in a short amount of time. That really creates a lot of cost efficiencies and also employee retention.”

FREIGHTWAVES: Why would companies not invest in this type of deep data learning in-house?

TOKMAK: “It’s hard! Not only the setup and the cost of it, but even finding the best engineers is really hard. We are a machine learning company and we attract the best of the best talent when it comes to that. 

“But as you said, there are companies who want to invest in some in-house machine learning themselves and go all in. We also work with these companies directly, to help augment their internal machine learning teams to build the best machine learning models to extract key information from documents and support their new product initiatives. 

“Flexport is one of them. They want control over how they build things, but still want a technical partner to go on this journey with them.”

Click here for more articles by Grace Sharkey.

Related Articles:

Flexport: Trans-Pacific deteriorating, brace for shipping ‘tsunami’

Flexport goes upstream with new Order Management suite

FreightTech 25: Flexport demystifies trade with more visibility tools