A quick dive into how our bet on flexible AI primitives is paying off in enabling agentic workflows.

I’m super excited by how our bet on primitives is paying off. We were super deliberate from the early days, and it’s now shining.

Key are three components, working in synergy.

1. Data Extraction Engine

First, our Data Extraction Engine. This is the AI that extracts and analyses information from documents, spreadsheets, images and raw text into an analyst-ready format. This includes the UX, workflows, and feedback loops to improve the process. Getting this right once is hard.

Getting it right in production and at scale is technically demanding, but doing it in a way that keeps underwriters in the loop and comfortable requires synergy between Product, Engineering, and Science (go team!).

2. Flexible Extraction Workflows

Second, our Flexible Extraction Workflows. This is the “target” for the Extraction Engine. Unique is our AI that allows users to create these workflows on-the-fly for emerging business lending opportunities. Beyond that, workflows can be adapted over time as requirements change, with flexibility to add or remove requirements without breaking things.

This version-controlled lifecycle management is something you just don’t see, and it’s because of our strong data science and data engineering that we have this natural inclination to data quality. (We also have extensive pre-built templates for most SME and Commercial loans).

3. Data Value

Finally, Data Value. We’ve found that the best way to make AI sticky is to make it useful. The way we do that is to be the best place to find correct answers, and the easiest place to answer a new question. Our chat agents can access data and compare with source documents, run scenarios, compare actual to forecasts, triangulate multiple versions of documents, uncover inconsistencies and enable the type of thinking that is so hard to automate at scale.

We’ve extracted high quality data, we’ve got the documents fresh from the borrower, what is critical is turning that into insights that drive competitive lending!

Bringing it together

Each alone is a technical engineering challenge, each contains a massive amount of nuance that overwhelms teams. Our commitment is delivering all 3:

• Engine: data extraction finely tuned to business underwriting, with carefully balanced capabilities and UX to keep humans in the loop.
• Workflow: perfectly designed for the intended purpose, and for the Agents assigned to them. Flexible, intelligent, versioned.
• Data: we care deeply about making data valuable. It needs to be accessible to other AI agents, humans and to legacy systems alike.

Reliable control planes are how it comes together.

I’ve spent years watching underwriting teams drown in documents and manual processes. Building something that actually fixes it has been a delight.

I’ll share a video soon with an update!

If any of this resonates, happy to show you what it looks like in practice.