This week it’s a quick retrospective, some learnings, and a look-ahead to two-dot-zero.

Intro

We’ve worked with some amazing people and companies over the last year.

When we started building, we knew one thing in our gut, putting LLM capabilities at the centre of data workflows was extremely powerful.

We were solving problems that had haunted us for years, and solving them very well.

Our earliest customers were delighted, and offered invaluable feedback that helped us shape and refine their experience.

But as anyone will tell you, working with AI is extremely complex. Our team is continuously challenged, and the knowledge around best practices is changing weekly.

Here is how we went from 1.0 to 2.0

The problems

We initially set out to replace complex copy and paste workflows in business lending.

We immediately realised that AI could not only simplify and accelerate credit analysts work, but could be the foundation for an entirely new way of running business lending teams.

In short, we discovered that we can unlock a growth oriented approach.

On the way, we’ve overcome business challenges such as:

• The broken telephone between broker, borrower, originator, underwriter, and portfolio manager. We learned speed and accuracy are essential.

• Providing alternatives to API extraction of Financial or Bank Data, which are met with resistance by borrowers, leading to piles of PDF and Excel docs. Not only do we remove that bottleneck, but we also unlock The Cross-Sell, increasing revenue with low risk and minimal overhead.

• Financial spreading and consolidating of non-standard financial statements, so that lenders can automate covenant compliance and monitoring. Getting this right leads to automated feedback loops.

• Bank statements issues, from justifiable irregularities through to bank errors all the way to tampering (across UK, US, and EMEA) which unlocks bank statement sanity checks to fight fraud.

Together, this is what enables lenders to Build a Lending Brain, just by improving their data extraction process!

What is great about these challenges is that they are often:

• not yet automated (as they are in Consumer and Sole Trader style business lending).
• existing solutions are extremely tedious to buy (if they exist at all).

What worked

For our 1.0 app, we knew we needed flexibility, we knew we needed to be able to move fast, and we knew that AI would move fast alongside.

Our early bet on data extraction as the most relevant capability started to pay off, and one of our early investments went into significantly expanding this capability for our vertical.

A quick recap of Challenges overcome and shared via this blog:

• Building fast feedback loops: This is at the core of everything we do these days!

• Turbo mode: High-volume documents that look simple and structured, but hide significant structural variance. Examples include long bank statements, invoices, and sales ledgers.

• Conversational data capture: Evaluating our conversational data capture and proactive lead nudges, when we only control one side of an evolving conversation with a range of loan applicant profiles. This powerful capability can reduce drop-off, reactivate churned leads and re-examine rejected leads.

• Context-Aware Confidence Score: Incorporating contextual information into confidence score, i.e., is this field correct given the context.

• Building trust with accurate citations and powerful financial validation pipelines, and a lot more, but obviously there are some insights worth hanging on to!

Technical development aside, finding the right product ergonomics for this capability was key.

A peek at the future

Over time, we ended up with a simple way of getting the most power to the users, with the minimal effort.

We’ve taken that idea, and ran with it!

Our 2.0 lending brain accelerates growth, enabling cross-sell and churned lead reactivation through intelligent and adaptive data capture.

1 - AI first workflows

We’ve done the work of finding out what works, and we have wrapped that with the simplest, most intuitive interface possible.

Our objective with this interface: giving the user access to all they need, including helpful onboarding and guidance, so that they could get going within minutes.

2 - Powerful feedback loops

At the core, we have a very powerful data processing and storage capability, allowing flexible document capturing, citation reviews, and feedback loops

This allows Fintechs to instantly get an AI native Deal system, with the simplest and most powerful lending brain available, with next to no engineering effort.

The Feedback capability, coupled with our increasingly powerful document analysis, leads to an extremely effective way to go from a new workflow to a refined and performant pipeline with minimal effort.

3 - Embeddable AI

What is possibly most exciting, is that we are now embeddable.

Lenders, Fintech lenders, and Fintech SaaS companies can use our app directly within their existing systems.

Within a day, you can solve the following engineering headaches:

1. Flexible, simple, powerful AI customisable schemas
2. Complex document extraction with citations and a PDF viewer
3. Customer facing chat, document extraction and validation
4. Customer FAQ, customer support and onboarding

We can most effectively solve this problem not by offering something end to end,
but rather by dropping a solution directly where it is needed (painkiller!)

This also unlocks the following business impactful capabilities:

1. Chat based underwriting in any language via Web Chat, Mobile, and WhatsApp
2. Real time origination, cross-sell and cashflow assessment using bank statements
3. Reactivation of churned leads with embeddable chat origination
4. Monthly monitoring with automated financial consolidation, spreading, and covenant checks

Next steps

All these features combined are what makes Sea.dev so powerful for Lenders.

Stay tuned for more. We are going live next week, so keep an eye out for socials, and if you want to get started, DM me!