Cutting a 6-7 day investment review down to 90 minutes with an AI agent workflow

The challenge

Kula is an investor that finds remote companies around the world that may otherwise struggle to secure investment. Once a target company is identified, that company fills out a detailed application covering its business, the investment it needs, and a wide range of supporting documents.

Reviewing that application properly meant checking it against Kula's own investment policy across several distinct areas, including KYC, compliance, legal and financial due diligence, then researching each applicant company individually to verify what had been submitted.

All of that sat on one person's shoulders. Every application meant six to seven days of manual research, cross-referencing policy documents by hand, and writing up findings, before Kula could even begin deciding whether to invest.

That single-person bottleneck capped how many applications Kula could realistically review at once, and repetitive research was being redone from scratch on every application, with no way to draw on what had already been checked before.

Our approach

1. 01

   Mapping the manual review process

   We worked through Kula's existing review process end to end, capturing exactly what a KYC, compliance, legal and financial reviewer each checked, which parts of Kula's own policy applied, and how the final decision was reached from that research.

2. 02

   Designing the agentic workflow

   We designed an AI agentic workflow where each application and its supporting documents are pushed through a series of skilled agents, with each agent focused on one review area, KYC, compliance, legal and financial among others, mirroring how Kula's own specialists approached the work.

3. 03

   Grounding agents in research and policy

   Each agent was built to research the internet for information on the applicant company and check the application against Kula's own investment policy, rather than relying on the application form alone, so findings are backed by evidence rather than assumption.

4. 04

   Building traceable, drillable summaries

   Individual agent findings are rolled up into a single, readable summary Kula can drill down from, with references built in throughout so any point raised can be traced straight back to its source, whether that's the applicant's own documents or external research.

5. 05

   Connecting the Microsoft 365 and AI stack

   We built the workflow across Power Automate, Power Apps, Power BI and Power Pages, connecting Kula's application intake through to the agentic review, and brought in OpenAI, Claude and Relevance.ai to run the agents themselves.

The outcome

What used to take one person six to seven days now takes around 90 minutes, with the AI agents working through KYC, compliance, legal, financial and other checks in parallel rather than one reviewer working through each area in sequence.

Reviews are now more detailed and broader than the manual process allowed, with duplicated research eliminated and every finding fully traceable back to its source, rather than resting on one reviewer's notes.

Kula no longer has a capacity ceiling on how many applications it can review at once, and the workflow runs at a lower cost than the manual process it replaced, expected to pay for itself within six months.