Wavemaker Impact · Venture capital · AI · Southeast Asia · 2025
Scaling investment due diligence without scaling the analyst team.
A climate-focused venture fund needed to scale its investment diligence without scaling the research team at the same rate.
I was contracted as an AI Principal to redesign the diligence workflow around AI and implement a production-grade AI system to augment their teams.
Within months, AI agents were deployed and allowed a 25% increase in due-diligence throughput.
~2 months
Forward-deployed POC
3 months
POC to production
+25%
Due-diligence throughput
Consistent AI quality
Compared to non-AI diligence
The situation
Investment diligence was becoming a constraint on fund scale.
Wavemaker Impact evaluates climate ventures across a wide range of industries and markets. Before committing capital, each opportunity requires substantial research, market analysis and investment due diligence.
This manual work created a structural constraint: doing more deals meant doing more diligence, which meant more staffing.
That heavily constrained their ability to scale.
The challenge
Due diligence is repeatable enough to augment, but too judgment-heavy to automate.
Some parts of diligence appear repetitive: estimate a market, map competitors, summarize interviews or assess an industry's carbon footprint.
But the underlying questions change substantially from one venture to another.
The challenge was therefore to augment the investment process without reducing complex investment judgment to a rigid template.
01
Every opportunity required a different analysis.
A climate software company, an industrial technology business and a new materials venture could not be assessed using the same research path.
The system needed a repeatable analytical structure while remaining flexible enough to investigate very different businesses and markets.
02
Important conclusions depended on judgment.
Market attractiveness, competitive positioning and venture quality could rarely be determined from one metric.
Analysts had to interpret incomplete evidence, weigh conflicting signals and decide which questions mattered most. AI therefore needed to support judgment rather than replace it.
03
Analysis was difficult to compare across deals.
Different analysts naturally approached research differently.
Sources, assumptions, notes and conclusions varied between diligence projects, making comparisons harder and creating lengthy discussions about what constituted a sufficiently attractive opportunity.
04
Throughput remained tied to analyst capacity.
Research, meeting synthesis, market sizing and competitor analysis consumed substantial analyst time.
Even with better processes, the number of simultaneous diligence projects remained constrained by the amount of analytical work the team could perform.
The solution
An AI diligence layer built inside the investment workflow.
Instead of starting with an AI architecture, I started inside the investment process itself.
I embedded myself with the due diligence team, working alongside members to understand how decisions were made, which tasks consumed time, and where better or faster analysis could move the needle.
We then built and tested the system on a weekly cadence against real diligence work.
01 · Forward-deployed into real diligence
Learn the investment process before automating it.
My first priority was to understand how the team actually performed diligence.
I worked directly within active investment projects: following research, reviewing evidence, observing how analysts developed hypotheses and seeing where judgment was required.
That made it possible to identify work that AI could genuinely accelerate from work where automation would add complexity without improving the investment process.
02 · Prioritize the bottlenecks
Automate what materially increased diligence capacity.
There were many possible AI use cases.
Instead of attempting to automate everything, we prioritized the work that was both time-intensive and sufficiently repeatable to benefit from AI.
This included market sizing, field note synthesis, adversarial research, industry benchmarking, meeting synthesis, carbon-footprint estimation and cross-checking research.
Lower-value opportunities were deliberately put on the shelf.
03 · Build with the investment team
Weekly usagereplaced long specification cycles.
New capabilities were introduced into active diligence projects on a weekly cadence, allowing analysts to review AI outputs while the underlying investment questions were still fresh.
Their feedback immediately exposed weak assumptions, missing evidence and outputs that looked useful technically but were not useful to an investor.
That shortened the loop between building, using and improving the system.
04 · Making AI trustworthy
Specialized agents owned specific analytical jobs.
I intentionally built a multi-step agent pipeline rather than relying on a single general-purpose AI assistant.
Different specialized agents were responsible for distinct analytical tasks, while others were specifically designed to be adversarial or verify other AI outputs.
Outputs could be traced throughout the AI pipeline, making outputs inspectable and trustworthy.
The business outcome
More diligence capacity without proportional headcount growth.
The result was akin to having an team of AI agents working in parallel with the investment team.
AI agents could research, synthesize and cross-check information in the background while investors concentrated on the work where their time was most valuable: developing investment theses, meeting founders, conducting field work and exercising judgment.
+25% diligence throughput
The investment team estimated that this allowed 25% more due-diligence work with the same existing team.
The productivity gain came from reducing time spent on research-heavy and synthesis-heavy activities rather than removing analysts from the process.
More analyst time on primary research
Routine analytical work increasingly happened in parallel with the team's activities.
This allowed investors to spend more time on founder conversations, stakeholder interviews, field investigation and challenging investment assumptions rather than assembling information manually.
Production deployment in 3 months
The initial forward-deployed proof of concept was built in 2 months.
The system then moved into the client's GCP environment within 3 months, turning the experiment into an operational capability rather than a standalone AI prototype.
Fund growth without matching headcount growth
When the fund eventually doubled in size, it was able to do so without doubling its due-diligence headcount, achieving the scalability objective that had motivated the project.
The AI diligence operating system
Many diligence questions, one analytical system.
Every investment required a different investigation. The solution was to build a common orchestration layer connecting specialized AI workflows with human investment judgment.
The diagram below shows a stylized and simplified version of the system.
AI did the legwork so that investors could focus on using their human, expert judgment.
By making research, synthesis and cross-checking happen in parallel, the system gave the team more time for developing investment theses, testing assumptions and speaking with the people closest to each opportunity.
That is what made the diligence capability scalable: AI augmented the analytical work, while the investment team remained accountable for the judgment.