EssayAn article by Mohammad Rasouli, summarized
How AI Platforms Are Changing the Game for Alternative Investment
Published in Middle Market Growth, the magazine of the Association for Corporate Growth, on January 24, 2024. Our summary of the argument, with a link to the original.
The argumentWhat the article says, in short
From scattered tools to one platform.
Funds are moving from fragmented, third-party AI tools to coherent AI platforms: one place that gathers the fund’s own data and serves every use case through a single interface. Large funds already have them. Smaller funds are still learning about them and need outside help. Large language models have made this affordable for smaller funds.
The article’s warning is that the largest risk is taking no action. Preparing infrastructure, cleaning historical data, learning how the fund wants to use AI and changing the culture all take time, so early movers stay ahead.
Why sourcing firstWhy deal sourcing and deal insight come first
Eight reasons to start with deal sourcing.
Large funds already use AI here to find returns, not only to process documents. The article gives eight reasons.
- A competitive edge
- Efficient discovery: more ground covered, faster
- Better quality of analysis
- Relevance tailored to the fund’s own strategy
- Analysis of non-traditional data
- Faster decisions
- Scale without a proportional rise in resources
- Continuous improvement as the platform learns
Why separate tools fall shortFive failure points of tools that work alone
Five places fragmented tools fail.
Not tailored to the fund
Generic tools do not reflect the fund’s thesis or way of working.
No access to the fund’s own data
So they never learn from the fund’s past decisions.
Privacy and security exposure
Sensitive data goes to third parties.
Fragmentation
Tools built for one task do not talk to each other, and building many is costly.
Generic output
Results lack the depth that high-stakes decisions need.
Seven stepsThe article’s approach for a fund that builds a platform
Seven steps to a platform.
Step 01: Define requirements and features
Adaptability, personalization, privacy and forward compatibility.
Step 02: Pick priority use cases
High return and relatively low risk, such as deal sourcing and insight; then extend across the investment cycle.
Step 03: Plan the data
Public databases, the fund’s proprietary data and confidential target-asset data, fed continuously.
Step 04: Choose models and algorithms
A mix enhanced with language models, matched to tasks and strategy.
Step 05: Decide who builds it
Outsourcing gives speed and cost efficiency; building in-house builds the fund’s own capability.
Step 06: Cost it in three layers
Infrastructure, training on public data and personalization. The first two can often be shared across funds; personalization is unique to each fund and carries most of the value.
Step 07: Manage risk, train staff, integrate
Review privacy and compliance regularly, train people, and connect the platform to existing systems.
PrivacyThe article’s three basic privacy practices
Three practices to keep data safe.
- Train on proprietary data only on a machine kept on the fund’s own servers.
- Train a separate model, with clear boundaries, for each asset covered by its own NDA, and delete it when the NDA period ends, so data from one asset never informs another.
- For highly sensitive data, such as HR records, search the documents instead of training a model on them.
OriginalThe full article, on the publisher’s site
Read the article
Published by Middle Market Growth. This page is our summary, not a reprint.