The AI returns gap is a data problem.
Bain finds 951 firms spending more on AI while returns stall. The blocker is reaching your own data, not the model. Only 7% run agents unattended.
Hi folks,
Bain said the quiet part out loud this week: companies are spending more on AI and mostly not getting the returns, and the top blocker is not the model, it is getting at their own data. That is the exact problem we spent the week building around for a fund.
The one thing
The returns gap is a data problem, not a model problem
- Bain surveyed 951 large companies. Most targeted 10-20% cost savings and landed at 0-10%, and only 7% run AI agents unattended in production (Bain).
- The No. 1 blocker is data access and integration, cited by 41%, ahead of compliance, budget and skills. The best performers cite it most, so it is not a maturity problem that solves itself.
What this means in plain terms:
- The constraint is reaching your own data and trusting the output, not which model you buy.
- Start with one workflow where the data is reachable and the answer is checkable. Get a win there first.
- Bain's line, worth keeping: "AI doesn't fix workflow debt; it locks it in, speeds it up, and makes it vastly more expensive to unwind."
In the mix
- Anthropic filed for an IPO (CBS News)
- Confidential draft S-1, the first frontier lab to start the public-markets process, on a $965B valuation and roughly $47B run-rate.
- Why it matters: the vendor-durability question conservative funds keep raising just got firmer. The S-1, once public, puts a lab's real economics on the record.
- Apollo and Blackstone are arranging ~$36B to finance Anthropic's chips (Yahoo Finance)
- A special-purpose vehicle buys Google's chips and leases them back to Anthropic, with Broadcom backstopping residual value. Closing early June.
- Why it matters: private credit is now funding the AI buildout directly. A real asset-backed structure, and a signal of where the largest credit allocations are heading.
- KPMG was named Anthropic's preferred partner to put Claude into PE portfolio companies (The Next Web)
- It is also rolling Claude to all 276,000 staff.
- Why it matters: the top of the market is productising "AI into portfolios". The mid-market version is what a partner does without the scale to build a captive team.
From our week
- Built the knowledge base (the "brain") for a fund
- One place the team asks a question and gets a sourced answer, instead of hunting across systems for something that already exists.
- Built the evals that make the answer trustable
- Automated checks grade whether the agent cites its source, opens the actual document rather than guessing off a snippet, and never silently recomputes a figure.
- This is the Bain gap in practice: an answer a GP can act on, not just a plausible one.
- Built the adoption and feedback layer
- A privacy-safe dashboard (metadata only, kept in Australia) to see whether the team actually uses it, and a feedback loop so the agent improves when an answer is flagged wrong.
- Hearing the same thing from other funds
- We are comparing notes with other managers on what good looks like, and the pattern repeats: the information exists, it is just scattered. The first win is making it reachable and checkable, before automating anything on top.
Levercon is the AI team for investment funds. We embed alongside your team to automate workflows, build custom tools, and compound operational gains. Ultimately we help Australian based investment funds leverage AI.
Book a strategy call.
See where AI fits in your fund.
We'll show you where AI lifts efficiency and performance across your organisation.