SPEC Research

The arbiter of AI edge for the world’s largest investors.

The State of AI Alpha

A quarterly research report on where AI is genuinely making money in investing, and where the edge is only being claimed. We read the market through published evidence and live fund data.

Published quarterly. The Q3 2026 edition is out now.

Open the report→
Accuracy on financial Excel tasks
Accuracy by workbook size
Easiest file
86.2%
Largest workbook152 cos · 8 funds
48.6%
Average across ten model configurations from OpenAI, Google and Anthropic. On the largest workbook, 152 companies across 8 funds, accuracy falls to 48.6%.
FinSheet-Bench, Ravnik et al. (2026) · arXiv:2603.07316
SPEC Test

An independent due-diligence test for a single AI investment strategy. Before you allocate, we check whether a manager’s edge is real or an artifact of how the strategy was built and measured.

S
Specification
Is the strategy conditioning on what it claims to?
P
Performance
Would the track record survive a journal referee?
E
Explanation
Can an independent judge reproduce the reasoning?
C
Configuration
Is the edge in the idea, or the layer it runs on?

Four falsifiable axes, each resolving to yes or no on evidence we can execute. Grounded in a catalog of 15 documented failure modes.

See how SPEC works→
AIMI, the AI Maturity Index

A public ranking of how ready the world’s largest funds are to actually run AI, scored on capability rather than on the capital they have announced.

27 sovereign wealth funds and 26 pension funds, ranked separately. Each cohort is weighted for how it actually uses AI.

Open the index→
AIMI score · sovereign wealth funds
Only GIC clears the 70-point Leaders band
GIC
72
ADIA
69
Norway GPFG
69
Temasek
67
Mubadala
63
NZ Super
49
cohort mean · 31
The top six of 26 scored sovereign funds. The cohort mean is 31 of 100, so most sovereign wealth funds sit well below this group.
AIMI v2, SPEC Research · scores as of 11 August 2026
About

SPEC Research is led by Milos Maricic, Principal.