Finance Performance

Is the finance industry actually changing, or just talking about it?

Tightness Diagnostic — 3/5

Traditional KPIs are intact — NAV, IRR, Sharpe, TER, AUM are universally measured. What is missing is a shared scoreboard for the agentic shift.

Why 3/5 — Standard KPIs work. New KPIs (agent-vs-human Sharpe, research compression ratio, agent-to-agent settlement latency, model-vs-reality variance under agent execution) are unowned.
What is broken — The industry measures the old game and calls the new game "innovation theatre". The metrics for the agentic era have not been published, so they cannot be reported, so the shift looks slower than it is.
What to teach — Reporting drives behaviour. Define the missing metric before the auditor does.

The numbers set the frame. Everything else is relative to this.

Finance is the largest industry by asset value. Every percentage-point shift moves trillions.

These are the signals that the edge is collapsing between institutional and retail.

Analyst time: Earnings call analysis — 2 days (junior analyst) → 5 minutes (AI plugin)
Research access cost: Bloomberg terminal — $24,000+/year → free AI plugin
Coverage capacity: 10 stocks/week (manual) → 50+ stocks/week (AI-assisted)
DCF build time: 2 days → minutes (same arithmetic; AI pulls and structures inputs)
Quant idea to production: 4 meetings + a sprint → one prompt to the agent → hours
Trading hours: Session-bound exchange + 300 staff to run 24/5 → 24/7 native rails with code

These are the metrics that would let an analyst — or a regulator — see the agentic shift in real time. Most are unreported.

Metric	What it measures	Why it matters
Agent-vs-human Sharpe	Risk-adjusted return of agent portfolios vs human PMs	First clean test of "are agents actually better?"
Research compression ratio	Analyst hours pre-AI / analyst hours post-AI	Productivity dividend — the labour-cost lever
Weekend volume share	Volume traded outside session hours / total volume	Proxy for agent participation; humans sleep
Stablecoin settlement share	Stablecoin settled value / SWIFT settled value	When this crosses 10%, correspondent banking unwinds
Agent-driver coverage	% of production agents with a named accountable human	Governance gauge — unaccountable agents are the risk
Memo-to-decision latency	Time from intent expressed → IC decision recorded	Compresses 10× under agent drafting

Signals that predict future structural change:

Retail active portfolio size (increasing = AI lowering research friction)
Micro-cap trading volume (increasing = AI enabling coverage of uncovered stocks)
Bloomberg terminal subscriber count (declining = institutional edge eroding)
DeFi protocol revenue (increasing = on-chain rails gaining trust)
Stablecoin settlement volume as % of SWIFT volume

Signals that confirm change already happened:

Retail investor alpha vs benchmark (measures if faster research = better decisions)
Bank revenue from correspondent banking (declining = stablecoin displacement)
Junior analyst hiring in equity research (declining = AI automation absorbing workflow)
RWA tokenization total value locked

Bounds that signal systemic risk if breached:

DCF model accuracy: fair value estimate within 20% of consensus (AI models on bad assumptions are still bad)
Stablecoin reserve transparency: full backing required or systemic risk rises
AI agent autonomy without human sign-off: guardrail on binding commitments
Agent-driver coverage below 100% in production: orphan agents are the next operational-risk class
Reed's-law compounding without provenance: if every agent learns from every other agent, the loop needs a traceable record or the failure mode is invisible until it is systemic

Which of these signals is already showing up in your portfolio — or in your workflow?

Where is the largest gap between AI-augmented institutional research and retail capacity?
At what stablecoin volume does SWIFT correspondent banking become structurally unviable?