Cambridge published a report this month showing 76% of large financial institutions cannot measure the value of their AI deployment. That same week, six US banks posted Q1 earnings. They cut 15,000 jobs and reported profits up 18% year-over-year. All six cited AI. Measuring AI value is apparently not that hard. You just have to be willing to read the number that actually changed.
This month, I am covering that story alongside two others: all three major model companies making the same product bet in the same ten-day window, and US regulators quietly creating a gap in model risk governance that nobody has a good answer for yet.
Free section is my take on all three.
The paid section has three items that will help you with the How.
A prompt to map your own agentic use cases.
A monitoring setup so your LLM tracks regulatory developments for you.
And a function-by-function breakdown of where AI displacement is most likely to land in your finance team, and what to do about it before it happens.
Claude in Action
I wasn't sure how fast this would fill.
The corporate track is down to one spot.
The mixed cohort opens May 26, and seats are moving.
If you've been sitting on this, now's the time to decide. Both tracks are the same three sessions, same material. The difference is whether your whole team goes through it together or you join a group of peers from other companies.
If you need a push: pricing for both tracks is promotional through the end of May. I'm on vacation in June, and when both tracks reopen in July, pricing goes up.
Three News Clusters in April
The agent push.
Three updates, one message.
Claude Opus 4.7 shipped April 16. GPT-5.5 shipped April 23. Google launched its Gemini Enterprise Agent Platform and AI research agents on April 22. All three landed in a ten-day window.
The more interesting part is what each of them was selling. Not reasoning scores. Not context windows. Agents: tools that plan work, move across systems, and finish tasks without you staying in the loop.
Claude Opus 4.7 is 64.3% on SWE-bench Pro, the benchmark most closely tied to real-world autonomous task performance. That's the highest score in the industry right now, and a 10.9-point jump from the previous version. It specifically outperforms on agentic tasks, where it leads GPT-5.4 by over 9 points. Vellum has a clean breakdown if you want the full picture.
So what: Your primary LLM matters less than you think right now. What matters is whether you're using any of them agentically. Most finance teams I talk to are still in one-step mode. You paste something in, you get something out, you move on. That still works. But the model companies have all decided, in the same month, that the next phase is agents. If you have not started thinking about what recurring tasks in your function could run without you managing each step, that conversation is overdue.
Regulatory moves on both sides of the Atlantic
On April 17, the Federal Reserve, OCC, and FDIC jointly issued revised model risk management guidance, replacing the framework that had governed bank model risk practices since 2011. Generative AI and agentic AI are explicitly excluded from scope. The agencies said they plan to issue separate AI-specific guidance. No timeline was given.
In the EU, August 2 2026, is THE DATE. That is when the AI Act reaches full enforcement. Credit scoring, loan approval, fraud detection, and AML risk profiling are all classified as high-risk AI systems. Non-compliance penalties reach 3% of worldwide turnover for high-risk violations. A Digital Omnibus proposal could push some deadlines to 2027, but that is not confirmed.
So what: Two different postures. In the US, regulators carved generative AI out of existing model risk rules and said something new is coming. That gap is real. You cannot rely on the 2011 framework to cover your AI systems, and you have nothing official to replace it yet. In the EU, you have a hard deadline and specific obligations. If you have AI systems touching credit or AML and you have EU exposure, mapping those systems against the Act's high-risk requirements is a Q2 task, not a Q3 one.
The research says measuring it is hard. The banks say it shows up in earnings.
On April 28, Cambridge Centre for Alternative Finance published its 2026 Global AI in Financial Services Report, produced with the BIS, IMF, World Bank, and WEF. It covers 628 organizations across 151 jurisdictions and is the most credible independent data I have seen this year on where AI in finance actually stands.
Top finding: 4 in 5 financial services firms are deploying AI at some level. Second finding: the impact to date is on efficiency, not on how the business actually works. And 76% of large financial institutions say they find it difficult to measure the value of their AI deployment.
That same week, six US banks reported Q1 results. JPMorgan, Citi, Bank of America, Goldman Sachs, Morgan Stanley, and Wells Fargo cut 15,000 jobs in Q1 while posting collective profits up 18% year-over-year. All six cited AI. Bank of America credited AI directly for 1,000 job cuts. Citi disclosed it is paying Anthropic, Google, Microsoft, and OpenAI to automate legal document review, trade invoicing, and account approvals.
So what: These two stories are not contradictory. Cambridge says 76% of large institutions cannot measure AI value. The banks figured out how to measure it. The difference is not the technology. It is that the banks stopped asking about AI adoption rates and started reading the answer off their income statement: fewer people, same output, higher margin. If you are still trying to build an AI ROI dashboard, you may be asking the wrong question.
The pattern across all three clusters this month is the same: the tools are moving faster than the governance, the governance is moving faster than most teams, and the results are starting to show up where they cannot be ignored.
In the subscriber section, I get into the how.
A prompt to map your own agentic use cases.
A monitoring setup so your LLM tracks regulatory developments for you.
And a function-by-function breakdown of where AI displacement is most likely to land in your finance team, and what to do about it before it happens.
Closing Thoughts
That is the April roundup. A lot moved this month, but the through-line is the same: the gap between teams that are acting and teams that are still watching is widening.
If you try the agentic use case prompt or set up the monitoring task, hit reply and let me know what came back. See you next Tuesday.
We Want Your Feedback!
This newsletter is for you, and we want to make it as valuable as possible. Please reply to this email with your questions, comments, or topics you'd like to see covered in future issues. Your input shapes our content!
Want to dive deeper into balanced AI adoption for your finance team? Or do you want to hire an AI-powered CFO? Book a consultation!
Did you find this newsletter helpful? Forward it to a colleague who might benefit!
Until next Tuesday, keep balancing!
Anna Tiomina
AI-Powered CFO
1