The AI That Pays Is Hiding in Your No Joy Work

Jamie Dimon says AI will bring huge efficiency to certain parts of JPMorgan. Their annual report already told us which parts.

The AI that pays is rarely new work. It is old work somebody has been doing by hand.

Jamie Dimon said this morning that JPMorgan fully expects AI will bring huge efficiency to certain parts of the company. Certain parts. Not all of it. He also said the bank is running nearly a thousand AI use cases¹. So which parts? That is the whole question, and most companies are figuring this out.

Turns out JPMorgan already answered it, in their own annual report. Over 80% of their wealth management front office uses the bank's internal AI. And here is how the report describes what they do with it. Every day, they look for opportunities to eliminate what they call "no joy" work and repetitive tasks².

No joy work. That is the whole strategy, sitting in a filing.

Why It Matters

Look at where it shows up. JPMorgan's chief analytics officer demonstrated the AI building an investment banking presentation in about 30 seconds, work that used to take a junior banker hours. Their advisors use it to read earnings transcripts and compare financial documents, and the bank says they now handle significantly more clients. A cash flow tool the bank built cut its corporate clients' manual work by close to 90%.

None of that is new work. All of it is old work somebody had been doing by hand.

And it shows up everywhere. In wealth management. In the pitch deck. In the research a seller does before a call. The pattern isn't a department. It's the antiquated process, wherever it happens to live.

The Three Moves

1. Ask the team what they hate doing. They will name it in about a minute, because they have been doing it by hand for years. Account research, reading filings and news before every call. Pitch and proposal prep, rebuilt from scratch every time. Well-Architected reviews run by hand. Security and compliance checks. Routine cloud operations and audit reporting. Monthly reporting, rebuilt by hand for every account, every month.
2. Check it against the pattern. The work AI fits looks the same everywhere. It repeats, so there is a pattern to learn. The rules are written down somewhere. You already know how long it takes, so the baseline is sitting there waiting. A person can check the output before it counts. And nobody would miss doing it.
3. Pick the number before you build. Name the business number this improves. Capture the baseline before launch, which is easy here because you already know how long the work takes today. Then set the date you check it. Do that and the go or no-go conversation settles itself, because the data answers it for you.

What To Do Monday

Walk over to the team doing the work and ask one question. What is the thing you do every week that you wish a machine would do for you? Write down the top five answers. Score each one against the pattern.

Then start with one small job and keep a person checking the work. Prove it pays. Add the data and the guardrails, so the safe way is also the fast way. Then let the person who built it show the rest of the team. Edelman, who has tracked global trust for 26 years across 28 countries, found people trust their own coworkers more than they trust any institution³.

Last week I shared MIT's number: 95% of enterprise AI pilots produce no measurable return⁴. The 5% that worked didn't find a better model. They found the no joy work.

Finding it is the first move. The second one is what you owe the team once the work goes away, and that is a bigger conversation than most leaders are having.