Boring Always Wins: Why Your First Transformation Project Should Be the Back Office

Last month I wrote about kAIos, the special chaos of AI tools that reprice, vanish, or get switched off mid-project. That post was about the vendor side of the mess. This one's about a decision that's entirely in your control, no matter how the vendor landscape shakes out: what backlogged project do you point AI at first.

Most CFIs get this backwards. They start with the exciting thing, a chatbot, a lending copilot, something a board member saw a demo of at a conference and now wants by Q3. That's a fine instinct if you're a bank with a nine-figure tech budget and a bench of engineers to babysit it. It's a rough way to spend your one shot at proving AI works if you're a $500 million community bank or a credit union with 40 employees and an IT department of two.

The data backs this up

A recent American Banker analysis looked at AI spending against headcount data across more than 21,000 companies. The finding: operations was the only job category that failed to see headcount growth at companies making heavy AI investments. Everything else, sales, engineering, customer-facing roles, kept growing. Operations didn't.

Standard Chartered is cutting 7,800 back-office roles by 2030 as it automates. At JPMorganChase, operations and support headcount shrank while client-facing roles grew 4%.

I know what you're thinking: great, more JPMorgan and Standard Chartered stats, thanks that’s very relatable for my small shop. Fair. But flip it around. These are institutions with very large budgets that can experiment and throw AI at anything: marketing, lending, trading, wherever. They tested it everywhere, and the back office is where it paid off first. That applies just as much to an $800 million bank as it does to a trillion-dollar one. It's a free answer key for the rest us, and it happens to point at the one place a community bank or credit union can actually afford to run a pilot without a team of data scientists.

What does improving the "back office" mean to you?

This isn't about anything exotic. It's about the stuff nobody at your institution would defend if you asked "does this really need a human doing it by hand?"

Re-keying new account and loan paperwork into the core system instead of pulling it straight off the scanned document is one example. Reconciliation is another: matching transactions and flagging only the actual exceptions instead of eyeballing every line. So is assembling the same recurring compliance report from data that's already sitting in three different systems, sorting incoming documents into the right queue, and cleaning up data entry that got fat-fingered the first time around.

None of it touches a member. None of it requires judgment. All of it is currently being done by a person who trained for something better than this, probably one of the same three or four people who also cover a dozen other jobs at your institution, because that's how staffing works when you're not a bank with 300,000 employees.

The real reason this matters isn't the cost savings

Cutting cost is fine. I'm not going to pretend it isn't part of the pitch. But if that's the whole pitch, you're missing the better argument.

The people doing this work right now are capable of more than data entry. The American Banker's piece makes a point worth coming back to: a lot of the best relationship bankers and lenders in this industry started in back-office and ops roles. That's where they learned how the business works before they ever sat across from a member or a business owner.

This matters more for a community bank than it does for a megabank, not less. A national bank can absorb a few hundred bored, underused employees without anyone noticing. A community bank or credit union can't. Every person on staff is already wearing three hats, and the member sitting across the desk from your loan officer can tell the difference between someone who has time to listen and someone who's racing back to finish yesterday's data entry.

How do you identify the first project?

Not every back-office task deserves to go first. It needs to happen often enough that fixing it moves a number, and the steps need to stay the same from case to case. A new hire should be able to follow a written checklist and get it right. And you need to be able to name one number, hours, errors, turnaround time, before you start, so you have something to compare against after.

If a task requires judgment on nearly every case, or it's the kind of thing where a mistake lands directly in front of a member, that's not your pilot. Save it for round two, after you've built some track record and some trust internally.

If more than one task clears that bar, pick the one everyone already complains about, the backlog that shows up in the Monday staff meeting or the report that always runs late. That's the one where a fix is obvious to the whole institution, not just to you, and that's what gets you buy-in for the second project.

You don't need a data science team for this

This is the part people talk themselves out of, usually right after reading a story about a megabank's AI budget. You don't need that budget, and you don't need to build anything from scratch. Most core and loan origination platforms already have automation add-ons for exactly this kind of work, and there are financial-services-specific tools built just for document intake and reconciliation, priced for a bank your size, not for Chase. The hard part was never the technology. It's picking the right first target, running a small pilot, and being honest about whether the numbers moved.

If you've already tried automating something in the back office, tell us whether it worked or turned into a bigger mess than the manual process. We want to hear it either way. Follow Shore Group on LinkedIn and share your story.