Where Our AI Work Stands

We’ve spent a lot of time this year looking at AI.

Not because we think every business needs an AI strategy, and not because we’re interested in adding another service just because it’s getting attention. We’ve been looking at it for a much more practical reason: there are a lot of repetitive, time-consuming processes inside accounting and finance, and some of them should be easier than they are.

So we started testing.

We’ve been building AI-supported workflows inside a handful of real client engagements, using real documents, real vendor information, and real deadlines. That has taught us quite a bit about where these tools are useful and, just as importantly, where they aren’t.

We’re not ready to package it up and call it a new service yet.

Here’s why.

The technology isn’t really the hard part

One of the more interesting things we’ve learned is that getting the software to do something isn’t usually the biggest challenge.

The hard part is everything that comes before it.

A purchase order has one item number and the vendor acknowledgment has another. A project name is entered three different ways. Freight terms change halfway through a job. Someone agrees to a change order in a conversation, but it never makes its way into the system.

People are pretty good at working around those inconsistencies. They know that two slightly different descriptions refer to the same thing because they’ve worked with that vendor for five years.

Software doesn’t have that same context.

So much of our work so far hasn’t been about installing an AI tool. It has been about cleaning up the process around the tool so the information going in is reliable.

That may not be the exciting part of AI, but it’s probably the more important part.

We’re interested in the boring stuff

There are plenty of impressive AI demos out there.

We’re more interested in the tasks nobody would ever put in a demo.

Matching vendor acknowledgments to purchase orders. Coding invoices to the right project. Keeping track of deposits paid and received across multiple open jobs. Making sure changes that happen during a project actually make it back into the financial information.

Individually, none of these tasks sounds particularly important.

Do them hundreds of times a month, though, and they start consuming a lot of someone’s time.

Those are the places where we think there’s an opportunity.

Not to replace the person doing the work, but to take some of the repetitive work off their plate and give them more time to deal with the things that actually require judgment.

Better information is the real goal

For us, the interesting part isn’t necessarily the automation itself.

It’s what becomes possible when the underlying information is cleaner and more current.

Cash forecasting is a good example.

A useful forecast needs to know more than what’s sitting in accounts receivable today. It needs to account for open commitments, upcoming vendor payments, customer deposits, project timing, and other things that may not show up neatly on a standard report.

If those inputs can stay organized and current without someone rebuilding a spreadsheet every Friday, the CFO has a much better starting point for making decisions.

That’s where we see the bigger opportunity.

So why haven’t we launched anything?

Because we’re still learning.

Before we put a formal service in front of clients, we want real examples of where this works and what kind of value it actually creates.

The tools are also changing quickly. What makes sense today may look different six months from now. We’d rather spend some more time working through that than rush to put a name and a price on something before we know exactly what we’re offering.

We’ll get there when we think there’s something worth offering.

In the meantime, we could use your help

If there’s a process inside your business that takes hours every month and seems to produce the same result every time, tell us about it.

Maybe it’s something your team has done manually for years. Maybe it involves a spreadsheet everyone hates. Maybe it’s one of those processes that works, but nobody really knows why it has to take as long as it does.

Those are the things we want to look at.

Bring it up the next time we talk or reply to the newsletter and tell us what it is.

The boring problems are usually the interesting ones.