Labor Day weekend closes out and the phone stops ringing quite so hard. Somebody says “finally, a chance to breathe.” Then you open the drawer.

Four months of receipts that never got coded. Vendor statements nobody reconciled, because reconciling in July was never going to happen. Seasonal staff who need to be offboarded, final-checked, and, if they were any good, remembered well enough that you can find their number in March. Reviews from June still unanswered. A shoebox of email addresses from people who came once and would probably come back if you ever wrote to them.

None of that is a crisis. That’s exactly the problem with it. It’s work that never becomes urgent enough to fix and never goes away either, and in a business whose revenue arrives in a fifteen-week window, it lands all at once.

The hiring you did in May is still on your desk

Staffing is where the seasonal admin bill is largest and least visible. Nationally, the number of employed 16- to 24-year-olds rose by 1.8 million between April and July of this year, a 9.1% jump in one quarter. In July, 25% of all employed young people were working in leisure and hospitality. That is the summer workforce, and in a region where the lakeshore, orchard country, and the campgrounds off the interstate corridors all staff up at the same time, it’s a substantial share of who’s behind the counter.

Michigan’s own numbers show what that churn costs in paperwork. In the most recent state figures from the Bureau of Labor Statistics, leisure and hospitality posted the largest quarterly job gain of any Michigan industry. That was a net increase of 7,478 jobs, produced by 44,820 gross job gains against 37,342 gross job losses.

Six jobs gained for every one that stuck. Each of those gains is an onboarding packet, an I-9, a payroll setup, a scheduling account, a training run, a set of logins, and then the same sequence in reverse. The net number is the economy. The gross number is your September.

What gets automated isn’t what actually hurts

New Census Bureau research on how AI is actually being deployed inside American firms found that 18% of firms used AI in a business function between November 2025 and January 2026. Among the firms that did, 57% used it in three or fewer functions, and the functions they picked were Sales and Marketing (52%), Strategy and Business Development (45%), and IT (41%).

Look at what isn’t on that list. Not intake. Not scheduling. Not reconciliation, not onboarding, not the follow-up sequence to the customer who came in July. The places AI has landed first are the places that were already interesting to work on. The seasonal crunch lives in the places that were always boring.

Retail is instructive too. As of May, retail trade businesses reported AI use of about 14%, against a national rate of 19.8%. The sectors that deal with the public in person, on a schedule set by the weather, are furthest behind, and they have the sharpest peaks.

Four places the deferred hours sit

The fall backlog is almost always in the same four rooms. What differs is which one is actually costing money.

• Operations The season that was never written down Opening and closing procedures live in one person’s head, so every May is a re-teach. Writing the season down while it’s fresh, which means now, is the cheapest thing on this list.
• Accounting & Finance Four months of coding, all at once Receipts, vendor statements, payroll for staff who already left. Automated capture and coding won’t make the fall close pleasant, but it turns a dig into a review.
• Customer service Intake is where the season is won The inquiry that arrives at 9pm in July gets answered Thursday, or not at all. Capturing the right details the first time and routing them to a human before they go cold is the highest-return automation most seasonal businesses skip.
• Marketing The list you never built Every peak season generates contact with people who would come back. Most of it evaporates because nobody had ninety seconds to capture it.

It does not replace your seasonal crew

Worth being clear about, since it’s the first worry when a shrinking labor pool is already the constraint. That same Census research found that 66% of AI-using firms rely on it solely to augment tasks, and that AI-related employment decreases occurred in only 2% of firms. That matches what the work looks like on the ground. Nobody automates the person who greets customers. They automate the twenty minutes that person spends re-typing an order into a second system.

When it isn’t the answer

Some of this shouldn’t be built, and we’d rather say so early than bill for it.

If your season is genuinely fourteen weeks and the deferred admin is genuinely forty hours a year, a system to handle it costs more than the forty hours. Do it by hand in October and spend the money elsewhere.

Maybe the process is about to change: a new POS next spring, a different booking platform, a partner buying in. Automating the current version is waste. Wait until it settles.

And if the backlog exists because the sequence is wrong rather than because it’s manual, automation just makes a bad process faster. If three people enter the same customer detail into three systems, the fix isn’t a bot. It’s deciding which system is the one that’s true.

The shoulder season is the build window

You cannot change how work moves while the work is moving. In July nobody can spare an afternoon to watch the intake process, let alone alter it. In the shoulder months the process is idle, the people who run it are still around, and the season is recent enough that everyone remembers exactly where it broke. That combination exists for maybe ten weeks a year.

So the useful version of this is small. Pick the one thing that hurt most this season; for most operations that’s intake or the fall close. Ship a working fix before the holidays, documented well enough that your team runs it without us. Not a six-month roadmap. One thing, working, in hand before the next season starts.

The alternative is the drawer, again, next September.

A note on the numbers: adoption surveys disagree with each other because they ask different questions. “Used AI in the past two weeks” and “used AI in a business function” measure different things, and gross job flows and net employment change measure different things again. Every figure above links to its primary source so you can check the definition yourself. We keep a running Sources page grading each citation by its distance from original research.