Advantage Solutions Grew Profit 116% While Its Closest Competitor Shrank 10%. The Real Advantage Was Built Three Years Before the AI.

THE CRAFT

Dave Peacock waited three years to write his first letter to shareholders. His explanation, in the letter’s opening sentence, is the most honest thing I’ve read from a public-company CEO this year: he felt it was best to be “further along in the transformation of Advantage Solutions before writing.” Most CEOs write the letter first and hope the transformation catches up. Peacock ran the sequence in the other order.

You’ve probably never heard of his company, and that’s the point of it. Advantage Solutions is the commercial infrastructure of American retail — more than 60,000 people working in 90 percent of U.S. zip codes, keeping 4,000 consumer brands stocked, merchandised, and sampled across roughly 85,000 store locations. When a shopper tries a sample at a warehouse club, when a store aisle is rebuilt so a new product is on shelf by morning, that’s usually Advantage. Nobody knows the name. Every retailer and every brand does.

It is also a business where the entire P&L is labor. Seventy million paid hours a year. In a company like that, margin isn’t made in the boardroom — it’s made in whether the right person, with the right training, shows up to the right store at the right hour, seventy million times. Which makes Advantage the purest test case I’ve found for the question every leader is being sold an answer to right now: what does AI actually do for a business built on people?

Here is what the AI pitches will not tell you. When Peacock arrived in early 2023, inheriting a leveraged roll-up whose profitability had fallen 16 percent, his first moves had nothing to do with AI. He sold fourteen non-core businesses. He consolidated a patchwork of systems into one. He built the data plumbing — ingesting, cleansing, separating the company’s information for actual use — and he shut down roughly 30 percent of the legacy systems that were running when he started. Then he centralized how the company hires, schedules, and deploys its enormous workforce. The AI came after all of that, layered onto workflows the company finally understood.

The result is sitting in plain sight in the segment numbers, and almost nobody has written about it. This brief is about the three unglamorous years that made the multiplier work — and what the one segment that got the full rebuild first can teach you about where your own margin is hiding.

THE OPERATOR

The situation

Advantage Solutions is what happens when the outsourcing logic of the 1990s meets the consolidation logic of private equity. Built through decades of acquisitions — it traces to a Southern California food brokerage in 1987 — the company became the biggest player in a business most people don’t know exists: the outsourced field force of consumer-goods retail. Three segments. Branded Services sells and merchandises on behalf of CPG manufacturers. Experiential Services runs the nation’s largest in-store demonstration and sampling operation — more than 40,000 teammates who executed over 4.5 million live sampling events last year. Retailer Services works the other side of the aisle: private-brand development and the field teams that executed 2.7 million planogram resets across 80 percent of U.S. grocery locations, often overnight.

The economics are unforgiving. This is an asset-light, people-heavy services model that earns a dime or less on the revenue dollar, carrying roughly $1.6 billion of debt from its roll-up years — more than four times its annual operating profit, with interest payments consuming close to half of it. Every basis of competition in this business reduces to the same question: how much value does each paid labor hour produce? When Peacock — a former president of Anheuser-Busch — took the job in early 2023, the answer had been going the wrong way since 2022.

And the clients were squeezed too, which matters because a squeezed client cuts its outsourced services first. The consumer landscape Peacock describes in his letter is the hardest CPG operating environment in years: tariffs pushing imported-goods prices roughly four points above trend, a consumer splitting in two — thrift-store traffic up 11.7 percent while mid-tier department stores fell 6.2 percent — and GLP-1 medications quietly rewriting food demand itself. Twenty-three percent of U.S. households now include someone on one of those drugs, and those users spend about 31 percent less on groceries. J.P. Morgan puts $30 to $55 billion of annual traditional food-category revenue at risk by the early 2030s. Squeezed this way, some large CPG clients pulled work in-house or exited entirely — and those losses landed on Advantage’s highest-margin work.

So picture the company Peacock inherited: a magnificent, invisible machine — 70 million labor hours touching nearly every store in America — running on decades of accumulated systems that didn’t talk to each other, with profitability falling, debt heavy, and its client base under structural attack. The obvious 2023 move, the one every vendor was selling, was to announce an AI initiative. He didn’t.

The move

The rebuild came in layers, and the order is the lesson.

The first layer was subtraction. Fourteen non-core businesses sold since 2022. More than $400 million of debt paid down, maturities pushed out to 2030. And inside the technology estate, the move almost nobody makes: Peacock’s team decommissioned approximately 30 percent of the legacy boundary systems that were in place when he arrived. Before the company added a single intelligent tool, it removed a third of the sediment. Every AI conversation you will have this year starts with what to buy. This one started with what to shut off.

The second layer was consolidation — the plumbing. The company collapsed its enterprise systems onto one platform, a migration that finished only this May, when the last major business unit went live. It stood up a common data foundation, with what Peacock describes as real capability for “ingesting, cleansing and segregating our data for better uses.” A new human-capital system arrives in 2027 to complete the modernization. None of this is exciting. All of it is the difference between a company that can be understood and a company that can only be described.

The third layer was the one that touches all 70 million hours: the centralized labor model. Historically, hiring, scheduling, and deploying field labor lived business unit by business unit — every group running its own version of the same work. Peacock centralized it: one model for how the company recruits, trains, schedules, and puts people in stores. The early, measurable effects showed up exactly where a labor company feels them first — the cost of hiring someone fell meaningfully in the first quarter of this year, by the CFO’s account, and the speed from hire to working shift improved.

Only then — on top of consolidated systems, a cleaned data foundation, and a centralized labor model — came the AI. It shows up in the company’s own capital-spending list not as a moonshot but as infrastructure: AI-based labor scheduling and deployment. Tools that decide, faster and better than a regional office juggling spreadsheets ever could, which of tens of thousands of field workers should cover which store, which event, which reset. AI woven into hiring — Peacock told analysts in May that AI-enabled staffing and scheduling tools “are already improving our speed and labor utilization.” An alert-based deployment system that reads demand signals and points sales reps at the stores where a brand’s opportunity is largest this week, rather than working a static route list. The company built over 5.3 million in-store displays last year; it is now transitioning that whole motion from coverage-by-habit to coverage-by-signal.

Notice what the sequence bought them. Because the labor model was centralized first, an improvement to scheduling propagates across the whole company instead of dying inside one business unit. Because the data was cleaned first, the AI schedules against reality — actual demand, actual skills, actual history — instead of against seven conflicting spreadsheets. This is why I keep telling you the order matters more than the tools: the same AI, bought two years earlier, would have automated the chaos.

The result

Advantage’s headline numbers for 2025 are not a victory lap, and Peacock doesn’t pretend otherwise. Net revenues of $2.998 billion, down 1.5 percent. Operating profit — the adjusted measure the company manages by — down 6.8 percent. His letter says it flat: “Our results fell short of plan and fell short of what this company is capable of achieving, which is my responsibility.”

The story is in the segments — because the rebuild didn’t land everywhere at once, the segments are a controlled experiment. Experiential Services, the events business where the centralized labor model and the AI scheduling landed first and deepest, grew revenue 8.3 percent to $1.02 billion and grew profit 34.1 percent, with margins expanding nearly two full points to 9.9 percent. Here’s the arithmetic that should stop you: event volume grew 5 percent, and profit grew 34 — profit growing roughly seven times faster than volume. That gap is not demand. That gap is the machine getting better at converting a paid hour into an executed event: the execution rate — events staffed and run as contracted — hit 94 percent in the first quarter, up year over year, and the cost of putting a new hire into the field fell. Meanwhile the two segments where the rebuild landed later both declined — Branded down 21.1 percent in profit, Retailer down 11.6 — under the same macro, the same management, the same balance sheet. Same company, same year, one variable moved.

The first quarter of 2026 says the gap is widening, not reverting. Experiential grew revenue 22.8 percent and more than doubled its profit against the prior year, on event growth near 20 percent and improving execution. Total company profit rose 16.4 percent on 5.8 percent revenue growth. And the cash discipline the rebuild was supposed to produce showed up on schedule: collection time fell from 64 days at the start of 2025 to a record 57.5 by December, the company converted 67.3 percent of its operating profit into free cash flow for the year — better than its own guidance — and paid down another $131 million of debt in the first quarter alone.

For the outside benchmark, look at SPAR Group, the closest public comparison in merchandising services — a much smaller company, about one twenty-fifth Advantage’s size, but working the same aisles for the same kinds of clients. In the first quarter, when Advantage’s events business grew 22.8 percent, SPAR’s revenue fell 10.3 percent and it swung to a loss. The tide is not lifting this category. Something inside Advantage is doing the lifting.

The second-quarter numbers came out this morning, and they sharpen the split rather than settle it. Revenue of $889.5 million was up 1.8 percent — about what analysts expected. Company-wide profit fell 12.2 percent. And underneath that decline the pattern held for a third straight period: the events business grew revenue 19.7 percent and profit 32 percent, while Branded’s profit fell 36 percent and Retailer’s fell 25. Of the three periods, this is the one I’d put the most weight on, because of what it’s measured against. The first quarter’s doubling ran against a weak prior-year comparison, which flattered the rate. This quarter ran against the strongest profit quarter of the prior year — and the rebuilt segment still grew by a third. Management left its full-year ranges where they were, with total profit still guided flat to down.

Now the honest take…

This is still a four-times-levered company reporting a net loss under standard accounting, guiding total profit flat to down for 2026, whose highest-margin segment is bleeding clients it may not get back. This morning’s numbers made that case harder, not easier: profit fell 12 percent at the company level and the net loss more than doubled, to $62.7 million. And some of the quarter’s cost improvement came from moving employees onto an outside provider’s payroll rather than from scheduling them better. Experiential’s surge also rides a genuine tailwind — retailers are structurally investing in in-store experience. Management itself won’t put a dollar figure on what the AI specifically contributes and points to 2027 as the year the technology benefits arrive in force. The bear case writes itself: a transformation narrative bridging a hard year.

What the bear case still can’t explain is the spread — or the record collections, or the cash conversion. Across all three periods, the segment that got the rebuild first grew profit by more than 30 percent every time. Branded, which never got it, fell every time.

That’s not a narrative. That’s a control group.

The Craft of AI read

Strip away the retail specifics and here is what I’d take if I ran a company where labor is the product — which, if you run a distributor, a carrier, a field-services firm, a healthcare operator, or frankly most businesses, you do. Every AI pitch in your inbox right now leads with a thing to buy — an agent, a copilot seat count, some platform with .ai in the domain — and it arrives before anyone selling it has understood how your work actually happens at the level of the people doing it. The tech lands first; the understanding is supposed to catch up later. Peacock ran the inversion: three years of making the company knowable — systems consolidated into one, data cleaned, one labor model, a third of the legacy systems gone — before asking software to make decisions. He fixed the workflow, then applied the multiplier, and the one segment that got the full sequence is the one whose margin turned.

There’s a second discipline here that gets no press because it’s the opposite of press: Advantage refuses to out-claim its own transformation. Analysts asked management in May to quantify the technology benefit; they declined and named 2027 as the realization year. Today’s earnings release doesn’t use the word AI once. The company’s realized numbers — the 34 percent, the 94 percent execution, the 57.5-day collections — are published and auditable, and its projections are labeled as projections. Compare that with the vendor deck on your desk, where the projected savings arrive with more confidence than the deployed ones. When a company’s claimed numbers and filed numbers are the same numbers, you can build on what it says. That’s not modesty. That’s what being early actually looks like from inside — less impressive per quarter, more impressive per decade.

And note what the sequence protects him from. Because the rebuild landed segment by segment, the P&L can see it — the board isn’t taking the transformation on faith, it’s reading Experiential’s business outcome. Because the spending is funded from operations while the company pays down debt, there’s no war chest to defend and no sunk-cost trap. The discipline compounds exactly the way PITT OHIO’s twenty minutes back per driver, per day did: quietly, arithmetically, in the results themselves, whether or not anyone writes a press release.

The transferable lesson is not “this takes three years.” It’s the order, and the honesty about where you are in it. If your data isn’t “ready for AI” — if pointing intelligent software at your schedule today would multiply guesses — then this year’s AI budget should buy understanding and subtraction, not seats.

That’s not falling behind. On the evidence in this brief, it’s the only version of ahead that shows up in the numbers.

Things to consider

• Your labor schedule is a P&L document. If people are a major cost line, the schedule is where your operating margin is actually decided — not in pricing meetings. Advantage runs 70 million paid hours a year; one percent of that, captured or wasted, is seven hundred thousand hours. Run your version: total annual paid hours, times loaded cost, times one percent. That’s the size of the prize sitting inside a workflow most leadership teams have never once examined.
• Subtract before you add. Peacock shut down roughly 30 percent of the legacy systems he inherited before layering anything intelligent on top. Ask your team what was decommissioned in the last twelve months. If the answer is nothing — if every initiative adds a system and none removes one — you are accumulating the sediment that will eventually make AI unworkable, and paying to maintain it in the meantime.
• Deploy where the P&L can see it. The rebuild landed first in one segment, so the results are visible, attributable, and defensible in a board meeting — 34 percent profit growth against two declining sister segments is an argument nobody can wave away. A capability spread thin across the whole company produces improvements everywhere and evidence nowhere. Pick the unit where the before-and-after will be undeniable, and finish it before you generalize.
• Make vendors separate deployed from projected. Advantage labels its projections as projections and lets its filed numbers carry the story. Demand the same structure from anyone selling you AI: two columns — results realized at named clients, and results projected for you — and watch how much of the pitch migrates to column two. Fund column one. Treat column two as a hypothesis you’ll test on your own numbers.
• A down year is not a reason to wait — it may be the reason to start. Peacock ran this rebuild while revenue declined, clients defected, and interest consumed half his operating profit — funding it from operations, not from slack. The rebuilt segment is now what’s holding the company up. If you’re waiting for a comfortable year to examine how your work actually happens, notice that Peacock didn’t get one and did it anyway.

THE WORKBENCH

Do this tomorrow

This one takes an hour, a notepad, and one team’s schedule.

Shadow the person who builds the schedule. Pick one operational team — dispatch, field service, store staff, production shifts — and sit with whoever assembles next week’s schedule. Count three things while they work: how many systems and spreadsheets they touch, how many pieces of information they need but have to guess at (actual demand, individual skills, travel time, history of what happened last time), and how many minutes the whole cycle takes. You are mapping the gap between how the schedule is made and what the schedule decides. In most companies the person deciding where hundreds of paid hours land each week is working from two spreadsheets, a group text, and memory.

Then price your business case. Take the team’s total weekly paid hours and estimate, from what you just watched, the share placed by habit or guess rather than by information — most teams land somewhere between a tenth and a quarter. Multiply by loaded hourly cost and 50 weeks. Write the annual number down. That is operating margin being spent today on scheduling decisions made blind — and it’s the number that tells you whether your next technology dollar should buy scheduling software or should first buy the demand signal, the skills data, and the clean history that any scheduling intelligence would need to beat the guess.

THE QUESTION

Dave Peacock closed his first shareholder letter with a disclosure I’ve never seen a CEO make: he used AI to help research and draft the letter itself, said so plainly, and then drew the line that matters — “the positions, commitments, and decisions in these pages are mine.” From the CEO of a 60,000-person company mid-rebuild, that’s the posture the whole story runs on: use the multiplier everywhere it helps, and never let it substitute for knowing what’s true. The transformation letter most CEOs would have written in year one, he wrote in year three, when the business’s own numbers could carry it.

So here is the question, and I’d ask you to answer it in under thirty seconds, honestly: if you pointed AI at your labor schedule (or any other inefficient workflow) tomorrow morning, what would it find to work with — a demand signal, clean skills data, real history — or seven spreadsheets and a group text? Peacock’s answer in 2023 was the second one, and instead of buying the AI anyway, he spent three years earning the first. The companies that skip those years don’t skip the cost. They pay it later, at the multiplier’s exchange rate.

---

Want this done for you?

You just ran the one-team version with a notepad.

My Ground-Up Workshop starts with the people in your business who do the work. One to two days in person, a synthesis week, and you walk out with a design: a ground-truth map of how your core workflows actually run, the one where redesign releases real operating margin, a target operating model with AI built in from the ground up, and a 30-60-90 day plan your own team can execute. I do a small number of these a quarter for $20,000.

I’m an investor in geniant. For more than 25 years, our craftspeople have done one thing: understand how work actually happens, craft how it should happen, then build software that works the way humans do.

When a workshop gives you the roadmap to execute or turns into a build, they’re who does it — one senior-led team, layered on top of your systems of record, in weeks rather than months.

Book a discovery call →

— Grant K. Baldwin