I spent the first four years of my retail career at the Gap in merchandise planning and allocation. And I loved that job. I loved the puzzle of it. But I also remember dreading Monday mornings with a passion.
Why?
Because I had to work through spreadsheet after spreadsheet just to order every style of denim across 40 size options before I could even start my actual day.
It was a grind.
And here’s what haunts me nearly 30 years later. For most retailers, the process hasn’t changed all that much.
All of whihc is exactly what I wanted to dig into in my latest conversation with Jeff Fish, Co-CEO of Intelo.ai, and Noah Herschman, the company’s Head of Merchandise Strategy. Both Jeff and Noah come from deep enterprise tech backgrounds (Microsoft and Salesforce), and they’re now laser-focused on fixing one of retail’s most persistently “under-teched” (my quotes) functions.
What they told me was equal parts validating, surprising, and genuinely exciting, so here’s my breakdown.
Why Merchandise Planning Got Left Behind
Merchandise planning is complicated in a way that most enterprise software never fully reckoned with. It’s part art, part science. And the ratio shifts depending on where you sit, e.g. whether you’re a luxury buyer trying to anticipate next season’s color trends, or a planner at a fast-fashion chain managing a thousand SKUs across hundreds of stores.
Legacy systems could input data and produce outputs, sure. But what they couldn’t do was capture the instinct of an experienced merchant. That gut feel about what’s going to sell in which store, which trends are going to break, or how sizing is going to shift as consumer demographics change.
So what did planners and allocators do?
They downloaded everything into spreadsheets and worked their magic manually.
Every.Single.Time.
Noah put the stakes into sharp relief. Somewhere between 6 and 12% of total revenue can be impacted by merchandising and planning mistakes. Having run large-scale planning teams of 50-plus people at Target, I actually think that number might be low. And we have almost no visibility as an industry into the true cost of these errors because the data never gets cleanly connected to the outcome.
What Changed After 2022
The pre-2022 AI world gave us ML models for demand forecasting. They were better than nothing, but they were essentially black boxes. You put data in, an output came out, merchants didn’t trust it, and back to the spreadsheets they went. What changed with the generative AI era is transparency. When you interact with a modern LLM-powered planning tool, it it tells you why. It shows its reasoning. And that, as Jeff explained, is what builds trust.
“Since ChatGPT launched,” Jeff said, “there’s a level of comfort now that if you have an interaction with a model, it will reason with you.” That reasoning layer, the ability to say “here’s what I changed, here’s what I assumed, here’s why the margin moved,” is precisely what merchandise planners have always needed and never had.
The Agentic Difference
What Intelo.ai is building takes this a step further. Their system is built on an agent-to-agent protocol, meaning the agents aren’t just responding to human prompts in isolation. They’re also communicating with each other across the organization in real-time. A planner working on a merchandise financial plan, an allocator running in-season optimization, and a buyer building an assortment plan for next year can all be interacting with the system simultaneously, with agents sharing data across functions and carrying updates across the organization automatically.
Noah gave me a concrete example that hit home. Imagine telling a system: “Give me my financial plan for the next six months based on these parameters.” Then saying, “Actually, I need to get an extra half-point of margin. Can you adjust?” In the old world, that was a nightmare. It meant starting from square one, redoung every calculation, and hoping like hell that you didn’t anchor the wrong cell in Excel. In our new AI world, this can all happens nearly instantaneously. And the systems can also tell you exactly what they changed and why.
Where the Early Results Are Coming From
The results Jeff and Noah cited aren’t theoretical. One of their luxury retail customers has already seen a 47% reduction in stockouts. A specialty retail customer has recorded a 42% reduction in broken sizes. And deployments are happening fast, in as little as three weeks in some cases, with an average of six to eight weeks.
The inventory rebalancing use case particularly stuck with me. In luxury retail, you typically buy once for a season and then live with it. The margin on those items is high enough that shipping a piece from one underperforming store to a store where it’s more likely to sell is critical. Intelo.ai’s agents are detecting those opportunities before a human planner would even notice a problem developing.
Store clustering is another area where the math starts adding up quickly. Most retailers cluster stores by square footage. For exammple, A stores get A assortments, B stores get B assortments. But the reality is much more nuanced. Demographics, local demand patterns, and historical sell-through data all shape which products will actually move in a given location. Intelo.ai’s agents can process all of that data and create dynamic, data-driven clusters. Noah put it plainly, arguing that just modifying assortments slightly based on real data versus gut feel or a traditional planogram can be worth millions in either recaptured lost sales or more productive inventory.
What the Job Looks Like in Three to Five Years
I closed the conversation by asking both Jeff and Noah how the actual day-to-day job of a merchandise planner or allocator changes as this technology matures. Jeff’s answer was that no one went to school hoping to spend most of their career doing data analysis in spreadsheets. But for most planners and allocators, that’s what the job has become. The transformation the three of us discussed is about getting those people out of the data work so they can do what they actually came to retail to do, to think strategically, work closely with design teams, anticipate trends, and plan for growth.
Noah added something that honestly stopped me in my tracks right near the end of the interview. The fear is that AI replaces these jobs. But when you do the math, the revenue impact of improved planning (we’re talking 12% loss sales improvement) isn’t remotely in the same ballpark as the cost of keeping those roles on payroll. So, the epiphany I had, is that the right move for a retail executive isn’t to reduce headcount. On the contrary, It’s actually to use AI as the steroid to extract maximum value from your organization first.
Get that right. Prove the model. Then figure out what your workforce looks like on the other side.
That’s an insight I’m going to be repeating for a long time. And it applies well beyond merchandise planning.
To hear the my full conversation with Jeff and Noah, watch or listen to this episode of the Omni Talk Retail Technology Spotlight Series wherever you get your podcasts:
Apple Podcasts | Spotify | SoundCloud | Amazon Music | YouTube
Be careful out there,
– Chris, Anne, and the Omni Talk team
P.S. See our past 8 years of wonderful Spotlight Series podcast guests, featuring roughly 200 movers and shakers in retail, by clicking here
Music by hooksounds.com
*Sponsored Content*
Omni Talk® is the retail blog for retailers, written by retailers. Chris Walton founded Omni Talk® in 2017 and have quickly turned it into one of the fastest growing blogs in retail.