There's a lot of noise about AI in retail right now. Most of it is vendor pitches and abstract think-pieces. Very little of it tells a planner what AI can concretely do for their job this week.
A quick grounding before specifics. Tools like ChatGPT, Claude, Copilot, and Gemini are good at language work. They can read, write, summarize, analyze, and explain at the level of a smart, well-read assistant who is occasionally overconfident. They're getting better at quantitative reasoning.
Out of the box, an AI tool doesn't know anything about your business. Ask it a cold planning question and you'll get a generic answer that sounds plausible and tells you very little.
That gap is what "skills" and "MCP" close. Both terms are starting to crop up in organizations everywhere. Here's what they actually mean.
What is an AI Skill, and How Does it Work?
New to AI? Start here. A skill is just a saved set of instructions. Think of it like a Standard Operating Procedure, except instead of handing it to a new analyst, you hand it to the AI. The AI follows it every time.
A skill is a packaged set of instructions (and sometimes scripts and resources) that an AI tool loads on demand to do a specific job well. Anthropic introduced the term for Claude, but the concept exists across every major AI platform under different names. Skills you build aren't locked to one tool.
Skills In Plain Language
A skill teaches the AI how to do your version of a task. Not a generic category scorecard. Yours. The format, the cutoffs, the flagging logic, the tone, the rules of the road. Built once, reused every time, by anyone on the team.
Skills come from three places:
- Built-in skills that ship with the AI tool itself, like creating Excel files or PowerPoint decks
- Custom skills your team writes for your workflows, usually in natural language, no code required
- Partner skills built by software vendors and published in a directory, designed to work with that vendor's data through MCP (more on that next)
A planning skill might look like:
- "Generate a weekly category scorecard from this data, formatted like our standard one, flagging any category off-plan by more than 3 points."
- "Run an OTB summary for next month, broken out by division, and call out any divisions where exposure exceeds plan by 10%."
- "Check current inventory against forecast and surface the top 10 SKUs at risk of going out of stock in the next 14 days."
What Using Skills Gets You in Practice
The work is repeatable (whoever runs the skill gets the same output). It's auditable (the instructions are written down, not living in a planner's head). And it's consistent (your scorecard looks like your scorecard, every time).
The AI only loads a skill when the task calls for it, so stacking up more skills doesn't slow anything down or clutter the AI's responses. It's there when you need it, out of the way when you don't.
Projects, Skills, and Prompts and How They Fit Together
If skills are new to you, it helps to understand how they sit inside the broader structure of how AI tools work. There are three layers, and each one does a different job.
A project is your persistent workspace; a named environment where context, files, and conventions live between sessions. Upload your category plan, your comp sales data, your markdown thresholds. Everything inside the project knows about each other.
A skill is your repeatable playbook for a specific task: the format, the flagging rules, the expected output. It lives in the project and runs whenever you need it.
A prompt is what you type right now. It can be a one-off question or it can trigger a skill. Either way, it's how you talk to the AI in the moment.
The point isn't to learn the terminology. The point is that once you have a project set up with the right context and a few key skills built, you stop retyping instructions from scratch every week. The setup pays for itself in the first month.
Terminology Varies by Platform But the Concept is the Same
Whether you're using Claude, ChatGPT, Copilot, or Gemini, this structure exists. The names differ. The logic doesn't.
What is MCP (Model Context Protocol)?
A skill teaches the AI how to do the job. MCP gives the AI the data to do the job on.
MCP stands for Model Context Protocol. It's an open standard that lets AI tools connect to a data source and pull live information from it. So when you run that "weekly category scorecard" skill, the AI isn't guessing or working off whatever you copy-pasted in. It's reading from your planning system in real time.
Skills plus MCP is the unlock. The skill is the how, MCP is the what. Together they take the AI from "smart language tool that doesn't know your business" to "actually useful in a planning workflow."
What Retail Planners Can Do With AI Today, No Integration Required
You can start using AI for real work this week, before you touch MCP or skills. Things any planner can try in a free Claude or ChatGPT account:
Draft a Weekly Category Writeup
Paste your numbers in. Ask: "Write a 200-word summary for my Monday business review covering the top three winners and bottom three losers in this data, and call out any categories that flipped from positive to negative this week."
Clean up Messy Data
Paste in a vendor's sales report that's formatted weirdly. Ask: "Reformat this into a clean table with these columns: Style, Color, Units Sold, AUR, Sell-Through %." It'll do it in seconds.
Pressure-Test Your Assumptions
Paste your forecast methodology and last season's actuals. Ask: "What's the most plausible reason these forecasts ran 8 points high? List five hypotheses ranked by likelihood."
Translate Between Teams
You're explaining a markdown decision to finance. Paste your reasoning in. Ask: "Rewrite this in language a CFO would respond to. Lead with the margin impact." Done.
None of these are transformational, but they're very useful, and they're available right now, with no integration, no IT involvement, no contract.
What Changes When AI Can Connect to Your Planning Data
The examples above all require copy-paste. You're moving the data into the AI by hand.
When MCP is in the picture, that step goes away. The AI can pull the data itself. So instead of "paste this report and ask for a summary," it becomes "ask Claude what categories are tracking under plan this week," and the answer comes back live, from your actual system, with your actual numbers.
That's where the practical use cases get bigger:
- A category review that took 90 minutes to build takes 5, because the AI is pulling the data and drafting the writeup in one pass.
- A finance partner asks the AI what your division's open-to-buy exposure looks like and gets the answer without bothering anyone in planning.
- You're in a reorder call and need current sell-through velocity on a top style. You query it directly from the dashboard you're already in.
What AI Still Can't Do For Retail Planners
It doesn't replace your judgment. It pattern-matches well, but it doesn't know:
- What your buyer is hearing in the showroom this week
- That a vendor's lead times have been slipping for two months
- That your CEO has a strong view on a category and will push back on this plan
- That last year's spike was a one-off promo that won't repeat
It's also wrong sometimes, confidently. You'll need to check its work, especially on numbers, the same way you'd check an analyst's first draft.
Used well, AI doesn't replace the planner. It removes the parts of the job that were mostly typing, formatting, and retrieval, and frees up time for the parts that aren't.
How Retail Planners Should Start Using AI
If you've never used Claude or ChatGPT for planning work, start there this week. Try the four examples above. The fastest way to understand any of this is to actually use it.
That's the small version of what AI changes for planners. Less typing, less formatting, faster turnaround on the writeups and back-and-forth that fill the week.
The bigger version is what happens when an AI tool can read your planning data directly, through MCP, and the rest of your organization can pull answers without routing the request through you. The work your team owns starts to shift, away from retrieval and toward judgment. And that's where your role gets more strategic.
