Your store sleeps. Your agent doesn’t.
NorthAIsle is an AI agent trained in commercial intelligence. Every Monday it tells you how much money you’re losing, where, and what to change.
For supermarkets, pharmacies, hardware stores, electronics, pet supplies, beauty and any other retail sector.
$95k to $238k a weekEstimate for $1,200M in annual sales. A planning assumption, not a result.
Your shelf is losing money. You can't see it.
Nine findings are hiding on this shelf. They are the kind the agent surfaces every week: a price above the market, a promotion that loses margin, a stockout on its way.
In a business selling $1,200M a year: $95k to $238k a week. Estimate.
0 of 9 findings, $0 this week
Move the light across the shelf.
I’m going to tell you where you’re losing money.
Monday, 7:00. Your business writes to you.
By Monday at 7:00 the decisions are waiting: what to change, what it’s worth and why. Nothing moves until a person approves it.
Your team opens the console.
This is the working software, running on Fieldstone Market, a fictional supermarket chain. The data is synthetic; the calculations are real.
Sell something else? See the same agent on your shelf$0
Where each SKU sits against the market
Each dot is a product. Further right, dearer than the market. Higher up, shoppers react more to price. Pick one and test a new price.
Did the last eight promotions make money?
Net incremental margin, after everything a gross-lift report leaves out.
What each bar means
- Margin on extra units
- Units above baseline, at the promo margin. Negative when the promo price is below cost.
- Subsidized base
- The discount given on units shoppers would have bought anyway.
- Cannibalization
- Margin lost on sister products that shoppers switched away from.
- Basket halo
- Margin from extra items in baskets that included the promoted product.
- Post-promo dip
- Sales lost in the weeks after, as shoppers work through stock they loaded up on.
- Vendor funding
- What the supplier paid per unit sold on promotion.
Same agent. Any shelf.
That was a supermarket. The method doesn’t change with the sector: price, promotion, range and availability are read the same way. What changes is what the agent knows about your shelf.
Synthetic examples. Estimates, not results.
Nothing changes in your store until a person says yes.
- The arithmetic is code.
- Elasticities, baselines and promotion breakdowns are calculated, never guessed by a language model. Run them twice and you get the same figure, so your finance team can audit every number.
- A person approves every decision.
- The agent recommends, shows its math and says what could go wrong. Then it waits. You set the limits it works inside.
- It runs in your cloud.
- Azure, Google Cloud or AWS, inside your own account. Your data stays where it already is.
- Your data never trains a shared model.
- What the agent knows is the method. Your numbers only feed the calculations for your business, and what is fitted for you stays with you.
One agent, six specialties.
- Price positions key value items and moves regular prices.
- Promotions decides which events to repeat, fix or drop.
- Assortment prepares delists, adds and facings for line reviews.
- Availability catches stockouts before the weekend does.
- Market Watch turns competitor noise into the moves that matter.
- AI Shelf measures the new shelf inside AI assistants.
Six weeks. Your own stores will tell you if it worked.
A pilot is set up like an experiment and agreed with you before it starts. Nobody asks you to trust a number on a website.
-
Data intake
Week 1
We map two years of sales, costs and your promotion calendar as they are. You do not clean anything.
A data health report.
-
Backtest
Week 2
The agent remakes your last 52 weeks of decisions. You see where it would have disagreed with your team, and what happened next.
The readout, and the rules signed off.
-
Live decisions
Weeks 3 to 5
A brief every Monday for the categories in scope. 20% of your stores carry on as usual, as the control group.
Three briefs and a record of every approval.
-
Readout
Week 6
Measured impact, adoption and a rollout plan sized on your own numbers, presented with your finance team.
The business case for rollout.
What you give
- Two years of weekly sales by product and store
- Unit costs
- Your promotion calendar, with supplier funding
- One sponsor, and one category manager per category in scope
What you get
- Three Monday briefs on your own data
- Every decision with its money and its math
- A plan for what comes next, whatever the result
How you will know it worked
Recommendations go live in test stores while matched control stores carry on as usual. The result is the difference between the two.
It is reported with confidence intervals and presented with your finance team, along with how many recommendations your team approved, changed or rejected, and why.
Size it for your business.
Change your sector, your sales or any assumption. The ranges are planning assumptions, not measured results, and a pilot replaces them with your own numbers.
Supermarket, $1.2B in annual sales
$4.9M to $12.4M
of gross margin a year. About $95k to $238k a week.
Estimate from planning assumptions, not a measured result. Pricing comes from positioning key value items and moving regular prices; promotions from fixing or dropping events that lose money; range and availability from delists and fewer stockouts. A pilot measures each lever separately.
Questions you should ask us.
Does it really work in my sector?
In any retail sector where prices, promotions and stock move week by week, which is most of the market. Two cases need the model adapted first, and we say so before a pilot: seasonal fashion, where the decision is when to mark down rather than what to charge this week, and very low-frequency, high-ticket categories, where each product sells too few units to estimate how shoppers react to price.
What data do I need to start?
Two years of weekly sales by product and store (or price zone), unit costs and your promotion calendar with supplier funding. Competitor prices help but are optional; the Market Watch specialty can collect public prices.
Do I have to replace my pricing or forecasting system?
No. It is designed to sit on top of what you have. Your forecasts become inputs, and the changes you approve go to the pricing and promotion systems you already run.
What stops a bad recommendation?
Your rules come first: margin floors, price ladders, legal rules and calendar locks. Every recommendation shows its confidence and its math, and a person on your team approves it before anything changes.
How much of my team’s time does a pilot take?
An executive sponsor and one category manager for each category in scope, part-time for six weeks. We bring the product lead and the data engineering, and a consulting partner adds two or three consultants part-time.
Who would I work with?
A consulting partner, who implements NorthAIsle under its own brand and runs the pilot with your team. We bring the product and the engineering behind it.
What does “trained” mean, if it never trains on my data?
The method is what is trained: how to read a price position, take a promotion apart, spot a stockout coming. That knowledge ships with the agent. Your data only feeds the calculations for your business, inside your own cloud. Estimates fitted for you stay there, and your data never trains a shared model or a language model.
Which AI models does it use?
Whichever one you have already approved. Elasticities, baselines and promotion breakdowns run as code, so the numbers do not depend on the model and can be reproduced and audited.
Which markets do you cover?
The United States and Latin America, in English and Spanish, with price logic that handles high inflation and local promotion mechanics such as bank-day discounts and interest-free installments.
Bring one category. See your first Monday.
For consulting partners: your brand, running in weeks.
The client stays yours. We bring the product, the agent and the engineering, and we stay out of the way.
Your firmMonday brief, fiscal week 44
Prepared for Fieldstone Market
Fictional retailer, synthetic data. Every decision waits for a person’s approval.
- White label
- Briefs and consoles carry your brand. The client sees your firm, not ours.
- In weeks
- Designed to deploy in the client’s Azure, Google Cloud or AWS account. A scoped pilot takes six weeks, and connectors are built with your team along the way.
- Your methods
- Your frameworks and playbooks load as the agent’s instructions, so it reasons the way your teams do.
- Margin on resale
- A subscription per banner, your margin on resale, and delivery services that stay with you.
Scope a pilot for a client.
Describe the client and the data they can share. You get a pilot brief you can paste into a proposal.