What Winston Knows
Any AI can reach your data.
Winston knows what it means.
Connect an AI to your systems and it will spot a discrepancy between them in seconds. It still won't know what caused it — because that answer isn't in any one system.
A connector is a checkbox, not a moat.
Connecting AI to business software is becoming a standard. That's real progress, and it will soon be table stakes — in a year or two, an AI connector will be a box to check on every platform's feature list, the way an API is today.
Which is exactly why it isn't a differentiator. When every system has one, the question stops being whether AI can reach your data. It becomes what the AI knows once it gets there.
The obvious rebuttal is to just feed it your context: upload the product catalog, hand it a spreadsheet of acronyms, let a frontier model figure out the rest. That works for the first layer. It falls apart on the ones that run the business.
Access is not understanding.
Post-trained on how regulated retail actually runs.
Winston runs on frontier models and adds models of its own, trained on real regulated-retail operations: how invoices and manifests are structured, how vendors abbreviate and how they bill, how categories and formats are organized, how inventory and compliance records are supposed to reconcile.
That shows up in different ways depending on the job. Here's the one that fits in a diagram.
One product. Five names. No report until someone fixes that.
01 — What comes in
WILDWOOD - Blue Dream .5g Liq Diamond PR 2pkPre-RollsInfusedWildwood Farms Blue Drm LD Preroll 2/.5 (cs 12)PREROLL—Blue Dream Diamond Infused Pre-Roll 2 x 0.5g — WildwoodFlowerPre-Roll InfusedSame product. Three systems. Three spellings, three abbreviations, three different ideas of what category it's in — and one of them bills in cases.
02 — The collapse
03 — One record
- Brand
- Wildwood Farms
- Product
- Blue Dream Liquid Diamond Infused Pre-Roll
- Format
- 2-pack · 0.5 g each
- Total weight
- 1.0 g
- Category
- Pre-Rolls
- Subcategory
- Infused Pre-Rolls
- Unit basis
- 1 case = 12 units
04 — What that unlocks
Only now is a price comparison possible. Before the collapse, these were four unrelated products.
Your cost is $14.40 a unit. At the median you still hold 48% margin.
Normalizing names is the boring part. It's also the part a connector can't do for you.
AI with access
“I found 4 products matching ‘Blue Dream’ across your stores. Prices range from $26.00 to $32.00.”
Winston
“That's one product, not four. You're 16% above the median on it, and it's your #3 pre-roll by units. Want me to draft the price change for your approval?”
Reconciling names is one example, not the whole of it. The same training is why Winston can read a vendor invoice in a format it hasn't seen before, tell a receiving miscount from a genuine compliance variance, and know that a case of twelve isn't twelve sales.
None of that comes from having access to your systems. It comes from knowing what the numbers in them mean.
Where that training data comes from
Winston's models learn from de-identified, aggregated operational patterns — how vendors abbreviate, how categories drift, how invoices are structured. Not customer or patient identifying information, and nothing that surfaces your business's numbers as your business's numbers. How we handle your data
The distinctions nobody wrote down.
Knowing the industry gets Winston most of the way. The rest is specific to your business — and none of it fits in a spreadsheet, because half of it was never written down.
A SKU is never just a SKU
It belongs to a product line, sitting inside a promotional scheme that treats your in-house brands differently from everything else on the shelf.
The right substitute and the wrong one
When something sells out, there's a right swap and a wrong one. Which is which depends on what your customers actually reach for, not what the category label says.
Two products, identical on paper
Same format, same price, same category. One carries margin. The other carries a vendor obligation.
Half of it lives in one person's head
It was never written down anywhere, because it's not a document. It's how your buyer thinks.
An AI with full access and no context still answers. It answers cleanly, confidently, and on distinctions it doesn't know exist.
Your data stays yours.
Training Winston to understand retail does not mean handing over your business.
Patterns, not records
Winston's models learn from de-identified, aggregated operational patterns — not in any way that could reconstruct your data.
No customer or patient identifiers
Identifying information about your customers and patients is not used to train models. You can also tell us to exclude any data category entirely.
Nothing crosses operators
Winston runs in a contained instance only your team works inside. There's no shared Winston, and your records don't show up in anyone else's.
Human in the loop
Money movement, POS writes, and customer communications always require your approval.
Full audit trail
Every action logged, reversible, and explainable. Tap any action to see why.
Encrypted and isolated
Your data is encrypted in transit and at rest, and isolated to your instance.
On the frontier models underneath. Winston runs on frontier models from Anthropic, OpenAI, and Google. They're used to run Winston, not to learn from your business — we don't permit them to train their foundational models on your data. They may hold input and output for up to 30 days for safety review, after which it's deleted.
On checks before anything posts. Winston validates its work before it writes to any system, and a person on your team approves it.
On our program. We maintain a SOC 2 Type II security program with ongoing security monitoring, and encrypt data in transit and at rest.
The full detail is in our Privacy Policy and Terms of Service.
Getting started is not a project. The project already happened.
Whoever owns the teaching owns it forever. Catalogs change, vendors change, and your state rewrites its rules. Winston exists so that job is mostly done when you arrive.
It arrives knowing the industry, then learns your specifics on the job. Tell it once that a vendor bills in cases and you stock in units, or that a field should stop auto-filling, and it remembers. Every correction from your team makes it better — and that knowledge stays with the business, not in one employee's head.
One multi-store operator spent the better part of a year getting frontier models to run their purchase-order work. They got there. But it took longer than planned, it ran through one person who became the bottleneck for the whole system, and keeping it working turned into a job nobody had budgeted. The AI was capable on day one. Making it reliable was the project.
Context compounds over time.
You ask for competitor pricing
Winston learns what matters to you
You mention margins matter more
Winston picks up on your priorities
Reports now lead with margin analysis
Winston adapts its output to your needs
Winston flags margin issues proactively
From reactive to proactive intelligence
Entire team benefits from what you taught Winston
Your insights scale across the organization
This isn't Winston versus connectors.
Your business is bigger than any one platform. The invoice arrives in email. The inventory balance sits in accounting. The delivery discrepancy traces back to a receiving shift in your workforce system. An AI attached to one platform reasons inside one platform, and confidently answers questions it cannot actually see.
Winston works across the stack, which is what lets it answer the why instead of flagging the what. It will use new connectors as they arrive — that's plumbing worth having.
The connection is the plumbing. The teammate is the point.
We're onboarding new stores now. Join the waitlist and we'll be in touch.
Questions? Talk to our team