AI Agents with Business Guardrails

For teams selling to the account base
The quote arrives ready; the seller approves, edits or discards it.
Sales Copilot shows why this customer today and which variables it used, always within the role’s discount ceiling.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.

For leaders who set how far AI goes
You decide how far AI acts on its own, action by action.
The 6 AI Agents matrix separates what an agent does on its own, what it only proposes and what it never does.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.

For IT leaders
The agent flags the rule conflict and changes none of them.
All 1,029 configurable parameters are scanned; a conflict that lives in the ERP is only reported and escalated to the team.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.

For volume buyers
The 80-line order is waiting in the cart by morning.
Procurement Agent builds the cart from an order in plain language; anything over the credit limit waits for a person to approve.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.

For network leaders
The network rebalances stock before it buys.
Rebalancing and alerting the agent does on its own, because both can be undone; a purchase suggestion goes to a person.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.

For brand owners
The campaign arrives ready and only goes live with your click.
The agent flags incomplete listings and expired campaigns on its own; it cannot touch the discount ceiling at all.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.
Platform · Governed AI
AI acting on a large operation can't hallucinate a price or break an approval limit. CWS Platform's 6 AI Agents consume the deterministic Core via MCP, under exactly the same rules as a human seller, within the Autonomy Gradient. They suggest and execute the reversible; the human decides the rest.
Inside the platform
The AI Agents at work, on screen
Three screens of the CWS Platform AI Agents: the Sales Copilot in chat with a workshop and with a building site, and the Catalog Enrichment Agent treating the product record, under the same rules as the rep.

Chat agent
The agent checks each part's fitment in the chat
The workshop asks for another brand, and the Sales Copilot answers with the equivalent part, the fitment confirmed for the car, the seller, the price and the stock, as the rep would.
Pain it removes: parts bought without knowing whether they fit the car.
CWS Platform demo screens with fictitious data.

Site material list
The site's material list becomes a cart in one chat
The engineer sends the stage's materials the way the site uses them. The agent finds each item, asks for the size when it's missing and returns the cart, with delivery under the account's rules.
Pain it removes: material lists retyped item by item by the rep.
CWS Platform demo screens with fictitious data.

Record treated
The agent treats the record on its own and stops at publication
The Catalog Enrichment Agent runs six steps on its own, from the counter photo to the spec sheet and attributes. SKU publishing sits in the yellow band of the Autonomy Gradient: the agent proposes, a person confirms.
Pain it removes: incomplete records that keep the part from being found.
CWS Platform demo screens with fictitious data.
New
AI Agents now serve buyers in Chat
The Sales Copilot, which already builds quotes for sales reps, now serves buyers directly in chat. It understands requests the way customers make them, checks that each item matches what was asked, compares suppliers, suggests equivalents and hands back a ready cart in the portal. It works for the part that has to fit the vehicle, the tool that has to meet the standard, the job-site materials list and the retailer's size-and-color restock. Same technology, same Core: the prices, stock and approval limits that apply to the rep apply to the conversation. And the Catalog Enrichment Agent now processes technical data at scale, such as parts fitment, the data that makes that conversation right.
Rules Auditor Agent
Audits 1,029 parameters in 45s, detects conflicts, prevents leakage. It suggests, the human decides.
View agent →Sales Copilot
Copilot for sales reps and buyers: accounts ranked by propensity, margin alerts, quotes built for you, and buyer service in chat that ends in a ready cart.
View agent →Procurement Agent
Semantic search, multi-seller comparison, builds the cart, reorders.
View agent →Merchandising Agent
Monitors the store, incomplete listings, suggests campaigns.
View agent →Catalog Enrichment Agent
Builds and enriches catalogs from photos and processes technical data at scale, such as parts fitment, with human validation.
View agent →Replenishment Agent
Redistributes across DCs, stockout alerts, suggests purchases.
View agent →Governed AI in 1 minute
Autonomy Gradient
AI governance by infrastructure, not by prompt.
Green
the agent acts on its own (reversible action, low risk).
3% discount, within the 5% threshold.
The order moves to checkout with no intervention.
Yellow
the agent proposes, a human confirms in one click.
8% discount, above the 5% threshold.
It waits in the manager's queue, with a mandatory reason code.
Red
the agent never acts. It informs and escalates to a human.
15% discount, above the 10% ceiling.
The interface stops it: the order is never even created.
Frequently asked questions
Can the AI Agents change a business rule?
No. The AI Agents consume the deterministic Core via MCP (Model Context Protocol), under exactly the same rules as a human seller. They interpret intent and suggest; the rule is executed by the Commerce Rules Engine, and the human decides what escalates.
What is the Autonomy Gradient?
It's AI governance by infrastructure, not by prompt. Each agent action falls into one of three zones: green (executes on its own, low risk), yellow (proposes, the human confirms) or red (never acts, escalates to the human).
If the AI goes down, does the operation stop?
No. The deterministic Core executes the rules by code, at zero inference cost. The AI Agents orchestrate that Core, but if the AI goes down, negotiation keeps running on the Core.
Can the agent serve buyers directly in chat?
Yes. The Sales Copilot serves buyers in chat with the same technology it uses with sales reps: it finds products, checks compatibility and specifications, compares suppliers and builds the cart, which the buyer completes in the portal with shipping and payment. Prices and terms come from the Commerce Rules Engine, and the rep follows the same cart in the portal, where they can adjust it together with the buyer.
Can the agent give a discount in chat?
Only what is already configured for that customer: price list, contract, campaign and coupon. The agent does not create discounts or negotiate outside the rules; negotiation stays with the rep.