AI Agent with Business Guardrails
Sales Copilot
Ask a B2B rep with three hundred accounts who they plan to call tomorrow. The honest answer is usually: whoever called in. The portfolio is not worked by priority, it is worked by whoever makes noise, and the customer who quietly stopped buying is exactly the one nobody notices until the quarter closes.
What it is
The agent that runs on the Assisted Selling Platform as the salesperson's copilot. It ranks the portfolio by purchase propensity, flags when an account goes inactive and when a negotiation starts leaking margin, assembles quotes automatically and suggests what makes sense to offer alongside. It is not a report the rep opens at the start of the month. It is the day's queue, rebuilt whenever the data moves.
The capability no one replicates
It orders the portfolio, and the order comes with a reason
A ranking without justification is a guess with a nice interface. Every position in the queue carries the why: this account used to buy every fortnight and stopped, this one is due to restock the item it always takes, this last negotiation closed below that customer's usual margin. The rep does not have to trust the score, they read the reason and decide whether they agree. That is what makes the suggestion accepted or dismissed on the merits, rather than obeyed blindly.
The margin alert arrives during the negotiation, not at closing
Margin leakage is almost never one bad decision, it is the sum of many defensible ones. Each discount fits the rule on its own; the combination does not fit the result. The agent follows the negotiation while it is happening and warns when the composition starts drifting from that customer's or that mix's pattern. Arriving before the acceptance is the difference between fixing and explaining.
The offer arrives ready, and the rep approves it
The inversion the agent proposes is this: instead of the rep starting from scratch and building the proposal, the proposal arrives built and they decide what to do with it. The quote comes with the reason, with the variables the agent used and with the message ready to send. The rep approves, edits or discards. It is worth saying what this is not: it is not the agent selling in the rep's place. What it gives back is time, and that time returns to the part of selling where the rep is irreplaceable, which is the hard conversation, the exception and the relationship. Every assembled offer passes through the Commerce Rules Engine, so what arrives ready is already inside what the company authorized.
What goes to the rep
The discount comes suggested. Sending is the rep's call.
The Sales Copilot suggests each line's discount and flags what is already within the ceiling, with no approval needed. Sending, editing or discarding the quote is the rep's decision.
On the buyer's side: the same copilot, in chat
B2B buyers already order by chat. What changes is who answers. The Sales Copilot handles that conversation with the same technology it uses with sales reps: it understands requests in text, voice, photo or list form, asks for what is missing and hands back a cart the buyer completes in the portal. The same mechanism serves automotive parts, industrial supply, distribution, construction, agribusiness and fashion.
In a few minutes of conversation, it:
- checks that each item matches the request: the part to the vehicle, the tool to the standard and size, the style to the color and size;
- asks for what is missing (year, size, unit) with buttons, instead of guessing;
- compares suppliers or sellers by price, stock and location;
- suggests equivalent brands, only when they meet the same specification;
- builds the cart, split by supplier, and sends the link.
None of this is decided by the AI. Prices, stock, suppliers and terms come from the Commerce Rules Engine, the same rules that apply to the rep. And the rep still owns the account: the cart the agent builds is the same one they open in the portal, adjust with the buyer and negotiate within their approval limit.
In chat · Buyer
Checks that each item matches the request.
Asks for what is missing (year, size, unit) with buttons, instead of guessing.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.
In chat · Maintenance buyer
Understands requests in text, voice, photo or list form.
Recognizes each item and checks the standard, size and unit of sale.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.
In chat · Retailer
Checks each variation against the stock it ships from.
Hands back a restock cart, one variation per line, at that retailer's price.
Illustrative scene with a real CWS Platform screen in a demo version. People, stores and on-screen data are fictitious.
Frequently asked questions
How does it decide which customer the rep should call first?
By purchase propensity, computed over that portfolio's own history: reorder rhythm, items the customer always restocks, time since the last order, and deviation from their usual pattern. The queue arrives with the reason written next to each position, and the rep is free to disagree. The order is an informed suggestion, not a mandatory queue.
Does the quote it assembles already fall within what the rep can approve?
Yes. The quote is assembled by consulting the Commerce Rules Engine, the same engine that validates what the rep does by hand, so price, discount and terms already come out within that rep's limit. When the requested composition goes past that limit, the quote is not silently blocked: it escalates for approval with the request on record, exactly as it would if the rep had assembled everything manually.
Does this become one more screen for the rep to open?
The agent runs inside the Assisted Selling Platform, which is already where the rep builds quotes and negotiates. The portfolio queue, the margin alert and the quote appear in the flow where the work already happens. A productivity tool that requires opening yet another system usually costs more attention than it gives back, which is why this agent has no screen of its own.
Does the buyer talk to a bot or to the sales rep?
To both, when it makes sense. The agent handles search, compatibility checks and the cart, the repetitive part. Negotiation, exceptions and the relationship stay with the rep, who picks up the same cart in the portal.
Does this only work for automotive parts?
No. Automotive parts are the hardest case, because every part depends on the vehicle. The same mechanism checks standards and sizes in industrial supply, the reorder list in distribution, job-site materials in construction, inputs in agribusiness and size runs in fashion. The technical data changes with each catalog; the conversation, the rules and the cart stay the same.