A quotation goes out faster, the customer accepts the proposed discount and the order is won. That looks good for sales. Whether the company earns more from it is another question. If an AI assistant prepares quotations, the limits of the commercial terms it may propose must therefore be agreed before it is put to use.
I would not start by asking how much discount the system can calculate. First, establish the threshold below which an order needs a different level of approval. A higher win rate is of little help if each additional order contributes too little towards covering ongoing costs.
A five per cent discount can cost a quarter of the contribution
Consider an explicitly illustrative calculation: the net price to the customer for a standard order is CHF 10,000. Its variable costs are CHF 8,000. These are the costs incurred for this order that would be incurred again for an additional, comparable order. The contribution margin is CHF 2,000. This amount is available to cover fixed costs and, once those are covered, generate profit; it is not the company's profit. The 2019 OpenStax management accounting textbook also explains this calculation. 1
A 5% discount reduces the price to CHF 9,500. If variable costs remain unchanged, the contribution margin falls to CHF 1,500. In this model, a 5% discount reduces the contribution per order by 25%. Achieving the same total contribution would require approximately 33.3% more identical orders: CHF 2,000 divided by CHF 1,500, minus one.
This is not a measured industry effect, a Fassbender client result or a forecast. The calculation assumes unchanged variable costs per order, sufficient capacity and no additional fixed costs. If the extra volume requires more staff, equipment or external services, the calculation must be revisited. It does, however, show why a small discount should not simply be treated as a minor commercial decision.
The discount threshold must reflect the order, not just the list price
A blanket rule such as “sales may approve discounts of up to five per cent” can be useful. But it is not enough on its own. Two orders with the same price can require different levels of work. Special transport, additional support or a customer-specific modification will change the calculation.
Before automating discount approval, I would therefore start with one recurring type of quotation. Finance and the person responsible for operations jointly define which costs enter the calculation and where the current figures come from. Management then sets the permitted limits. The model must not infer the minimum contribution from past quotations. Historical exceptions do not confer current pricing authority.
AI can explain a proposal, bring together approved terms and flag missing information. The definitive calculation and the threshold check must use transparent calculation rules. A persuasively worded proposal is no substitute for this check. In its voluntary 2023 AI Risk Management Framework, NIST recommends clearly defined responsibilities for managing risk. The framework provides guidance, not certification or a legal guarantee. 2
An approval rule that works in day-to-day use
For the selected type of quotation, I would set out the following on one page:
- ◆Calculation basis: The net price after all discounts and the confirmed variable costs, including relevant additional services.
- ◆Permitted limits: The minimum contribution the order must achieve and who may approve a discount within that limit.
- ◆Exceptions: The named role authorised to approve an order below the threshold and the justification that person must document.
- ◆Insufficient information: No automatic approval if reliable costs are unavailable or an additional service has not been clarified.
That page must actually be used in the quotation process. A rule left in a folder does not protect margin. The proposed discount, the cost basis used and the decision should remain traceable in the quotation record. If purchase prices or the scope of work change, an old calculation cannot simply continue to apply unchanged.
There can be good reasons to accept an order below the usual threshold. Entering a market or maintaining an existing customer relationship may justify a deliberate exception. This requires an accountable decision with a defined scope. Expected follow-on orders remain an assumption unless they are contractually secured. The assistant must not turn that hope into an apparently substantiated financial return.
At the next sales review, I would go through a small selection of actual discounted quotations with Finance. Do the costs used in the calculation still hold up after the work has been delivered? How often was an exception approval needed? Did the order won achieve its expected contribution? This shows whether the threshold works in practice or merely looks sensible on paper.
If an agreed transformation plan already includes these quotation processes, the AI Transformation Execution Partnership can help embed calculation rules, approvals and the internal handover in day-to-day operations. The value must be assessed against orders that have actually been fulfilled. Not against the number of quotations sent.