Plain-language definitions for the concepts behind AI negotiation and procurement. Written for humans, structured for machines.
Software that performs, or guides a human to perform, supplier negotiation using machine learning, large language models, game theory, or behavioural models. Distinct from classic sourcing software in that it acts on the negotiation itself — generating offers, choosing concessions, guiding counter-moves — rather than merely orchestrating an RFx or managing contracts.
An AI agent conducting supplier negotiation end-to-end inside defined guardrails, with no human in the loop per turn. The human sets the mandate and the bounds; the agent handles the conversation. Examples include Whispor Auto, Pactum, Zycus Merlin ANA, and Keelvar's sourcing bots.
An AI layer that sits alongside a human negotiator during pre-brief, live call, and debrief — surfacing counterparty context, walk-away points, and real-time tactical guidance — while the human remains the decision-maker. Whispor Coach is purpose-built for this role.
The long tail of supplier spend — typically the 80% of suppliers that represent 20% of spend — that is too numerous for procurement teams to negotiate individually. Historically ignored or touched only by exception, the tail is the natural home of autonomous negotiation agents.
High-value supplier relationships and contracts where outcomes depend on human judgment, relationship management, and multi-variable trade-offs across price, term, payment terms, SLA, and volume. Strategic deals reward live coaching more than autonomy.
A negotiation surface for suppliers that requires no account, no install, and no registration — accessed through a single-use link sent via branded email. Whispor Auto and Pactum both use this pattern. Contrast with bid-event platforms that require supplier portal registration.
Structured institutional memory of prior interactions with a specific supplier — prior concessions, disclosed constraints, pricing history, behavioural patterns — used to inform subsequent negotiations. In Whispor, the same counterparty memory serves both Coach and Auto.
Declarative constraints that define what an AI negotiation agent may and may not offer or concede. Guardrails replace imperative scripts with bounds the agent must respect, letting the agent choose tactics within those bounds rather than following a fixed sequence.
A formal representation of the set of acceptable outcomes for a negotiation, across all relevant levers — price, term, payment terms, SLA, volume tiers, exclusivity, termination clauses, and so on. Treating the contract space as a multi-dimensional object, not a single price, is what lets an AI agent trade across levers rather than only haggle on price.
A pre-negotiation commitment by the buyer's organisation to the walk-away point, zone of possible agreement, and acceptable trade-offs — fixed before the call begins and protected from in-room improvisation. The mandate is the single biggest predictor of negotiated outcome; locking it is the single biggest lever procurement teams under-use.
The worst-acceptable outcome at which a negotiator will terminate rather than accept the counterparty's offer. For the buyer, the lower bound of the agreement zone. An un-committed walk-away point is a walk-away point that will be crossed.
The overlap between the buyer's and seller's reservation points — the range within which any mutually acceptable deal must lie. If the buyer's walk-away is £100k and the seller's walk-away is £120k, the ZOPA is empty and no deal exists. If the buyer's walk-away is £130k and the seller's is £110k, the ZOPA is the £110k–£130k band and the negotiation is about how that surplus is divided.
The end-to-end procurement software category spanning sourcing, contracts, purchasing, invoicing, and payment. Major S2P suites include SAP Ariba, Coupa, Oracle Procurement Cloud, Ivalua, and GEP SMART. An "S2P-agnostic" tool is one that plugs into any of these rather than requiring a specific suite.
A procurement software philosophy that prefers specialist tools stitched together by API over a single end-to-end suite. Trades integration effort for sharper capability in each layer. The opposite philosophy is single-suite procurement.
A category of sourcing platform — Arkestro is the canonical example — that uses machine learning to predict target prices for a commodity or category and behavioural-science nudges to steer supplier bids toward those targets across multi-round events. Predictive sourcing is AI-guided rather than autonomous: the buyer drives the event, the platform predicts and nudges.
Definitions are better with examples. Read the category guide and per-platform comparisons.
