Does it require a language model?

No. Many negotiation agents use structured offers and mathematical strategies.

What is a reservation value?

It represents the value assigned to not reaching the proposed agreement, such as using an alternative.

Can both parties benefit?

They may find a package both prefer to their alternatives, but that depends on the available options and preferences.

AI Glossary / AI and machine learning / Intermediate

Automated negotiation

Definition

Automated negotiation is a process in which software agents exchange, assess or revise proposals to seek an agreement under defined rules. Agents act according to represented preferences, constraints and authority, which may be supplied by people or organizations.

What it means

Negotiation involves more than selecting the lowest price. Delivery time, quantity, service conditions and other issues can matter together. Two parties may value those issues differently, creating opportunities for a mutually acceptable package.

An automated negotiator needs a way to represent what its principal values and what it must never accept. A utility function can score acceptable outcomes, while hard constraints rule out prohibited ones. A reservation value represents the value of the alternative to reaching an agreement.

For business readers, the key distinction is between recommending an offer and committing to a deal. The system's authority must be explicit. A technically attractive agreement can still be unsuitable if the preference model omits an important obligation or relationship.

How it works

A protocol defines who can make offers, how acceptance works and when negotiation ends. An agent's strategy determines which proposals it makes and which it accepts within that protocol.

The agent may know little about the other party's preferences. It can infer possibilities from responses, but those inferences remain uncertain. Strategies can balance pursuing a better outcome against the risk of reaching a deadline without agreement.

After an offer is accepted, the surrounding process must confirm the exact terms and the agent's permission to commit. Negotiation software should also handle withdrawals, ambiguous responses and failed communication. Evaluation considers agreement quality, failed negotiations and adherence to constraints, not just whether the agent obtained a lower headline price.

Three practical examples

A scheduling agent exchanges acceptable time slots with other agents while respecting each participant's availability.
A procurement experiment evaluates price-and-delivery packages within limits set by a buyer.
A resource-allocation system negotiates capacity among services that have different priorities and deadlines.

Think of it this way

Think of giving a colleague a clear mandate before they discuss a purchase: acceptable options, priorities, a walk-away point and whether they can sign. Software needs an even more explicit version of that mandate.

The analogy has limits because a person may notice an unspoken concern or seek clarification. An automated agent can optimize only within the information and procedures it has.

Worked example

Comparing packages with a simple utility score

Suppose a fictional buyer weights price suitability at 60% and delivery suitability at 40%. Offer A scores 0.9 on price and 0.4 on delivery, giving 0.6 × 0.9 + 0.4 × 0.4 = 0.70. Offer B scores 0.7 on price and 0.9 on delivery, giving 0.78.

Under this simplified additive model, B is preferred despite its lower price score. If delivery after Friday is prohibited, any such offer must first be rejected regardless of its weighted score.

The example assumes scores are meaningful and comparable. It does not prove the weights reflect the buyer's real preferences or that issues are independent. A review should examine those assumptions before allowing an agent to negotiate binding terms.

Illustrative case study

A fictional events company tests an agent for arranging equipment hire. The initial preference model strongly rewards low cost, and the agent proposes a cheaper collection location that would create a difficult transport journey.

The coordinator adds transport constraints and clarifies which trade-offs may be made without review. The agent then produces candidate packages for staff acceptance, and the team records why each offer was preferred.

The case illustrates preference elicitation as part of automation. A sophisticated bargaining strategy cannot compensate for a missing business requirement.

Common misconceptions and mistakes

Automated negotiation always means autonomous contracting. Some systems only recommend proposals.
Agreement is always better than no agreement. A valid alternative can be preferable to an unacceptable deal.
The opponent model reveals the other party's true preferences. It is an estimate based on limited evidence.

Business use

Automated negotiation can help when many similar exchanges must be coordinated and acceptable trade-offs can be specified. Start with a bounded domain and explicit approval limits.

Assess outcomes against the value of the alternative, total operating cost and important non-price conditions. Keep a record of offers and accepted terms. For complex relationships or ambiguous commitments, a system that prepares options for a person may provide more value than one given broad authority too early.

Check your understanding

An offer has the highest weighted score but violates a non-negotiable delivery deadline. Should the agent accept it?

Answer: No. Hard constraints should be enforced before comparing acceptable outcomes by utility.

Related terms

Utility function
Preference elicitation
negotiation agent
reservation value
multi-agent system
protocol
constraint

Learn next

Read utility function and constraint. Describe a small negotiation using permitted issues, unacceptable outcomes, a fallback option and the authority to accept an agreement.

Utility function

Sources

Automated Negotiation with Gaussian Process-based Utility Models: https://www.ijcai.org/proceedings/2019/60. A negotiator's backup plan: Optimal concessions with a reservation value: https://ir.cwi.nl/pub/34279. Primary research examines utility uncertainty and fallback values. The package scores and events-company case are original teaching scenarios.

Previous: Automated essay scoring Next: Automated reasoning