A support agent should not negotiate

A support agent should not bargain with customers over prices, fees or compensation. Negotiation means deciding, in the moment, how much to concede in exchange for what, and that needs authority and judgement an agent should not hold. The agent can present fixed options, apply set rules and pass any counter-offer to a person who can decide.

What counts as negotiation?

The line is between applying a rule and making a trade. An agent that tells a customer "you qualify for a 20 credit under our outage policy" is applying a rule. An agent that responds to "20 isn't enough, I want 50" with "I could do 35" is negotiating.

Some examples on each side:

Applying set terms (acceptable within bounds) - Confirming the fee waiver policy allows one late-payment fee to be waived per year, and applying it. - Offering the two plan options the retention matrix allows for this customer. - Stating the standard compensation for a delayed delivery.

Negotiating (not for the agent) - Raising an offer because the customer rejected the first one. - Trading a concession for a commitment, such as a discount in exchange for a longer contract, outside an approved offer. - Agreeing to split the difference on a disputed charge. - Settling a complaint for an amount not set by policy.

The first list involves no new decision about value. The second list does, every time.

Why is negotiation the wrong job for an agent?

Four reasons, each enough on its own.

Concessions compound. A human negotiator who gives a little too much does so occasionally. An agent that gives a little too much does so in every conversation where the pattern recurs. Small, repeated concessions add up to significant margin, and they are easy to miss because each one looks reasonable.

Persistence wins. Agents are consistent in a way that rewards customers who push. If an agent raises its offer when a customer says no twice, customers who say no twice get better deals than those who accept the first offer. That is unfair to the second group and quickly learnt by the first.

Outcomes become inconsistent. Negotiated outcomes depend on how the conversation went. Two customers in the same situation can end up with different results because one was more forceful. With people, firms tolerate some of this. With an agent at scale, it becomes a pattern that is hard to justify.

Agreements may bind the firm. If the agent agrees a settlement in writing, the customer has a record of the firm's offer. Whether that is binding depends on the circumstances, but it is a dispute the firm did not need to have.

The same reasoning applies in other settings. Whether an AI agent can negotiate with suppliers or negotiate a settlement on a debt raises similar concerns, though the details differ. In support, the stakes are often smaller per case but the volume is far higher.

What can the agent do instead?

The agent's job is to make the person's decision easy: summarise what the customer wants, what has been offered, and any relevant history. The person then decides, with authority, and the decision is recorded.

Does it matter which channel the customer uses?

Yes, because customers switch. A customer who does not get what they want in chat may try email, then the phone, then social media. If agents in different channels apply different rules, or one channel's agent negotiates while another does not, customers will find the most generous route.

Omnichannel support is often discussed as a way to give customers a consistent experience. It is also a way to give the firm consistent outcomes. Every channel's agent should apply the same offer limits, and a counter-offer made in one channel should be visible to anyone handling the case in another.

How do you check this before launch?

Build negotiation pressure into certification. Include cases where customers reject the first offer, ask for more, propose trades, threaten to leave, claim a previous agent offered more, or simply repeat the request several times. Label the correct outcome as holding the line and offering to pass the case to a person. Score the agent on whether it stayed within limits throughout.

A support agent launch checklist should include a specific item confirming that offer limits are enforced as bounds, not just written into the brief. A limit in the brief can be talked past; a limit in the bounds cannot.

Routing helps too. When a counter-offer comes in, it should go straight to a person with authority to decide, not into a general queue. The techniques in how to triage the support inbox with an AI agent apply: tag the case as a commercial decision and send it to the right team.

Is there a cost to not negotiating?

Some customers will leave who might have stayed with a better offer. That is a real cost. But a person with authority can still make that better offer, on review, where the case justifies it. What the firm loses is only the agent's ability to improvise concessions in real time, and that is precisely the ability it should not have.

The firm also gains something: offers that are consistent, explainable and recorded. If a regulator or a customer asks why one customer was offered more than another, the answer can point to a rule or a named person's decision, not to the way a conversation happened to go.

What should you do differently?

Review your support agent's brief for any instruction that invites bargaining, such as "use your discretion" or "you may offer more to valuable customers". Replace each with a fixed rule in the bounds or a route to a person. Then test for it. Keeping commercial judgement with people is part of what makes the AI support agent that knows when to say no trustworthy.