Support Agents Are Everywhere. Selling Agents Aren't. Here's the Real Reason.

Support agents are solved. Selling agents are the frontier.

The numbers tell the story. Gartner expects AI agents to autonomously resolve 80% of common customer service issues by 2029. Salesforce expects half of all service cases handled by AI by 2027. Support is the proving ground, and AI has basically passed.

Selling agents are a different story. Only about 58% of teams even expect to have agents that support sales - and that mostly means recommending products or qualifying leads, not closing. A selling agent that actually completes a transaction is still rare.

You see the same hesitation on the buyer side. When Contentsquare asked shoppers in late 2025, only 30% said they'd let an AI agent complete a purchase for them. A Gartner survey this year found willingness to let AI make the actual purchase decision topped out around 11% for low-stakes categories. People want AI to help them shop. They're not ready to hand over the credit card.

So both sides, the company selling and the customer buying, are stuck at the same line. Help, yes. Close, not yet.

Why support agents were easy and selling agents are hard

Think about what a support agent risks getting wrong. A bad answer is annoying. You apologise, you fix it, you move on. The blast radius is small.

Now think about a selling agent. It applies a discount that wasn't approved. It quotes a price your finance team never signed off on. It closes a deal on terms you'd never offer. Suddenly the mistake isn't annoying, it's revenue, margin, and a contract you have to honour.

That's the real gap. It was never about whether the agent could hold a good conversation. It's about whether you can trust it near the part of the funnel where money changes hands.

PwC has been clear that agentic AI isn't plug-and-play, it needs real oversight and controls, especially when an action touches revenue or compliance. That's exactly the part of the funnel a selling agent lives in.

Trust isn't a feeling. It's a set of guardrails.

Here's the shift in thinking. "Do I trust this agent?" is the wrong question. The right question is: "What rules does this agent have to follow, and who enforces them?"

When you frame it that way, the problem gets a lot smaller. You don't need an agent you trust blindly. You need an agent that physically can't go outside the lines you've drawn.

A few guardrails matter most.

Discount limits it cannot cross. A selling agent should know exactly how far it can go on price, and stop there. One company we work with maps out where a voucher is allowed, where an exception needs a human, and where the answer is simply no. Their finance team's biggest fear isn't the agent talking to customers. It's the agent creating pricing exceptions someone has to clean up later. Bake the limits into the system and that fear goes away.

Pricing and packaging it can't invent. The agent should sell from the same catalogue and the same rules as every other channel. Not its own interpretation of your pricing. The same source of truth your website, your sales team, and your partners already use.

A clean handoff to a human. Trust grows fast when the agent knows what it doesn't know. Small deal, standard plan, approved discount - let it close. Enterprise sector contract, custom terms, a discount outside the matrix, pass it to a person. The agent that knows when to stop is the one you'll actually let run.

Scoped access to data. An agent helping at checkout needs to see the basket. It does not need the keys to everything. Give it exactly what the task requires, authenticated as the right user, and nothing more. When we tested early versions, handing an agent too much data actually made it worse, it got overwhelmed and lost the thread. Less, scoped tightly, beat more.

None of these are AI problems. They're commerce problems. Which is the whole point.

The guardrails have to live in one place

Here's where most companies will trip. If your discount rules live in your CRM, your prices live in your billing system, and your checkout logic lives in custom code an engineer wrote three years ago, then your "guardrails" are scattered across three systems that don't agree with each other.

A human rep can muddle through that. An agent can't. It will do exactly what each disconnected system tells it, and the gaps between them are where the expensive mistakes happen.

This is why a selling agent needs a commerce layer underneath it: one place that holds your offers, your pricing, your discount rules, and your approval logic, and applies them consistently no matter who or what is doing the selling. Think of it as the rulebook the agent reads from. Define the rules once, and every channel follows them: your website, your sales team, your partners, and now your agent.

Without that layer, every new agent is a new way for your pricing to drift out of control. With it, the agent is just another well-behaved channel selling by the same rules as everyone else.

Start where the stakes are low

You don't have to leap straight to "agent closes a £200k enterprise deal." Almost nobody should.

Start where a mistake is cheap and the path is clear. Low-value, high-volume orders. Standard plans. Self-serve customers who'd happily skip the sales call anyway. Let the selling agent close those, inside tight limits, and watch what happens.

That's how trust actually gets built. Not in a strategy deck, but in a few hundred small orders that went exactly right. Once your team sees the agent stay inside the lines, the conversation about bigger deals gets a lot easier.

The takeaway

The companies pulling ahead aren't the ones with the smartest agent. They're the ones who did the unglamorous work first: pinned down their pricing rules, set hard discount limits, defined where a human steps in, and put it all in one place an agent can read.

Trust isn't something you feel your way into. It's something you build, one guardrail at a time. Get the guardrails right and the scary question "would you let your AI close the sale?" stops being scary. It just becomes the next channel.

Thinking about letting a selling agent sell for you? Start with the guardrails. We help SaaS teams put their offers, pricing, and discount rules in one governed commerce layer, so every channel, agent included, sells by the same rules. Talk to us about agent-led selling.