What we do not let AI do

Most teams do not worry about AI when it is drafting a polite reply. They worry when it starts sounding certain about things it should not decide. In a dealership, that line matters.

A customer asks for a part exchange figure, wants to know if a car is definitely available, or expects a promise on finance, delivery or a complaint. One wrong message can create confusion, disappointment or a compliance problem that takes much longer to put right than it did to send.

That is why the most useful question is not what AI can do. It is what it should never do without a person involved. The answer says more about trust than any demo ever could.

Five tips for setting up your own AI agent

Before the guardrails, the practical part. If you are about to put an AI agent in front of customers, these five habits matter more than any feature list.

1. Ownership. Have someone in the dealership who is in control of all AI edits and tweaks. Feedback flows through them, and they make sure every AI message sounds like your brand.
2. Treat it like a new employee. Put it through the same onboarding you would give anyone joining the team.
3. Workflows. Map out the most time-consuming workflows you would like to hand over. Explain the objective, the usual messaging flow, and when to flag a conversation to a human.
4. Start small and grow. Limit the AI at the start. As you, your team and crucially your customers get more accustomed to it, let it handle more.
5. Continual editing. Once it is set up, do not forget it. Plan regular checks to update your instructions so the agent keeps up with what is new for you.

AI should help with speed, not pretend to have authority

There are plenty of jobs AI can do well. It can answer common questions, collect details, suggest appointment times, route the conversation, and keep replies moving when the team is busy or out of hours.

Those tasks are helpful because they are repeatable and low risk when they are set up properly. They save time without asking the system to make judgement calls that belong to a trained member of staff.

The trouble starts when AI is treated like a decision-maker. Customers do not always hear a difference between a suggestion and a commitment, especially on WhatsApp where messages feel quick, personal and final.

If an automated reply sounds definite, the customer may reasonably assume the dealership stands behind it. That is exactly why some boundaries need to be strict, not vague.

We do not let AI quote prices or make promises

There are some messages we do not want AI sending on a dealership's behalf. Pricing is one of the clearest examples.

A vehicle price can depend on stock status, trim, prep, location, optional extras, campaign rules, finance terms, part exchange, and what has changed since the data was last updated. Even when a system has access to useful information, it may not have the full context needed to phrase that answer safely.

So we do not think AI should quote prices as if it were a salesperson with authority to commit the business. The same applies to valuation figures, finance offers, discount promises, deposit terms, delivery dates, complaint outcomes and anything else that could be read as a firm undertaking.

It can explain the next step. It can gather the details a human needs. It can say that a member of the team will confirm the right answer. What it should not do is sound confident where confidence would be misleading.

Some conversations should always reach a human

Not every message is suitable for automation, even if the wording looks simple at first glance. The real meaning often sits behind the words.

A complaint is not just a request for information. A customer saying they feel unsafe in a vehicle is not a routine booking query. A message about finance can carry regulatory sensitivity even when it sounds casual.

That is why there should be clear categories that always go to a person. Complaints, vulnerable customer situations, legal threats, finance-specific questions, requests to vary agreed terms, safeguarding concerns, and anything that could put the dealership in a difficult position should be handed over quickly.

The same goes for emotionally charged conversations. If a customer is angry, distressed or confused, the right response is usually empathy and judgement, not pattern matching.

Good guardrails are not there because AI is bad. They are there because customer communication involves trust, context and accountability, and those still sit best with people.

Confidence matters, but uncertainty matters more

A useful AI agent is not the one that answers everything. It is the one that knows when not to answer.

In practice, that means looking at confidence before a reply is sent. If the system is not sufficiently sure what the customer means, what the policy allows, or whether the available information is current enough, it should stop and pass the conversation on.

That handover matters just as much as the reply itself. The customer should not be left wondering why the tone changed, whether they have been ignored, or if they need to repeat themselves from scratch.

The better approach is simple. Acknowledge the question, explain that a team member will confirm the details, and carry over the conversation history so the human can pick up without friction.

This is where restraint beats ambition. A lower-risk answer with a clean handoff is often the better customer experience than an instant reply that creates doubt.

Guardrails need rules, records and review

Saying AI has guardrails is easy. The hard part is being specific about what those guardrails actually are.

At a practical level, that starts with policies on what the agent can answer, what language it can use, what sources it is allowed to rely on, and which topics trigger handover. If those rules only exist as a general intention, they will not hold up when the edge cases arrive.

It also means keeping a record of what was said and why. When a conversation is escalated, the team should be able to see the original customer message, the AI response if there was one, and the reason for the handoff.

That audit trail matters for internal learning and for accountability. If a process needs tightening, you can only improve what you can inspect.

Review matters too. Stock feeds change, dealership processes change, campaign wording changes, and compliance expectations change. A safe setup is not something you configure once and forget.

Compliance is not a feature, it is a habit

It is tempting to think of compliance as a box to tick before launch. In reality, it shows up in small decisions every day, especially in messaging channels where customers expect quick answers.

A reply can be warm, short and helpful without drifting into advice, promises or statements that need formal confirmation. That takes careful design, approved wording, clear escalation paths and regular checks on live conversations.

Teams also need clarity on responsibility. If AI is helping with first responses, who owns the templates, who reviews exceptions, who updates the rules, and who signs off sensitive changes? If nobody owns those questions, risk slips through the gaps.

This is one reason shared visibility matters. When conversations sit in one place instead of being scattered across personal phones, it is far easier to supervise what is being said and step in when needed.

Used properly, AI can reduce risk by making approved responses easier to use and by spotting when a case needs attention. Used carelessly, it can scale inconsistent messaging very quickly. The difference is rarely the model alone. It is the discipline around it.

You are in control

Fuzey has its own guardrails that protect every customer on the platform, and those are not optional. On top of them, you set your own critical constraints for your dealership.

That means you decide what the AI can and cannot say. Which topics it answers, which wording it uses, which questions always go to a person, and which parts of the process it is allowed to move forward on its own.

At Fuzey, AI is there to handle the repetitive work, keep conversations moving and bring the right cases to the right people. The point is not to replace responsibility. It is to protect it.

In a dealership, trust is built when customers get the right answer from the right person at the right moment. Keeping you in control of the limits is how that stays true.