AI agents in marketing: from hype to sales

For two years now, I’ve been hearing the exact same question from clients: ‘Where do I paste the website link and drop in a Google Doc so the agent starts talking to customers on Instagram and Telegram by itself?’ After industry conferences, companies often expect near-instant autonomy — hand over a few instructions, and the AI sells, advises, and closes enquiries entirely on its own.

SendPulse works with more than 200,000 customers across over 180 countries, and every day I see the gap between what gets sold as an AI agent and what actually happens inside companies. The hype is largely deserved: the technology really is powerful. But working with AI teaches you one thing quickly — without clean data and a clearly structured process, no agent performs the way the advertising promises.

The illusion of full autonomy

Today’s AI agents are good at analysing large volumes of data, producing communications at any scale and handling routine queries. By nature, though, an agent is less an autonomous operator than a capable intern: it gets through a lot of useful work, but the final call has to stay with a person.

That’s the central paradox of the ‘age of agents’. We expect the agent to work from the team apart, on its own — yet it entirely depends on that team: on the quality of the data, the instructions and the scenarios it’s given. Call it ongoing supervision. Agents need regular, sometimes daily checks to confirm they’re doing exactly what was intended, rather than improvising in situations where improvising costs the company a customer.

Real cases: from click to sale

AI delivers its most measurable results where a real gap exists between marketing spend and actual conversion to sale. This is one of the most common problems in marketing: the advertising budget is spent, and leads either never arrive at all, or they arrive and never convert. That is the stage where the money vanishes most frequently.

One area where the SendPulse team sees consistent results is opening an AI conversation on Instagram or WhatsApp the moment someone clicks an ad. It beats a standard website form: the user lands in a conversation instead of filling in fields, and the exchange feels more natural. If their interest holds up, sales gets far more context about the customer before the first direct contact. For businesses with simple products, an AI conversation in a messenger sometimes takes the user all the way to purchase without a manager involved — our clients’ cases certainly bear that out.

One condition is crucial, however. Don’t hand AI tasks that are too broad or too vague; accuracy will drop sharply. With a narrow script, where you know exactly which advert the user responded to and what they are asking about, you can set a firm rule: stick to the script. 

The moment a question falls outside it — or anything ambiguous comes up — pass the conversation straight to a human. That boundary is where companies most often lose money, and it is also the easiest place to build in the control that rules out unpredictable behaviour.

Why smaller businesses find AI easier to adopt

The question I am asked most often is where to start when the budget is tight. A vast proportion of SendPulse customers sit in exactly that segment — in Ukraine, Brazil, and other Latin American markets. Here, small and medium-sized enterprises (SMEs) have a distinctly non-obvious advantage.

The big players usually take years to roll out new technology: complex architecture, endless sign-offs, multi-year implementation plans. Small businesses have nothing to do with any of this. They can pick up solutions available today and launch them within the month, with no long-term strategy attached. 

This also means the small business owner needn’t become a prompt engineer overnight: they can carry on working with familiar tools and slot controlled AI precisely where it solves a specific problem — handling standard queries, say, or first-pass lead qualification.

Five rules for launching AI agents safely and effectively

• Don’t hand every process to AI.

Not every process needs its own agent. Plenty of routine jobs — auto-replies, reminders — are already handled well by familiar tools such as chatbots and automated mailings.

• Start where the gap between spend and conversion is widest.

Usually that’s the point after the ad click and before first contact with sales.

• Write narrow scripts, not broad ones. AI performs far better with a tightly defined task than with free rein.

• Build in a handover to a human. This isn’t extra insurance; it’s a required part of the architecture, and protects the company from reputational damage.

• Don’t wait for the perfect moment. Small businesses don’t need years of preparation — you can start with one narrow use case this month.

The hype around AI agents isn’t going anywhere, and much of it is well-deserved: the technology performs exceptionally well on clearly defined tasks. However, that is no reason to expect a flawless turnkey solution, nor to roll AI out everywhere at once.

Companies of any size should be building practical experience with AI now, even in narrow, limited scenarios. That’s the preparation that keeps you from falling behind a year or two from now.