AI agents in marketing and sales: what they can and can't do

AI agents can already reply to enquiries, qualify leads, book calls and draft follow ups for a person to approve. They still slip on longer tasks, and customers want a way out. According to Gartner, August 2026, 87% of customers say it is essential that companies using GenAI offer a way to reach a human agent.

How many companies are using AI agents?

Use of AI is common. Running agents at scale is not.

In McKinsey's State of AI survey, August 2026, nearly 9 in 10 respondents report regular use of AI in at least 1 business function. Agents are further behind.

According to McKinsey, August 2026, 40% of respondents from organisations with more than $1 billion in annual revenue are scaling AI agents. That is up from 27% a year earlier, while smaller organisations stayed flat at 22%, McKinsey found.

Sales teams are trying them. According to Salesforce, February 2026, 54% of sellers say they have used agents, and nearly 9 in 10 plan to by 2027.

Marketing teams are earlier. Salesforce reported in May 2026 that 75% of marketing organisations use AI in some form, but only 13% use self directed agents.

The key figures:

• According to McKinsey, August 2026, nearly 9 in 10 respondents report regular use of AI in at least 1 business function.
• McKinsey found in August 2026 that 40% of respondents from organisations with more than $1 billion in annual revenue are scaling AI agents.
• According to Salesforce, February 2026, 54% of sellers say they have used AI agents.
• Salesforce reported in May 2026 that only 13% of marketing organisations use self directed agents.
• In a May 2025 benchmark by Salesforce AI Research, leading AI agents succeeded on about 58% of business tasks set in a single turn.
• The same May 2025 benchmark found success fell to 35% when tasks ran over several turns of conversation.
• According to Gartner, August 2026, 87% of customers say an option to reach a human agent is essential when companies use GenAI.
• Gartner predicted in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027.

What can AI agents do for marketing and sales today?

The best jobs for an agent are frequent, well defined and easy to check. They tend to be conversations that follow a pattern.

• Reply to new enquiries by email, text or chat within seconds, in the brand's tone.
• Ask qualifying questions and record the answers against the contact.
• Offer real availability and book calls into the right calendar.
• Follow up with people who went quiet, and stop when they reply.
• Answer common questions from approved information about services, process and policies.
• Draft emails, social posts and review replies for a person to approve.

Each of these jobs has a clear end point. The agent either booked the call or it did not, and a person can check the record in seconds.

Sellers expect the savings to land in specific tasks. According to Salesforce, February 2026, they expect agents to cut prospect research time by 34% and email drafting by 36%.

The commercial upside sits in the same place. McKinsey's August 2026 survey found revenue gains are most often attributed to AI used in marketing and sales.

Where do AI agents still fall short?

Accuracy drops as tasks get longer. In a May 2025 benchmark by Salesforce AI Research, leading AI agents succeeded on about 58% of business tasks set in a single turn.

When tasks ran over several turns of conversation, success fell to 35%, the same May 2025 benchmark found. The agents also showed almost no built in awareness of confidential information.

Projects stall too. Gartner predicted in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027. It pointed to escalating costs, unclear business value or inadequate risk controls.

Gartner's June 2025 release also estimated that only about 130 of the thousands of agentic AI vendors are real. Judge an agent by what it does inside your system, not by its label.

Then there is judgement. Pricing, complaints, anything emotional and strategy still need a person. So does checking that what the agent says is still true when the business changes.

Do customers want to deal with AI agents?

Many do, provided there is a way out. According to Gartner, August 2026, 50% of customers say their interactions are easier when companies use GenAI.

The same August 2026 Gartner survey of 3,566 B2B and B2C customers found 87% say an option to reach a human agent is essential. Customers push back when AI acts as a barrier to reaching a human, Gartner said.

Gartner's advice is not to make GenAI a mandatory first step for every issue. An agent should open the door to help, not guard it.

So tell people they are talking to an AI agent from the first message. Offer the human option in plain words, not buried in a menu.

When should an AI agent hand over to a person?

Write the handover rules before the agent goes live. These are the triggers worth building in:

• The person asks for a human, in any words.
• The conversation turns to price, discounts, contracts or refunds.
• The person is upset, complains or raises a legal or safety concern.
• The agent cannot find the answer in approved information.
• The person repeats a question the agent has already failed to answer.
• A qualified lead is ready to talk to someone who can make decisions.

At handover, the agent passes the full conversation to a named person and tells the customer what happens next. Nobody should have to repeat themselves.

Out of hours, the agent should stay honest about what it is. It can take details, book a callback and say when a person will reply.

Test the rules before launch. Run 20 real enquiries from past months through the agent and check every handover it should have made.

What does this mean for your marketing system?

An agent is only as good as the system it works in. According to Salesforce, February 2026, 51% of sales leaders with AI say disconnected systems are slowing down their AI initiatives.

So build the foundations first. That means a clean pipeline with stages that match reality, services and policies written down, and follow up that works without AI.

Then add agents with guardrails. They never offer discounts, they hand complaints to a person and nothing goes public without approval.

Start with a single job, such as replying to new enquiries out of hours, and add more once the numbers hold.

Measure an agent the way you would measure a new hire. Track time to first reply, conversations resolved without help, handovers and their reasons, bookings created and every answer a person had to correct.

Agents can also work from outside the CRM. Through MCP, assistants such as Claude can read a pipeline, update contacts and draft follow ups from a chat.

Afoofa builds AI agents into each bespoke system once the pipeline, data and follow up underneath are in place. Every agent works from approved information and hands over to a person by design.