Marketing in the AI agentic enterprise: Lessons from Salesforce

When Salesforce piloted an AI agent to engage with prospective customers, it worked around the clock, qualifying leads and delegating tasks to human sales representatives. There was one problem: it did not stop, with some regular and at times demanding requests for the human team. The company had to tell what Leandro Perez, Senior Vice President and Chief Marketing Officer of Salesforce Australia and New Zealand, described as an "overactive intern" to slow down.

The anecdote, shared at the UNSW Marketing Analytics Symposium Sydney (MASS) 2026, captured a tension that every organisation adopting AI agents will need to resolve: the technology needs the right guardrails and governance to move from pilot to production and use with customers. For Mr Perez, the story illustrated a broader point about why marketers – rather than IT teams alone – should take ownership of the AI agents that now represent brands to customers.

In a fireside chat with Nicolas Chu, Professor of Practice in the School of Marketing at UNSW Business School, Mr Perez outlined how AI agents are changing business models, marketing metrics and the skills that professionals need to remain relevant. He drew on his own path from a UNSW computer science degree to leading marketing at one of the world's largest technology companies, and on Salesforce's experience as both a vendor and user of its own technology.

From software engineer to CMO: how a technical foundation shapes marketing strategy

Mr Perez did not follow a conventional route into marketing. He studied computer science at UNSW, worked as a software developer and later built product marketing and corporate messaging teams at Salesforce's headquarters in San Francisco. In that role, he worked with Salesforce co-founder and CEO Marc Benioff, whom he described as the company's de facto chief marketing officer.

"I don't approach most problems the same way as traditional folks in my team," Mr Perez told the MASS 2026 audience. "Obviously, at UNSW, when you become a software developer, you think of everything as a system. And in fact, when you get given a problem, it's an interesting challenge to solve. Whereas most people, if they have a problem and have to fix something, they don't want to do it. Actually, I really run towards those."

Subscribe to BusinessThink for the latest research, analysis and insights from UNSW Business School

That systems-oriented thinking, he said, carried through into everything he did in marketing. Every campaign, every process and every customer interaction was something that could be broken down into components, measured and improved – the same way a software engineer would refactor a piece of code. He combined that technical discipline with what he had learned from Benioff about narrative and messaging, and it was that combination, he argued, that defined what modern marketing required.

What the agentic enterprise means for marketing and customer engagement

Much of the conversation centred on what Salesforce calls the "agentic enterprise" – a model in which AI agents collaborate with each other and with humans to deliver business outcomes. Mr Perez was careful to distinguish this from earlier applications of AI at the company. Salesforce had been using AI for over a decade, he noted, from next-best-action recommendations for sales representatives to send-time optimisation for email, processing one trillion AI predictions every week. The shift to agentic AI was something different in both scale and kind.

"It's not what you would expect when you interact with an LLM yourself. You're getting a response after you ask. It's having agents across departments working together to deliver an outcome," Mr Perez said.

Salesforce tested the approach on itself first – a practice the company referred to as being "customer zero." It launched a public-facing agent on help.salesforce.com that could handle customer support queries in conversational language. That agent had held 2.5 million conversations with an 87% deflection rate, freeing support engineers to focus on more complex work.

Photo gallery: UNSW Marketing Analytics Symposium Sydney (MASS) 2026

Mr Perez pointed to external examples as well: PepsiCo had deployed agents that connected field technicians restocking fridges with sales teams engaging customers and back-office operations, all working from a shared knowledge base to remove friction across the customer journey.

For CMOs, Mr Perez said, the rise of AI agents represented an opportunity to expand their influence. He argued that because agents were the front door of a brand – requiring the right tone, the right messaging and the right customer experience – marketing was as well positioned as any department to lead the orchestration. "I think marketing should have a seat at the table, and I think we're better positioned because we're natural storytellers. We care more deeply about the brand and that experience you have with the customer," he said.

How AI pricing models are reshaping the way technology companies sell

Prof. Chu pressed Mr Perez on a structural change underway across the software industry: the shift from subscription-based SaaS pricing to consumption-based and outcome-driven models. Salesforce had built its business on the principle that customer success equalled company success – if a customer was not getting value, they would leave the platform. But the introduction of AI agents had complicated the pricing equation.

Learn more: How businesses can leverage AI agents (without crossing the line)

Mr Perez explained that Salesforce now offered three pricing approaches: traditional per-seat licensing, consumption-based pricing tied to agent actions, and open-ended trial arrangements designed to encourage adoption. The challenge was that the value of an agent action varied widely across use cases. He cited one case where a supermarket wanted to use an agent for price checking – an action worth less than a cent – while for a bank, a single mortgage origination process might be more expensive.

"I'd say we're in an interesting time where people want to explore new use cases. They want flexibility, but they also have a CFO that they need to report to, who is expecting outcomes that are defined," Mr Perez said.

Beyond the marketing qualified lead: how analytics must evolve for the buying group era

The conversation turned to marketing measurement, an area where Mr Perez said Salesforce had undergone a significant internal transformation. For years, the company's marketing analytics had centred on traditional signals: event badge scans, webinar registrations and PDF downloads. But as the company launched new digital engagement platforms – including an online streaming service called Salesforce Plus and a learning platform called Trailhead – the data inputs multiplied.

Salesforce unified all of those data streams – marketing, learning, paid and organic advertising, service adoption, customer success and sales pipeline – into a single data layer. The result was a move away from the marketing qualified lead as a primary metric. "We moved from a signal like a marketing qualified lead, which is really just one touch point, to a much more robust approach. And this is where AI comes in – because it's very hard for any one individual to know what's going on with all of those signals," Mr Perez said.

He also highlighted the rise of the buying group as a force that demanded change. Enterprise sales no longer involve one decision-maker. A single opportunity might include a buyer, their manager, the CIO, a chief data officer, a chief AI officer and even the CEO or the board. Tracking engagement across all of those contacts as a collective opportunity, rather than as isolated individual leads, required a fundamentally different approach to marketing analytics.

Mr Perez said his team had shifted focus from pre-acquisition metrics to a model that tracked the full customer lifecycle: acquisition, adoption, expansion and renewal. He framed the change as essential for any CMO who wanted to maintain influence.

"If you want to have a seat at the table, I talk to our local GM or our global CEO, for example. You need to be tied to those business metrics. Most CMOs I know who have been in their seat for a long time are the ones who align that way," he said.

The skills marketers need in an AI-driven world: curiosity, resilience and domain expertise

Asked what skills and mindsets marketers needed to remain relevant over the next three to five years, Mr Perez said there was no single answer, but resilience and adaptability mattered most. He described taking his team on an annual offsite after Salesforce's major conference, where, for the past four years, he had introduced progressively more demanding AI challenges. What started as playful experiments with generative AI video tools had evolved into structured problem-solving sprints where cross-functional teams had 60 minutes to solve real business challenges using three or four AI tools.

"I think for me, it's about instilling this idea of being comfortable with change. You need to lean into it," Mr Perez said. 

Learn more: Understanding customer decision-making: The Amazon challenge

He was most emphatic about domain expertise. With natural language now the interface for AI tools, Mr Perez argued that the ability to code in a specific programming language mattered less than understanding the business problem. If a marketer were the subject-matter expert who could precisely describe a problem and evaluate whether the AI's solution was right or wrong, that skill would be more valuable than technical execution ability.

"I think a lot of marketers forgot that that was the key skill: understanding the product, understanding the solution, and what the customer needs, as opposed to becoming a specialist in a specific channel or being able to execute a certain way," he said.

On hiring, Mr Perez noted that Salesforce had removed university degrees as a prerequisite for roles, shifting from role-specific to behavioural interview questions while placing greater emphasis on collaborative skills. In a company where marketing was, as he put it, a 'team sport', he said the ability to navigate a matrix organisation, resolve conflict and prioritise competing demands from multiple stakeholders counted for as much as technical proficiency.

Key takeaways for marketing and business leaders

The fireside chat outlined practical implications for professionals navigating the transition to AI-augmented business operations. First, marketers should see customer AI agents not as an IT project but as a brand and customer experience initiative. Because agents interact with customers in conversational language, they need to be on brand – and that gives marketing a legitimate claim to lead or co-lead the deployment. Waiting for the IT department to define the agenda risks ceding control of the customer experience.

Second, organisations deploying AI agents need to unify their customer data before those agents can work together. The experience at Salesforce, where separate departments built agents drawing from separate data sources, is a cautionary example. Without a single source of truth, agents operating in silos will create fragmented customer experiences rather than resolving them.

Learn more: How AI recommendations can balance privacy with accuracy

Third, the shift in software pricing models toward consumption-based and outcome-driven structures means that marketers must become more fluent in commercial language. Communicating value will increasingly require articulating the business outcome a product delivers, not just its features. Product marketing teams in particular will need to be responsive to different pricing preferences across markets.

Fourth, the marketing qualified lead is no longer sufficient as a primary performance metric. The rise of buying groups, the proliferation of digital engagement channels and the move to post-acquisition metrics such as adoption and expansion all demand a more integrated approach to analytics. CMOs who tie their reporting to pipeline and business outcomes – rather than campaign vanity metrics – will retain influence with the C-suite.

Finally, the most durable skill in an era of AI is not technical execution but domain expertise. Professionals who understand their customers, their market and their product deeply enough to describe problems with precision and evaluate AI-generated solutions with rigour will be the ones who thrive. For hiring managers, this means placing greater weight on behavioural competencies, collaborative instincts and adaptability than on mastery of any single tool or platform.