Salesforce busts AI myths

Artificial intelligence is entering a more critical phase, as organisations move beyond early adoption and are testing whether investments translate into real business value.

This was a focus of a recent virtual media briefing held by Salesforce and Cloud23, which examined how companies assess AI strategy, including the assumptions shaping investment, infrastructure, and implementation decisions. Cloud23 is a South African technology advisory firm that helps organisations implement and optimise Salesforce’s global suite of cloud-based CRM and AI solutions.

Myth 1: Market valuation is an adequate way to determine if a company’s AI strategy is solid

“If you’ve seen the markets, it’s a wild ride out there,” said Salesforce Africa lead Linda Saunders. “Share prices are all over.”

Market volatility has made investment decisions more uncertain, she said, with share prices fluctuating across the sector. Organisations that once relied on stable valuations from established software companies now face a more complex environment when assessing technology investments.

She pointed to Allbirds as an example, noting that the footwear company saw stock decline before securing $50-million to invest in GPUs and reposition towards AI. The shift led to a sharp increase in share price of around 600% in a single day, raising questions about how investors assess credibility in the AI space. She described this as part of a broader trend of “AI washing”, where struggling businesses adopt AI positioning, creating risk for investors as the market continues to determine how to value such transitions.

(See Gadget’s perspective on the Allbirds AI washing here.)

The presence of AI in a company’s positioning, said Saunders, is not enough to determine credibility, and that organisations need to assess how AI is being applied and the journey taken to develop those capabilities.

“It’s not just about whether there’s AI in the title. It’s also a lot about competitiveness and the journey that the organisation has taken with AI.

“It’s hard to ignore the share prices, but the market will eventually figure out how value is created in the segments. Right now, nobody really gets it, and we’re all figuring it out.”

Myth 2: Real AI companies own the foundation models and a data centre full of GPUs

Investment in AI infrastructure is accelerating, with estimates pointing to trillions being spent on data centres and compute capacity. This level of investment creates pressure on companies to generate returns, influencing how AI systems are designed.

“What they’re basically doing is they’re saying, ‘we’re creating a lot of this token capacity’. Now, if you have a shareholder and you don’t return that money in terms of the investment, they make you do all sorts of crazy things to get back to invest value,” said Saunders.

This pressure can shift focus towards utilisation, rather than outcomes, as companies look to recover the cost of infrastructure.

“I like to call it token maxxing. What it means is in the solutions, they’re making sure that they’re building out solutions that maximise token consumption, so that they can get a return on investment.”

Rather than focusing on ownership of infrastructure, she pointed to an approach centred on measuring what that compute delivers in practice.

“What we’ve done is look at our token consumption. We’ve said, ‘Great, you’ve consumed the service. How do we take that token consumption and turn it into something meaningful from an organisational perspective, so we can understand whether we are getting the ratios right?’”

At the same time, rapid advances in large language models are reducing differentiation between providers. Capabilities are converging, costs are declining, and switching between models is becoming easier.

“The view here is that LLMs and many of these models are actually a commodity.”

This shift places greater emphasis on flexibility and risk, particularly the ability to move workloads between providers as the market evolves.

“I’ve asked a lot of CEOs and CIOs in the last few weeks, ‘What if somebody did something crazy at OpenAI and the business shut down tomorrow? How flexible is your organisation in moving that workload somewhere else?’”

Ownership of infrastructure is not necessarily required for AI strategy, with flexibility and outcomes emerging as more important considerations.

Myth 3: Buying a technology solution – a new AI/CRM tool to plug in and get quick wins

The assumption that AI can be implemented through a plug-and-play solution is not reflected in current outcomes, with most initiatives failing before reaching production.

“If we zoom out and we look at what’s been happening in this sort of AI space, 95% of AI pilots are failing,” said Saunders.

Organisations are increasingly examining why these failures occur, with technology forming only one part of the challenge.

“The platform is a critical component to whether or not that pilot will be successful. The wrong platform, and for sure, you will be a statistic.”

Organisations need multiple systems in place, she said, starting with access to reliable, actionable data and clearly defined processes.

“For AI to work, it needs a system of context with credible, clean, actionable data. It needs a system of work that links to processes and gives the technology a defined role, especially in the agentic world.”

This requires defining how AI operates within the organisation, rather than assuming all tasks should be handled by advanced agents.

“When we’re talking about automation and looking at your organisation, don’t automatically assume everything is an agent. Everything doesn’t need to be an agent. Some processes might be two or three steps back from that.”

Failure rates are also linked to organisational factors such as change management, risk, and business design, rather than the technology itself.

“60% of AI projects risk being abandoned, not because the tech’s wrong, but because we under-index on what it means to become an agentic organisation.”

She said successful implementation requires a structured and deliberate approach, rather than relying on technology alone to deliver results.

“Buying a technology solution, it’s really not just plug and play. There’s a lot of intentionality that goes with it.”

This challenges the idea of plug-and-play AI, reframing adoption as an organisational shift that requires changes to how work is designed and executed.

Beyond the myths: What actually drives success

AI has introduced a virtual workforce, which can effectively remove traditional capacity constraints.

A week prior to Saunders’ presentation, Marco Hernansanz, Salesforce EVP and GM for Southern Europe, Middle East and Africa, told Gadget that success still depends on fundamentals such as strategy, customer understanding, and execution.

“When you think about what makes a company successful, you need the right strategy, you need to know your customers well, understand their needs, and then execute well,” he said.

According to Hernansanz, wider access to AI tools is lowering barriers to entry, increasing competition, and shifting advantage beyond technology itself. Instead, differentiation lies in how organisations understand their markets and apply AI within their operations.

This simplifies execution but does not reduce competition. Instead, pressure shifts to areas such as positioning, customer insight, and brand strength. AI becomes part of a broader competitive equation, where outcomes depend on how effectively organisations combine technology, data, and execution.

* Jason Bannier is a data analyst at World Wide Worx and deputy editor of Gadget.co.za. Follow him on Bluesky at @jas2bann.