Building an AI-Powered Business in 2025: Six Key Questions for Executives

This white paper aims to help CEOs, their leadership teams and boards articulate the strategic preferences essential for crafting an effective artificial intelligence (AI) strategy. It highlights the use cases that deliver real ROI at pace, both for efficiency gains and pursuing real competitive advantage.

Our goal is to support Australian and New Zealand businesses to move beyond experimentation to genuine AI-powered business transformation as we shift from the ‘art of the possible’ to the ‘art of the practical'.

The race is on to deliver on the promise of enterprise AI. Global AI spending is expected to reach $632 billion by 2028, according to IDC . McKinsey estimates that AI will generate $2.6-$4.4 trillion in value annually, and the vast majority of that value, 75 per cent, is expected to come from front-line functions, such as customer operations, sales and marketing. McKinsey also estimates that AI could boost Australian labour productivity by up to 1.1 per cent a year by 2030.

We believe the most pressing challenge right now is to determine what role AI should play in next year’s operating plan. Enterprise leaders need to secure clear, well-informed answers to the following six questions in order to build resilient future-proofed organisations that can both endure and adapt.

1. What can AI do today?
2. What kind of infrastructure is necessary for AI to work?
3. What kinds of use cases warrant consideration?
4. How can “trapped value analysis” help prioritise our options?
5. Where is the ROI and what are the critical metrics?
6. How important is it to act fast?

C-suite executives in Australia and New Zealand are feeling pressure from all sides. Boards expect to see how all this experimentation will materially impact the fundamentals of the business. Customer expectations are rapidly shifting in response to the experience of consumer AI, making traditional processes seem clunky and unnecessarily bureaucratic. Meanwhile, the absence of industry best practices makes it hard to know how much AI is enough, to keep up with competitors.

Over two decades ago, Salesforce pioneered the first multitenant cloud platform, setting a precedent in the industry. Since then, Salesforce has significantly expanded its footprint, and now serves over 150,000 customers from various industries and regions.

Four years ago, Salesforce embarked on a mission to completely transform its platform from the ground up. The launch of Agentforce at Dreamforce 2024 represented the culmination of this extensive effort.

With the groundwork laid, we now have the right conditions for accurate, governable and sustainable enterprise AI use cases.

Any AI system is a combination of UI + Model + Data. Collectively, the features of our platform above materially impact the viability of enterprise AI, by improving the performance of AI systems at every level.

• Surfacing results and responses in a timely way within the flow of work, unlocking value from prior investments (e.g. in predictive AI), and ensuring customer and staff engagement with future investments.
• Similar to the way an iPhone is intuitively usable out-of-the-box, configurable prompts embedded into the UI avoid the need for every user to have sophisticated prompt engineering skills. This lowers the costs of re-skilling and governance of generative AI.
• Empowering agents with an extensive range of standard and custom-built workflows that can be combined into sophisticated tasks with real value for customers or employees, and can be completed in a predictable and repeatable way.
• Zero copy integration with any enterprise data source allows AI and especially Agents that can be informed by the full breadth of intelligence across the organisation, whether stored in Salesforce or another system or data platform.

• Grounding prompts in data that is highly relevant, leveraging customer data within Salesforce, and making data sources and streams accessible and intelligible through Data Cloud.
• Grounding prompts in a metadata context that is highly structured, providing the model with a company-specific and industry-specific conceptual scheme to help it understand meaning, and discern what matters, when interpreting a prompt.
• Providing a secure gateway architecture that allows selection of the right model for the job, while ensuring privacy and security with zero retention, toxicity detection, and other features of our Trust Layer.

The key point here is that because the missions for each zone and their metrics of success are so different, it is critical for executive management to put each AI initiative in service to one and only one zone. When two or more zones both want to put their hands on the tiller, the boat tends to travel in circles.

Despite almost every company making significant data strategy investments, 94 per cent of business leaders feel they should be getting more value from company data. Only 39 per cent of companies believe they are successfully turning data driven-insights into competitive advantage.

The challenge lies in the complex data strategies companies are forced to employ. With an average enterprise using over 1,000 applications, those who have attempted to organise their data develop massive and complex data ecosystems. While successful in centralising data from an array of sources, these ecosystems are difficult to operationalise as the data remains static and stuck in back-end solutions not designed to activate data within the everyday workflows and applications business teams use to engage with their customers. They fall short at the “last mile” of activation, leading to poor customer experiences driven by disconnected or inaccessible data.

Our customers have shared stories that illustrate this problem more clearly. From marketing campaigns that miss their target due to delayed data, to service alerts that blanket entire regions instead of focusing on only the affected customers, to sales reps from the same company unknowingly cold-calling the same buyer multiple times in a day, these issues stem from a lack of accessible, actionable data.

Data Cloud connects all of a company’s data to Salesforce's foundational metadata layer, offering a shared language used by all applications, unlocking a range of user-friendly tools that make data easier for everyone to work with, especially AI agents.

Focusing on the data problem can, however, drive companies into a monolithic approach to building AI capability, or delay it indefinitely. 'Let's solve our data problem first, then we’ll start our AI transformation’, is a tempting strategy, but it delays action, delays learning and workforce engagement, and strongly biases the organisation to invest only in AI initiatives that fit within the Productivity Zone.

Organisations can get initiatives in other zones moving faster, and can unlock the power of a multi-speed transformation, by thinking in terms of data products. You can think of a data product like a small brain, specifically created for an AI agent, so that it knows just what it needs to know to do its job. The key, however, is not to build a separate data infrastructure for each of these brains, but to virtualise that infrastructure, and the associated metadata, through Data Cloud.

One of the biggest benefits of this approach is this: just as Agent Builder empowers business users to take responsibility for defining an agent’s role and instructions, Data Cloud allows those same business users to share responsibility for the relevance and reliability of the data that informs that agent’s behaviour. Here we are applying a principle we’ve lived by for over 20 years: you get faster, more effective digital transformations when you bring control closer to the business and closer to the customer, and you do that by enabling configuration through clicks, not code.

Even better, this approach frees up rare and expensive data science and engineering resources to focus on those incubation zone initiatives that are truly unique to your organisation’s competitive advantage.

The answer to this question depends on how far the disruptive innovation has progressed through the technology adoption life cycle in your geography and industry. There are four stages in the life cycle that matter, and each one comes with its own imperatives:

This is a time when disruptors need to act fast in order to gain first-mover advantage. Exploiting this kind of opportunity is best suited to start-ups funded by venture capital and unencumbered with any other obligations. It is not critical for established enterprises to participate at this time, but if there is interest and bandwidth, they can activate their Incubation Zone to explore the possibilities.

For a handful of use cases, it is important to act fast, but not for technological reasons. Rather, because there is an urgent need to address a mission-critical business process problem and the current solutions are simply not cutting it. Start-ups can focus on a single use case in a single industry in order to put their business on a solid footing, no longer dependent upon raising another round of funding. Established enterprises can free their operating model from a bottleneck that is bringing down their overall performance, initially by tackling it in their Incubation Zone, and then - once the solution has proven out - transitioning to either the Performance Zone or Productivity Zone, depending on where the bottleneck resides.

This is when the technology has caught on, adoption is accelerating like crazy, and either you are going to catch the wave or the wave is going to catch you. At this stage, everyone must act as swiftly as they can. Vendors need to capture as much market share as possible, and customers need to modernise their operating models as fast as possible. The pressure is on the Productivity Zone to accelerate deployment and adoption, and on the Performance Zone to deliver value realisation. If inertial resistance is stubborn and you are falling behind, this warrants activating your Transformation Zone.

The technology has now become reliable infrastructure, and the urgency to act is replaced with the opportunity to optimise. This is a great time to do a comprehensive end-to-end trapped value analysis of your entire enterprise, leveraging Six Sigma and DMAIC disciplines, to identify and target processes that have been holding you back for years. Here the Productivity Zone can shine, and over the technology adoption life cycle, the bulk of the ROI is likely captured at this stage.

that have a clear business benefit, and move the organisation forward in its transformation. AI systems should be judged not by the number of parameters they hold, but by whether they remove productivity bottlenecks and deliver unexpected value to customers.

There are no free lunches, especially in this economic and environmental climate. It’s crucial that the AI systems you choose enable you to pick the right combination of capabilities — whether it’s predictive, generative or agentic AI — along with the right model, prompt augmentation, and UI for the job.

We’ve already seen examples of solutions delivering significant cost savings by switching back to a predictive model for a resource-intensive component. We’ve also seen examples of solutions that avoided an expensive governance challenge by changing the UI to avoid it altogether. Choose the system that will allow you to balance accuracy and power against the total financial and carbon cost of ownership.

No-one knows better than your employees where sophisticated and powerful technologies add significant value, and where a simple workflow is good enough. The sooner you engage them in refining the problem to be solved, the sooner your AI investments may start generating outsized returns, and you build trust with employees along the way. Inviting a wider employee community into the process is also a great way of mitigating the risk that can arise if there is less diversity in your data science team.

Start the way you want to continue, by exposing employees to what is possible, as early as possible, in the actual context of use. In doing so, you’ll not only deliver value faster; you’ll learn what sort of accuracy‌ matters, and what level of autonomy is manageable. By doing this, you’ll encourage a more intelligent backlog of use cases by building a familiarity with common technical or data constraints.

Your AI systems should help protect the integrity of your data, and your customers’ privacy. They should make it easier to enforce roles, scope and guardrails for agents. They should also allow you to create the kind of mindful friction that will encourage safe and responsible use at critical decision points.

so that AI responses and actions can be informed by what the data means and which data matters to the business and its customer relationships. Ensure that the data informing the agent or assistant responses is transparent and intelligible to the business unit that the AI system is serving.

To recap, here are the six questions you need to develop clear answers to in order to incorporate AI effectively into a 2025 operating plan:

1. What can AI do today? Specifically, what can it do that might be material to success in the coming year? Who can we talk with to get a deeper understanding of our opportunities?
2. What kind of infrastructure is necessary for AI to work? Do we have this infrastructure in place today, or do we need to invest, and if so, in what exactly and how much will it cost?
3. What kinds of use cases warrant consideration? Specifically, have we identified one or more use cases that could materially impact our success in 2025, and if so, what are they?
4. How can “trapped value analysis” help prioritise our options? Where should we be targeting trapped value in our operating model today, and how big a boost would we get from releasing it?
5. Where is the ROI and what are the critical metrics? Specifically, what zone will host our AI initiative, and to what metrics are we going to hold ourselves accountable?
6. How important is it to act fast? Where do we think we are in our local technology adoption life cycle, and based on that conclusion, what time-to-value objectives do we intend to set?

The advisory services and account management teams at Salesforce are deeply engaged in helping our customers embrace AI as part of their core infrastructure. They are experts at answering these questions and pressure-testing plans to apply AI to trapped-value use cases.

We would welcome an opportunity to engage with your team to help you make your plan a huge success.

• Employees lacked visibility into customer interaction history, resulting in longer, less efficient service calls.
• Technicians had limited job specifics, making it difficult to provide precise appointment times.
• Professional customers had restricted interaction methods outside business hours for tasks like inventory checks and orders.

Fisher & Paykel is using Salesforce's AI, data, and CRM tools to enhance customer experiences:

• Agentforce will automate simple tasks such as answering FAQs, processing orders and scheduling appointments, allowing employees to focus on building customer relationships and driving sales. Agents will provide technicians with job details for better planning and accurate appointment times.
• Data Cloud unlocks trapped customer data, creating a complete customer view for accurate agent outputs. Marketing Cloud uses this data to create personalised marketing segments, automated customer journeys, and more relevant customer communications. In 2023, targeted emails saw a 206 per cent increase in unique opens and a 112 per cent increase in unique clicks.
• Service Cloud offers global customer support, linking to the call centre for instant customer insights and seamless conversation continuity, reducing service rep training time by 76 per cent. Agents will be able to book on-site technicians during calls using Field Service.

• Legacy systems hindered daily experiences for tradies and internal teams.
• Lengthy onboarding and customer service wait times.
• Siloed data and no operational insights.

By leveraging Salesforce to drive a data and AI transformation hipages can:

• Reduce tradesperson onboarding time from 60 to 10 minutes using Agentforce
• Use Agentforce to pilot an autonomous Onboarding Agent for rapid license checks, saving time and building trust.
• Equip customer-facing teams with AI tools and rationalised legacy systems using Service Cloud.
• Consolidate data and gain visibility into operational data using Data Cloud.
• Create a single view of product engagement via Data Cloud for more connected conversations.
• Use AI to enable customer-facing teams to make retention offers and package recommendations.

A consultant to the high-tech sector for over 35 years, Geoffrey Moore is a best-selling author, renowned public speaker, and trusted advisor recognised for his profound understanding of dynamics of disruptive innovations and the strategies to capitalise on them. He balances his consulting between venture-backed start-up companies and established public enterprise.

As an author, Moore has published seven best-selling books addressing the marketing, strategy, and organisational challenges that are triggered by disruptive innovations. His first book, Crossing the Chasm, published in 1991, has sold over 1 million copies, been translated into more than a dozen languages, and to this day, now in a third edition, represents the go-to-market playbook for entrepreneurs seeking to grow their customer base beyond the early adopters to the pragmatist mainstream. His most recent work, Zone to Win, written in collaboration with the executive teams at Salesforce and Microsoft, addresses the challenges publicly held enterprises face, balancing commitments to profitable but aging legacy lines of business with promising next-generation opportunities that require deep investment to get to scale.

As a keynote speaker, Moore delivers between 20 and 40 speeches each year, specialising in thought leadership topics that help audiences better grasp the potential impact of emerging technologies on their industries and enterprises. The phrases and metaphors he introduces have become part of the language of high-tech one often hears about: chasms, tornadoes, systems of engagement, or the hierarchy of powers.

As an advisor to CEOs and their executive teams, Moore has worked with some of the most iconic high-tech firms, including Adobe, Autodesk, Cisco, HP, Microsoft, and Salesforce. He has also helped emerging companies rise to prominence, including Aruba, Box, Docusign, Gainsight, and Proofpoint. Advisory efforts include strategy check workshops, thought leadership narratives, executive development, and organisational design.

This report is for informational purposes only and is subject to change. While we endeavour to assist our customers in their AI transformation journey, Salesforce, its subsidiaries and employees, and Geoffrey Moore, (collectively, ‘We’) do not warrant or guarantee the accuracy, suitability or completeness of any information, links, or other items in this report. To the extent permissible under applicable laws, We exclude liability for any representations or warranties made. As this is not legal or professional advice, we encourage you to obtain your own prior independent legal and professional advice, and to make your purchase decisions based on your unique needs and the Salesforce products and features that are currently available to you.

The report contains the trademarks, confidential and proprietary information of Salesforce, Inc, its subsidiaries and Geoffrey Moore, the respective owners reserve and retain all rights, titles and interests. Salesforce customer stories reflect the experiences of actual customers, but your experience may vary.