Boomi chairman and CEO Steve Lucas. Lucas is in Sydney for Boomi World Tour, the company’s flagship Australian event, on September 2.Michael Barnes, chief analyst, enterprise IT at global technology research firm Omdia, says some businesses started experimenting with AI without really knowing what success would look like.Experimenting with new technology is crucial, but only up to a point. “Fear of missing out is not a good enough reason to undertake large-scale AI initiatives, but that’s still a major driver for a lot of firms, including a lot of Australian firms,” he says.“The way around that, how do you deliver more measurable returns? You start by asking a really basic question: what are we hoping to get out of this initiative?”From pilots to the P&LThe answer is changing the way companies fund AI.Barnes says businesses are putting less emphasis on poorly defined enterprise-wide initiatives and more on applying AI to individual functions where the result can be measured.That might mean using AI to improve the productivity of a customer service team, automate part of a finance function or perform a defined task more efficiently. The measure of success depends on the job and the expected outcome rather than the amount of AI deployed.“We’re seeing less of a focus on enterprise-wide AI rollouts,” Barnes says. “It’s much more about domain or function-specific AI-enabled value delivery.”Michael Barnes, chief analyst, enterprise IT at Omdia. Those individual deployments still need company-wide controls over governance, risk, cost and data.Lucas makes a similar distinction between deploying AI and producing a result. The companies that succeed will not necessarily be those running the most pilots.“It’ll be the companies that can connect AI to their P&L,” he says.That means examining existing processes and asking whether they would be built the same way if designed today. Lucas points to functions such as accounts payable, general ledger activity and market research as examples where businesses can start with the desired outcome and then decide how AI should be used.Barnes says many initiatives stumble because companies have not established what they are trying to measure.“You can’t measure it if you can’t even name it,” he says.Experimentation tends to be driven by proving the technology works. Putting AI into daily operations raises a different set of questions.“As organisations start to think about operationalising AI, that requires a very different approach,” Barnes says. “That’s where issues of cost, of control and of risk all become central.”The value of what companies already ownAccess to powerful AI models is becoming less of a differentiator as businesses gain a wider choice of providers and models.What competitors cannot readily reproduce is the information accumulated inside a business: its transactions, customer relationships, operational knowledge and intellectual property.Barnes says that makes corporate data central to the commercial value of AI.“Ultimately, creating real value from AI comes from leveraging AI in context,” he says. “What’s the context? It’s your organisation’s data, it’s your insights, it’s your customer knowledge and relationships. It’s your institutional knowledge as a firm. That’s where the real differentiator still lies.”Boomi’s research found 94 per cent of Australian organisations view integration, data access and governance as a key priority.Lucas says data only provides that advantage when it can be accessed reliably by AI systems.“Data only creates that advantage when it’s connected, easily accessed, trusted, governed and of high quality,” says Lucas.Boomi describes this as data activation for AI. Its Enterprise Platform connects applications and data and provides tools to manage and orchestrate AI across existing business systems.Orchestration matters because businesses may not need the largest or most capable model for every job. Different models can be assigned different tasks according to cost, capability and the information they need to access.“AI models provide intelligence, but orchestration of that intelligence creates an outcome,” Lucas says.That leaves companies deciding how much AI capability to buy from outside providers and how much to keep under their own control.Counting the costThe price of AI makes that decision more pressing.AI services commonly charge according to usage, including the number of tokens processed by a model. As deployment spreads through a business, usage-based pricing can make future costs difficult to forecast.Barnes is unequivocal about the problem.“Token costs are untenable. They’re simply too high, they’re too unpredictable,” says Barnes.At the same time, businesses are still trying to establish what they receive for that spending. Barnes says the answer is not to abandon external models but to match the technology to the job.“It’s all about horses for courses,” he says. “It’s about choosing the right model to meet your needs, balancing cost constraints against desired outcomes.”A company performing a relatively simple task may not need to pay for the world’s most capable model. Barnes says organisations can instead consider whether a “good enough” model gives them the performance, cost and control required for a particular job.He also expects the huge investments being made by frontier AI companies ultimately to affect what customers pay.“The investments that have been made are astounding by any measure,” Barnes says. “Those investments need to be recouped in some way. That’s just the way the world works.”High and unpredictable usage costs are therefore pushing companies to consider using a mix of models selected according to need.“It certainly pushes organisations towards looking beyond frontier models towards open weight and even open source alternatives, where they’ll have more control and more predictable costs,” Barnes says.Lucas takes the argument further. He warns businesses against becoming dependent on a single frontier AI provider when neither the eventual cost of the technology nor its pricing structure is settled.“We don’t know what the true cost of AI is,” he says.His “Not rent it. Run it” argument, outlined in Boomi’s recent white paper, Economics Always Wins, is not that companies should build every AI model themselves.Rather, businesses should retain control of their proprietary data and decide which models should be used for particular workloads instead of becoming dependent on one provider.“The bottom line is, economics will win out here,” Lucas says.Australia’s fast-follower advantageBarnes thinks Australia’s traditional caution on technology adoption could work in its favour.Australian companies have historically been pragmatic adopters of technology rather than determined to be first at any cost. In a field developing as quickly as AI, waiting long enough to learn from early adopters can be an advantage.“I would still classify Australian organisations as fast followers,” Barnes says. “This space is moving incredibly quickly, so there’s nothing wrong with being even a few months behind. It’s potentially not a bad place to be to learn from what else is happening elsewhere.”The discussion has already changed markedly from two years ago, when many businesses were still testing what generative AI could do. Barnes says the organisations he speaks to are now asking how the technology can perform specific jobs and deliver defined outcomes.Barnes says companies are now concentrating on cost, measurable value and control. Data sovereignty is also gaining attention as businesses consider where proprietary information resides and how it is used by AI systems.For CFOs and boards, the next stage of AI is less about keeping pace with the technology than determining where it earns its keep.The winners may not be the companies with the biggest AI budgets or the most pilots. They will be those that can put a number against the result.To find out more about the economics of AI and how organisations can take greater control of their AI costs, data and infrastructure, read Boomi’s white paper, Economics Always Wins.