Australian enterprises are handing the plumbing of their integration work to artificial intelligence (AI) while keeping a tight rein on where agents themselves are allowed to act.

Take Cochlear, for example. The supplier of hearing devices is doing 70% of the Boomi work behind its enterprise source planning (ERP) transition with the vendor’s AI assistant, turning jobs that once ran to months into days or hours.

In addition, brick manufacturer Brickworks has been using connectors to link applications to master data, while Bakers Delight has opened online ordering at 20 stores, thanks to an integration layer that took a few days to stand up.

Agents are a different proposition. Those who spoke at the recent Boomi World Tour earlier this month in Sydney were consistent about what is holding them back – data that is clean but stripped of business context, governance bolted on after the fact, and runaway token costs.

Brickworks CIO Keiren Lawrence said the company had been a Boomi user for years, though previously only for point-to-point integrations. Around two years ago, it went through a technology transformation aimed at making data more trustworthy, and “Boomi was a big part of that”.

Brickworks started using connectors to link applications to master data. Connector reuse rose 6.4 times, and the cost of a connector fell to a seventh of what it had been. Aggregate use of the Boomi platform grew 700%, but costs went up only 50%.

The most recent project connected the product information management system to the content management system behind the website – no small task, given that Brickworks juggles multiple brand names and complicated product hierarchies. Separately, a change from local to centralised procurement has been accompanied by improvements to invoice ingestion that save around A$4m.

“There’s still a lot to do,” said Lawrence. Plans include using Boomi Data Integration to replace some of the traditional data flows, simplifying processes without adding risk or weakening governance.

Engaging Boomi partner Atturra has helped Brickworks innovate, she added.

“Everyone wants stuff done faster,” said Jason Frost, executive general manager of data and integration at Atturra. Boomi Companion – which lets users build and manage integrations using natural language – is a game changer, he said, but there is still room for partners to add value.

Brickworks is taking a “purposeful and targeted” approach to AI, according to Lawrence. Concerns include the risk of cost blowouts, and the resulting uncertainty over payback.

Two applications under consideration are enhancing the existing ERP system by matching payments with invoices, and ingesting sales orders that arrive as PDF files or emails.

Business processes and data should be standardised before automation is added, said Lawrence, and it is worth remembering that “people bring significant value … to the organisation”. They need time to learn new things and adapt to change, so pushing ahead too quickly means the human side will slow things down.

Governance should not be a bolt-on, but something the platform supports, said David Ireck, Boomi’s chief technology officer (CTO) for Asia-Pacific and Japan. That includes routing queries to the right models and carrying an individual’s access rights through to the agent working on their behalf. Agents will try any route they can to reach the information they need to complete a task, he warned, so security and access controls matter more than ever. As Computer Weekly has reported, there is an important difference between delegating authority to an agent and allowing an agent to impersonate the user.

As in some other areas of technology, ANZ businesses are fast followers, he said, with many waiting to learn from others’ mistakes. Those include failing to put the right data and security foundations in place, lacking the governance mechanisms to keep agentic AI under control, and spending too much on tokens.

The work done by chief data and analytics officers in recent years to clean up their organisation’s data has been necessary, but not sufficient, for AI projects, said Irecki. Cleaning data into a “golden record” and replicating it to the systems that use it is one thing; agents do not necessarily understand the business context. An “active customer” might be defined in one system as someone who has made a purchase in the past 180 days, and in another as someone who has visited the website in the past 30 days. Unstructured data throws up similar problems.

Boomi’s forthcoming Knowledge Hub will connect to structured and unstructured data sources and supply context, security and governance for agentic and retrieval-augmented generation (RAG) applications, addressing most of those issues, said Irecki.

Bakers Delight has just started taking online orders for pickup or delivery via Uber from 20 stores, to compete with the convenience of supermarket home deliveries.

The average in-store basket is worth around A$10, but one of the first online orders came to more than A$50, suggesting the anticipated incremental sales will materialise. Customers are being encouraged to place orders the night before so the relevant store can adjust its bake quantities.

Two things are expected to drive online pickup orders: customers buying larger quantities than usual – say, dozens of rolls for a social event – and shoppers making sure a favourite product has not run out before they reach the store, perhaps on the way home from work.

Online ordering has been stealth launched, and Bakers Delight will not promote it until it is widely available across the store network, said CTO Keng Ng.

The combination of sales (Shopify), fulfilment (Deliverit), loyalty (Talon.One) and customer segmentation (Klaviyo) software led the bakery franchise to use Boomi as an enabler and accelerator of integration.

“Retailers will always underinvest in technology,” said Ng, as they want Qantas service at Jetstar prices. “Franchisees don’t care what’s behind the scenes,” as long as it works and does not cost any more. Boomi was affordable, needing only a few days’ work to implement, and its no-code system means there is little variation between the work of the best and worst users. That said, engaging Boomi professional services was money well spent, he said.

The new environment means Bakers Delight can move away from “carpet bombing” its customers towards a more targeted approach to promotions. It knows that different customer groups behave differently – mums, dads and kids among them – and that there is little point promoting products containing bacon in some neighbourhoods. Geolocation, used with the customer’s permission, opens up further options: a previously regular customer who has not bought anything lately but can be seen in the vicinity of a store could be offered a A$1 discount voucher valid for an hour.

The updated systems provide real-time rather than next-day visibility of customer behaviour, giving the company the ability to react within the day, said Ng.

Plans for the future include closer integration between the point-of-sale and marketing systems to support more personalised conversations between store staff and customers, and to enable AI-driven upselling suggestions. That will require customers to scan their loyalty QR codes earlier in the transaction rather than at the time of payment – something that could be encouraged by having the scan unlock members-only prices.

Cochlear is in the middle of an ERP transition, and 70% of the supporting work in Boomi is being done with the aid of Boomi Companion, said Tim Woo, head of digital solutions at the hearing implant maker. Tasks that would previously have taken months are now being completed in days or hours. It took only three days, for example, to create an agent that has been able to handle 30% of incoming calls.

Woo reminded his peers to be selective about the application of generative AI. Some things can – or should – be done deterministically, and Cochlear is using a combination of both approaches at Boomi.

Plans include running models locally, both third-party and Cochlear’s own. It is an uncertain world, but Cochlear needs to be around to support its patients for the rest of their lives, said Woo, so financial and environmental sustainability are important.

Irecki noted growing interest in running open-weight models on premise as a way of improving the returns from AI projects. That has sovereignty benefits, he observed, as it involves moving the model to your data rather than the other way round. But Boomi chief product and technology officer Ed Macosky warned that “you can’t go all-in on open weight”, as some tasks need the advanced models available from hyperscalers.

• Nucleus Network taps AI agent to screen clinical trial volunteers

  

  By: Aaron Tan

• Trust Bank cuts incident triage time to two minutes with AI agents

  

  By: Aaron Tan

• Token bills to push most enterprise AI workloads onto open-weight models

  

  By: Aaron Tan

• UiPath’s Cartographer agent interviews staff to map how businesses run

  

  By: Aaron Tan