Verticalisation refers to AI applications that are trained for specific industries and use cases. “If you’re in finance, as an example, you’re training these models to understand credit risk, how that process works, claims adjustment, things like that,” he says. “That’s what we see with verticalisation.”
For Boomi, whose platform connects enterprise systems and manages data flows across organisations, the rise of vertical AI is less a trend to observe than a commercial reality already reshaping its customer base across Southeast Asia. The companies succeeding, Irecki notes, are not necessarily those with the most sophisticated models; they are the ones with the cleanest data infrastructure underneath.
The gap between confidence and readiness
Ask most CEOs whether their organisations are AI-ready and the answer is typically yes. The reality beneath that confidence is often considerably more complicated. Irecki describes the mismatch as one of the industry’s most pressing near-term risks: not that AI will fail outright, but that it will appear to work while quietly producing unreliable outputs.
See also: AI push faces a two-speed reality
“I think the biggest gap is that AI works. We’ve seen AI work in many situations, for many different use cases, but it runs into problems,” he says. “Those problems tend to be because the business doesn’t have good data liquidity or good data quality within the organisation. Then when you look around scaling AI in your organisation, you have to worry about explainability and transparency. You need to start understanding lineage — how did the AI come up with that answer?”
A related and growing concern is what Irecki terms AI sprawl, or the proliferation of AI tools being adopted quietly across business units, often without the knowledge or oversight of central IT. A human resources team using an agentic model here, a sales or finance team plugging into a third-party AI there. The result is a fragmented, ungoverned landscape of models being fed company data with no audit trail. This could also lead to cumulative financial burden as users may not fully understand the cost of AI answering each query or executing each task.
“A lot of the conversations we currently have are around putting robust governance into your business before embarking on some of these AI projects — or at worst, putting governance in parallel, so you’re able to identify what these agents are doing in your organisation,” he says. “And if they’re making errors, hallucinating, and/or using excess token counts, you can either attempt to fix them on the fly or have a kill switch to shut them down and disable them.”
See also: Singapore embraces the AI revolution
To address this, Boomi has built what Irecki describes as an AI agent control tower — a governance layer capable of detecting anomalies such as unusually high token usage, drilling down to the specific AI agent responsible and either suggesting parameter adjustments or disabling the agent entirely.
Southeast Asia’s frontrunners
Across Southeast Asia, the sectors moving fastest on vertical AI tend to share a common profile: they are regulated, data-intensive and have well-defined workflows that lend themselves to automation.
Financial services leads the way. Banks and insurers are deploying AI for know-your-customer (KYC) processes and expanding into risk and compliance use cases. Logistics is another area of strong traction. Irecki illustrated this with an example from the stage at Boomi’s recent user conference: instead of calling an application programming interface (API) to a single carrier such as Singapore Post to arrange delivery, an AI agent can now contact multiple shippers simultaneously, receive responses in real time and determine which can ship a parcel at the fastest and cheapest cost at that point in time.
“That’s the kind of infiltration we’re seeing with AI,” he says. “That’s where we’re seeing the value of doing an agentic transformation of your workflows.”
Healthcare is progressing more cautiously, with back-end integration between electronic health records systems laying the groundwork for broader AI adoption. Higher education is also on Irecki’s radar: universities, particularly those that depend significantly on international students, are exploring AI to personalise the student experience in an increasingly competitive market. Retail, meanwhile, is emerging as a quiet achiever, with brands deploying AI for customer personalisation and predictive inventory management.
Not every sector is keeping pace. Manufacturing is lagging, hampered by the divide between operational technology and information technology systems and by entrenched legacy infrastructure. Small and medium-sized enterprises face a more fundamental challenge: many lack the data liquidity, data quality foundations and API coverage needed to make AI work reliably at all.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
Separating scalers from proof-of-concept purgatory
Across all sectors, Irecki returns consistently to the same answer when asked what distinguishes organisations that successfully scale AI from those that remain mired in pilot mode.
“For me, it always [about] having integrated data. So, having good data in your core and (customer relationship management) CRM systems with your third-party providers and having lineage if possible,” he says. “Because as you feed this data into these models and they have an outcome that might be a compliance risk, you need to understand how the agent determined that.”
Boomi’s approach to this problem centres on data synchronisation. When a record changes in one system — a customer’s address, for instance — that change is validated and, if approved, replicated into every application across the organisation. The result is a single, reliable view of each entity, ensuring that whether a customer service representative opens the CRM, a finance team member checks the invoicing system, or a warehouse manager looks up the warehouse management system, they are all working from the same data.
Beyond data quality, governance architecture proves decisive. Successful deployments set clear guardrails and human-in-the-loop thresholds from the outset. In practice, this means defining a confidence threshold — Irecki cites 80% as a working benchmark — below which an agent flags a decision to a human operator for validation before the process continues.
“The other thing for me is around people and culture,” he adds. “There’s a lot around people understanding how to use AI and how they can use it as part of some of those processes for compliance and audit trailing. And on the government side, we have things like the Monetary Authority of Singapore that’s trying to tell financial institutions to focus on AI governance — because if there’s a standard or an approach that’s a common way for interoperability, then we see that in the broader sense. Whether we get to something like the EU AI Act for our region, I’m not sure. But having interoperability between countries around how they share data and do things is important.”
AI readiness as a valuation factor
One of the more consequential observations Irecki offers is the degree to which AI and data readiness is beginning to shape enterprise valuations and mergers and acquisitions (M&A) activity. Acquirers, he says, are placing a growing premium on targets with clean data estates, strong integration coverage and documented governance frameworks.
“I think probably the key point I see is that, as you’re looking at M&A now, businesses value companies that have clean data states — their data is solid, they have good integration, they have API coverage — because then when you bring those businesses into your own, it’s going to be a lot easier to integrate them,” he says. “And equally, you might get some further competitive advantage, because if they have laid that foundation and the business you’re acquiring is, potentially, already running AI agents and models on top of that foundation, you might be buying capability that you don’t have in the AI space as well.”
Irecki expects this to be formalised over time, with data lineage, governance policies and AI infrastructure treated as standard due diligence items, much as cybersecurity posture is today.
Start with governance
For companies that want to adopt AI but have not yet built the integration backbone to support it, Irecki’s counsel is deliberate and unglamorous. He describes himself as a governance-first person, someone who believes the architecture of oversight should be established before the AI, not retrofitted afterwards.
“Find those use cases that provide some immediate value to the business and that you can grow from, almost like a crawl, walk, run scenario,” he says. “If you do find those one or two high-value use cases, when you scale, you’ve got to be mindful of observability and governance and ensure that people are adjusting for drift and things like that, because you want these models to continue to provide value to the business.”
For Irecki, the long view offers a clear enough precedent. “We’ve been in this position before with very transformative technology — maybe it’s overhyped, or maybe the bubble does burst, but the underlying technology continues on,” he says. “We saw this with the internet and the dotcom bubble. The internet is still here and it’s evolved and it’s found a key foundation for everything we do in our lives today. And we’ll see the same with AI.”
