Asia-Pacific M&A activity is in full swing. But a critical problem lurks beneath those heady numbers. Most organizations rushing into these deals have no idea what AI systems they’re actually acquiring — or what shadow AI already runs inside their own walls.
David Irecki, chief technology officer and senior director of solution consulting for APJ at Boomi, frames the issue: “Despite 30 to 40 billion [dollars] being invested in generative AI, 95% of enterprise AI pilots delivered no measurable business impact.” He’s referencing the MIT study that identified integration — not algorithms or compute power — as the primary failure point. And when two companies merge, this integration challenge multiplies to the nth power.
Stranded assets and interoperability deficit
The problem really starts with an unruly data architecture. “When they bring these deals together, both businesses have very different systems. They have very different data models and have very different AI approaches,” Irecki explains. In APAC, cross-border deals compound these issues with regulatory compliance requirements and varying levels of legacy system complexity.
It’s no surprise that data quality and integration issues consistently rank as major barriers to AI adoption. The same challenges can become catastrophic in M&A scenarios where two fragmented data estates collide.
Take healthcare, for example, which, together with fintech, is experiencing heavy M&A activity. The problem manifests as clinical systems, patient portals, and administrative platforms, each running separate AI implementations. None are sharing patient identifiers or treatment protocols. Then consider this likely scenario: multiple AI-powered scheduling systems across merged clinic networks, each optimizing for different variables (provider availability, equipment utilization, patient preferences) with no unified logic. The result: patients receive conflicting appointment options, and AI systems actively compete for resources.
Data integration is not an unknown issue. Yet many organizations still treat AI readiness as optional in M&A. But “if we want to actually get real value out of AI and a real return on investment, data liquidity and data quality are two key pillars,” says Irecki.
Unquantified technical debt (UTD) and governance leakage
Another area where M&A due diligence often overlooks the challenge of shadow AI. It’s a fact that most organizations cannot inventory their own AI deployments, let alone assess a target company’s AI landscape. “AI is really democratized across organizations, so it’s no longer confined to IT,” Irecki warns. “There are a lot of people out there who swipe the credit card, use ChatGPT in their own world, and you don’t know what data is going out.”
Line-of-business teams build agents to automate processes. Marketing deploys generative AI for content. Sales uses AI for forecasting. These shadow AI systems operate outside central IT oversight, creating ungoverned sprawl. When acquisition discussions begin, the acquiring company often has no systematic way to detect these systems.
Irecki draws some similarities to how APIs were treated: “I just see so much similarity between the two worlds.” Just as organizations spent years discovering rogue APIs, they now face the same challenge with AI agents. “You would have to understand what agents are out there, because it’s a black box until you get into it.”
The governance implications extend beyond mere inventory. Different regions implement AI governance differently. A Western company acquiring a China-focused business faces fundamentally incompatible approaches to data handling and model deployment. Without unified governance frameworks, merged organizations cannot ensure compliance, manage risk, or prevent agents from accessing inappropriate data.
EBITDA erosion: Contradictory insights and customer 360 failure
Even when organizations successfully catalog their AI systems, a deeper problem emerges: most AI implementations lack the context-rich data required to deliver value. “Context is king when we’re talking about AI,” Irecki emphasizes. “You can definitely feed or train a model with generalized data from your business, but to get key insights that will really accelerate your business in a new direction, you need context from data.”
Context means entity resolution at scale. Irecki illustrated this challenge when he recounted how one AI model said he loved playing badminton while another claimed he played squash. Neither was correct, and both reflected fragmented data sources.
In M&A scenarios, fragmented data sources play out more seriously: the same customer might appear as “Customer ID 12345” in one CRM, “Employee #67890” in HR systems, and “[email protected]” in support platforms. Without master data management and entity resolution, AI agents trained on each system generate contradictory insights. One model might recommend premium service upgrades based on customer value; another might flag the same person as a credit risk based on employment data; a third might route their support tickets to junior staff based on ticket history. The AI works perfectly in each silo and produces chaos in aggregate.
Post-close value dilution and system brittleness
Post-acquisition, companies typically face multiple instances of core systems, including numerous finance platforms, disparate HR systems, and fragmented customer and supply chain data. AI models trained on these silos “will fail to generalize,” Irecki argues. “You need to get to a single source of truth. Data maturity isn’t just a technical milestone for these M&A plays. It’s a strategic enabler to scale AI enterprise-wide.”
Architectural choices matter. Traditional point-to-point integrations create brittle dependencies that shatter during M&A. When Company A’s finance team deploys an AI agent that directly queries Company B’s legacy ERP system post-merger, you’ve hardcoded assumptions about data location, schema, and access patterns. The moment IT begins system consolidation by migrating from the acquired ERP to the parent company’s cloud platform, those direct connections break. The AI agent fails silently, or worse, continues operating on stale data.
Here, the Model Context Protocol (MCP) is emerging as a standard for AI agent interoperability, promising to standardize how agents discover and invoke tools across systems. But “MCP is still an evolving standard, lacking security capabilities,” Irecki cautions. Without authentication, rate limiting, and audit logging built into the protocol, organizations must layer these capabilities themselves. Mature API infrastructure provides exactly this governance foundation.
Mandate for data due diligence: Pre-close conditions
Organizations that discover AI sprawl post-merger face extended remediation timelines. Technical teams must reverse-engineer undocumented systems, reconcile conflicting data models, and rebuild governance from scratch. It’s work that delays actual integration and diverts resources from growth initiatives.
Pre-deal AI audits offer a more efficient path: automated scanning for agents built in common platforms (like Salesforce Agent Force, Microsoft Copilot, and ServiceNow), interviews with business unit leaders, and review of cloud spend for undocumented AI services. Implementing baseline governance frameworks, such as inventory systems, access controls, and audit logging, prevents the post-merger scramble to understand what’s actually running in the merged organization.
CDOs and AI leaders helping to evaluate potential acquisitions in APAC need to audit data architecture, integration maturity, and AI governance before financial modeling and entering serious M&A discussions. Without this foundation, those impressive M&A deal values represent unquantified technical debt and integration risk that will sabotage post-merger value realization.
Image credit: iStockphoto/Iuliia Efimova
