The build vs buy dilemma: why most finance teams are hedging their bets on a hybrid model

Unless stated, all statistics referenced below come from "Building An AI-Ready Finance Function", a commissioned study conducted by Forrester Consulting on behalf of Airwallex, June 2026.

51% of finance teams are running hybrid AI models, and the split between what they build and what they buy follows a pattern that has nothing to do with capability.

In a recent Forrester study commissioned by Airwallex, a senior finance director at a UK tech company confessed: “We almost always buy rather than build. Our engineers will say, 'We can build that, it's straightforward,' but once we walk through requirements like auditability, SOC 1 reporting, and control frameworks, it becomes clear how complex it really is." 

That tension plays out in finance teams across the world right now. The AI capability is within reach. Your engineers probably can build the model, train it on your data, and get it producing useful outputs within a few months. The part that derails the project comes later, when your auditors start asking about explainability, your compliance team flags the control gaps, and your CFO realises nobody budgeted for the two-year regulatory wrapper the model needs before it touches a single financial report.

Forrester Consulting surveyed 1,279 finance decision-makers earlier this year. The headline finding: 51% already run hybrid models, combining in-house development with external vendors, with the percentage predicted to rise to 57% within the next 12 months. The binary build-or-buy framing no longer reflects reality, most finance teams have already picked both.

The 57% middle ground

Only 26% of organisations plan to build entirely in-house in the next 12 months. Only 17% plan to fully outsource. The overwhelming majority have landed in the middle (57%), and they've landed there on purpose.

This split is also directional. Over the next 12 months, finance leaders expect full outsourcing to drop from 26% to 17%. Finance teams want more ownership of their AI workflows, not less. They're pulling workflow design in-house while pushing compliance infrastructure out to vendors.

That trajectory tells you something about where confidence sits. Finance leaders trust their teams to own the logic of how AI fits their workflows. They don't trust themselves to maintain the regulatory scaffolding around it at the speed auditors and regulators demand. So they're keeping the steering wheel and outsourcing the engine maintenance.

43% of organisations expect AI capabilities in the next 12 months to come from vendor platforms that will require additional configuration or integration with their existing finance systems. Buy the foundation, build the differentiation on top. That's a deliberate architectural choice, and it's one that lets finance teams move faster without absorbing the compliance overhead that slows DIY builds to a crawl.

The compliance moat that kills DIY builds

Your engineering team can probably build the AI model in a quarter. The part they haven't priced in is the two years of regulatory wrapping that comes after. Finance operates under SOC 1, SOC 2, SOX, PCI DSS, and a patchwork of regional requirements that shift by market, entity, and transaction type. Any AI that touches financial reporting, reconciliation, or payments needs documented audit trails, explainability logic, and internal controls that satisfy external auditors before it processes a single transaction.

That's a multi-year, multi-million-dollar build on its own. And the costs don't flatten after launch, because regulators keep raising the bar, auditors keep tightening their standards, and your compliance team needs to maintain version control and documentation updates that nobody scoped in the original project plan. Most engineering teams quote the model. They forget to quote the maintenance contract on the regulatory wrapper around it.

Consolidation without connectivity

The data captures a telling detail here: 84% of organisations still require manual steps to complete finance workflows, even where they've consolidated onto fewer platforms. You'd expect fewer platforms to mean fewer gaps, but data stays fragmented because the platforms were never designed to talk to each other at the workflow level. Teams end up with three tools instead of ten, and the same manual handoffs in between.

The auditor bottleneck

Auditors are becoming a constraint in their own right. Traditional sample-based methods weren't built for AI-generated outputs at scale, so audit teams are defaulting to the most conservative approach available. More manual checks, more documentation requests, more friction. Organisations further ahead in AI execution are already pushing their auditors to move beyond manual validation toward real-time, technology-driven approaches. As one UK senior finance director noted in the Forrester study, there is growing expectation for auditors to plug directly into systems via APIs rather than relying on manual requests and sampling.

Where the line falls between build and buy (and what moves it)

The default split is starting to crystallise. Buy the regulated infrastructure, from payment rails and FX execution to card issuing, compliance frameworks, and audit trails. Build the proprietary logic, the custom treasury routing, bespoke risk models, and company-specific approval chains.

But the line isn't fixed. It moves depending on three things. How standardised your workflows are, how connected your data is, and how much regulatory exposure the function carries.

Function maturity determines the split

If your treasury team runs on clear parameters and binary outcomes, you've got fertile ground for custom AI development. A treasury analyst in Singapore routing multi-currency payments can build proprietary logic because the rules are well-defined and the data flows are predictable. Tax teams fall into the same category. Judgment-heavy functions like corporate development and investor relations stay closer to vendor-provided AI because they need tools that surface patterns across messy, unstructured data, not automation that follows a script.

Your AP team and your FP&A team operate at different levels of data maturity, workflow complexity, and regulatory exposure. The build-buy line should sit in a different place for each of them.

The orchestration layer is the real decision

The US CFO of a SaaS company captured the challenge in the Forrester interviews: individual AI tools work within their own walled gardens but don't communicate across systems. When you need human review for exceptions, context that spans multiple platforms, or logic that connects workflows across functions, you need an integration layer that sits above any single vendor's capabilities.

That's the crux of the architectural decision. You can buy five best-in-class point solutions and still end up with the same disconnected workflows you started with. The layer that connects those tools and lets your team build custom logic on top is where the real value sits.

The Forrester data backs this up. 66% of decision-makers prioritise orchestration capabilities across the broader finance stack when choosing vendors. 65% want flexible, platform-based architecture that supports modular adoption. The vendor that wins the deal connects your existing systems and gives your team a foundation to build on.

The talent bottleneck in the middle

The hybrid model sounds clean on a whiteboard. In practice, it creates a new kind of resourcing challenge that most finance teams haven't solved yet.

53% of Forrester's respondents cite limited experience operating AI-enabled finance processes as their top barrier to scaling. Only 15% have rolled out structured AI talent strategies across their finance function. Most are still running foundational AI literacy training, which falls short when your architecture demands people who can straddle the build-buy boundary with confidence.

The hybrid model needs a specific profile. Someone who understands both the AI and the governance, who can configure a vendor platform while designing proprietary workflows that fit your company's approval logic and reporting structure. That cross-functional skill set barely existed two years ago. It's now the scarcest resource in finance transformation.

Organisations further ahead in AI execution are tackling this by hiring data scientists and analytics specialists directly into the finance function, then blending them with traditional finance professionals. The Head of Commercial Finance at an Australian ecommerce company described the shift. Training that was once roughly 80% technical accounting and 20% softer skills has reversed, with a much heavier emphasis on AI usage, effective prompting, and critical thinking.

The organisations still routing AI requests through IT are learning that the latency costs them more than the headcount savings. When a treasury analyst has to file a ticket to adjust an AI workflow, business context evaporates somewhere between the JIRA board and the sprint backlog.

What the gap means for your infrastructure choices

The build vs buy question has already been answered by 1,279 finance leaders across the world. The answer is both. The open question is where you draw the line, and whether your financial infrastructure gives you the flexibility to draw it in the right place. When your payments, FX, accounts, and spend data sit on a connected platform, you remove the data fragmentation that blocks AI from scaling. You buy the compliance-grade infrastructure and start building the proprietary logic that gives your finance team its edge.

Source: Unless stated, all statistics referenced in this article come from "Building An AI-Ready Finance Function", a commissioned study conducted by Forrester Consulting on behalf of Airwallex, June 2026.