Build vs. Buy: How to Give AI to Everyone Without Paying for Everything

Most organizations aren’t losing the AI race because they chose the wrong tool. They’re losing it because they asked the wrong question. “Should we build or buy?” sounds like a strategy—but it’s actually a trap.

The real question is: what problem are you trying to solve, and what does your organization need to own versus what can someone else build better and faster than you ever could? Get that answer right, and the build-vs.-buy decision becomes much clearer. Get it wrong, and you end up with eight disconnected AI tools nobody uses, custom solutions that nobody can maintain, or a vendor lock-in that’s quietly reshaping your workflows around someone else’s roadmap.

At Innovation Summit 2026 in Pittsburgh, three members of HIKE2’s own team took the stage to unpack exactly this challenge—sharing the frameworks, failure modes, and practical principles they use every day with clients across financial services, law, insurance, the public sector, and high-tech. The conversation was candid, specific, and immediately applicable. Here are the key takeaways.

About the Session

The session was led by Amy Gradnik, Principal of Strategic Services at HIKE2, and featured Christina Natale, Director of Legal Strategy & Advisory Solutions, and Ken Kwan, Enterprise Architect.

Between them, they bring experience spanning litigation practice at the Department of Justice, private equity and finance, and enterprise data architecture across banking, healthcare, advertising, and retail. The conversation drew on real client engagements and the hard-won lessons that come from helping organizations move from AI ambition to AI execution.

Escape the Binary Trap: Build vs. Buy Is the Wrong Frame

The session opened with a provocation: the entire build-vs.-buy debate is built on a false premise. It implies that every AI decision is a binary choice between two endpoints—build everything custom, or buy an off-the-shelf solution—when in practice, almost every successful AI strategy is a deliberate hybrid of both.

Ken Kwan offered the clearest articulation of where to draw that line: you buy the commodity, but you build your competitive advantage. The intelligence layer—the large language models, the AI engines, the reasoning capabilities—is increasingly commoditized. Companies like Anthropic, OpenAI, and Google are investing billions to make that layer better. An individual organization isn’t going to out-engineer them. But the data that feeds those engines, the workflows that govern how outputs are used, the organizational knowledge that gives the AI context—that’s yours. That’s where your differentiation lives, and that’s what you need to own.

Christina Natale added the operational dimension: organizations that fall into the binary trap often do so because they never started with the actual problem. A vendor demo is exciting. A new AI feature from an existing SaaS provider is easy to say yes to. But those decisions, stacked on top of each other without a governing strategy, produce what the session called “AI sprawl”—a landscape of disconnected tools that don’t talk to each other, create redundant costs, and leave employees more confused than empowered.

“You buy the commodity, but you build your competitive advantage. The intelligence will keep getting better—but your data, your workflows, your organizational knowledge? That’s the part you own.” — Ken Kwan, Enterprise Architect, HIKE2

The practical implication: before any build-vs.-buy decision, organizations need a clear picture of their current technology ecosystem, their strategic priorities, and where AI can genuinely move the needle—versus where it’s just adding complexity. Amy Gradnik framed it as a question of organizational vision: how does your AI strategy support your broader business strategy? Without that alignment, every individual tool decision becomes a guess.

Your Data Is the Differentiator—But Only If It’s Governed

One of the session’s most grounded and practically urgent insights was about data. Not data strategy in the abstract, but the unglamorous, foundational work of knowing what data you have, where it lives, whether it’s accurate, and whether the AI you’re deploying can actually access and trust it.

Ken described a project where a retrieval-augmented generation (RAG) system built on an internal knowledge base consistently returned wrong answers. The AI wasn’t broken. The data was—outdated, contradictory, and ungoverned. The lesson landed hard: “garbage in, garbage out” is not a new concept, but it takes on new urgency when the system consuming your data is an AI that will confidently present its errors as facts.

Christina connected this to a pattern she sees consistently with clients, particularly in document-intensive industries like law and financial services: even organizations with sophisticated document management systems often have governance processes that are fundamentally manual. A human has to save the document. A human has to version it correctly. A human has to put it in the right place. When that doesn’t happen—and it often doesn’t, at scale—the AI that was supposed to unlock the value of that institutional knowledge is operating on an incomplete, unreliable picture.

The implication for AI strategy is direct: data readiness is not a prerequisite that you check off before the real work begins. It is the real work—and it’s ongoing. As Ken put it, data governance is now AI governance. Organizations that invest in clean, well-structured, well-documented data assets will get dramatically more value from every AI tool they deploy, whether they built it or bought it.

“This is the system that is connected to our very precious, very well-governed data. That’s where you really want to focus that cost—on the uses that are going to rely on your data, your sensitive client data, the data your business has spent all this time and money to build. That is the data that you want to be using enterprise-grade AI for.” — Christina Natale, Director of Legal Strategy & Advisory Solutions, HIKE2

There’s also a cost dimension that often goes unexamined. Enterprise AI tools charge for usage—tokens consumed, queries processed, compute utilized. Using expensive, secure, enterprise-grade AI for questions that could be answered by a public tool is waste. Using a public tool for queries that touch sensitive client data is risk. Christina’s framing was precise: reserve your enterprise AI investment for the use cases that depend on your proprietary data. That’s where the cost is justified and where the returns are real.

Four Layers, One Framework: How to Architect AI That Actually Scales

The session’s most structural contribution was a four-layer enterprise architecture framework for thinking through AI deployment—one that HIKE2 uses with clients across industries to move from “we want to use AI” to “here is how AI integrates with what we already have and what we still need to build or buy.”

The four layers, in brief:

• The data and content foundation (is your data clean, governed, and accessible?);
• The AI engine and model layer (which reasoning capabilities are you using and how are they selected?);
• The orchestration and integration layer (how does AI connect with your existing systems and workflows?); and
• The user interface and delivery layer (how do users actually interact with AI outputs in the course of their normal work?).

What makes this framework valuable isn’t just its comprehensiveness—it’s its diagnostic function. Rather than approaching every new AI initiative as a greenfield problem, organizations can map their current state against each layer and ask: where do we already have what we need? Where do we have gaps? And for each gap, is the right answer to build, buy, or fix what we already have?

The user interface layer deserves particular attention, because it’s where AI adoption most often breaks down in practice. If using an AI capability requires an employee to navigate between multiple systems, log into a separate tool, or change how they work in ways that feel like added friction rather than reduced friction, adoption will stall regardless of how powerful the underlying technology is. The goal is to surface AI outputs in the context where work is already happening—not to create a new destination that people have to find their way to.

For organizations in complex, regulated industries—law firms managing matters in case management systems, insurers working inside claims platforms, financial services teams operating within compliance workflows—this principle is especially important. The AI layer needs to integrate with the operational layer, not sit beside it.

Design the Future Workforce Before the Technology Decides It for You

The session closed with a question from the audience—an HR professional asking how workforce and talent leaders should be thinking about AI’s impact on roles and skills. The response was one of the most candid and practically useful exchanges of the entire session.

Amy was direct: the goal is not to have your entire organization building bots. That’s not where most people create their highest value, and pursuing it as a strategy will exhaust the organization without delivering commensurate returns. The goal is to ensure that everyone understands how to use the tools, knows where to go for help, and has a clear picture of how their role is evolving—not disappearing, but changing.

“What are you designing your workforce to be for the future? What are the skills that you have today? Where do you want to invest in skills? And what do you think the technology should be doing for you that maybe you have people doing today?” — Amy Gradnik, Principal, Strategic Services, HIKE2

Christina took the point deeper, returning to the process mapping work that underpins the four-layer framework. The key question for workforce design is the same as the key question for technology design: which things that humans are doing today do we want to replace with technology, and which things genuinely require human judgment, relationship, and accountability? Mapping the current state of work—task by task, system by system, person by person—is the only way to answer that question with confidence rather than assumption.

This matters for reasons beyond efficiency. In industries with apprenticeship models built into their talent pipelines—law firms developing associates, financial services firms training analysts, insurers building underwriting expertise—there’s a real risk that automating entry-level work eliminates the learning path that produces senior talent. Organizations that don’t design for this proactively will find themselves five years from now with AI handling the tasks that used to develop their next generation of experts, and no clear plan for how that expertise gets built.

The session’s final takeaway, articulated by all three speakers in different ways: the organizations that will win with AI are not the ones that deploy it fastest. They’re the ones that deploy it most deliberately—with a clear strategy, a governed data foundation, an architecture that reflects how work actually happens, and a workforce that’s been brought along rather than left behind.

Watch the Full Session

The full Build vs. Buy session goes deeper on vendor lock-in and portability risk, the economics of token-based vs. seat-based AI pricing, how to evaluate AI tools across each layer of the enterprise stack, and real examples from HIKE2 client engagements across legal, financial services, and the public sector.

Not Sure Where to Start? That’s Exactly What We Help With.

Whether you’re staring down an AI strategy that needs to be built from scratch, an existing tool landscape that’s grown too fragmented to manage, or a workforce transformation that needs a human-centered design approach—HIKE2’s advisory team has been in that room with clients across every major industry.

We help organizations ask the right questions first, architect for scale second, and deploy with confidence third. If you’re ready to move from AI curiosity to AI capability, let’s talk.

Contact HIKE2 to start the conversation →