I’ve been in three separate conversations in the past fortnight where enterprise IT leaders are agonizing over whether to build or buy AI capabilities. And in all three cases, they’re asking the wrong question.

It’s not “build or buy.” It’s “what are we actually trying to achieve, what’s our realistic capability level, and what can we sustain long-term?”

But that doesn’t fit on a slide deck as neatly, so we keep pretending the decision is binary.

Why Everyone Wants to Build

The appeal of building is obvious. You get exactly what you want. You own the IP. You’re not dependent on a vendor’s roadmap or pricing structure. You can iterate quickly based on your specific needs.

And there’s ego involved, let’s be honest. Building your own AI feels innovative. Buying a SaaS product feels pedestrian.

I’ve sat in meetings where leadership teams convinced themselves they needed to build because “our requirements are unique.” They’re usually not. What they actually mean is “we don’t want to change our processes to fit a product, so we’ll build something that fits our dysfunction.”

The Build Trap

Here’s what actually happens when enterprises decide to build AI capabilities in-house without the proper foundation.

First, they hire a couple of data scientists or ML engineers. Smart people, expensive people. Those people spend six months just getting access to the data they need, because your data governance is a mess and your data architecture predates the iPhone.

Then they build a prototype. It works. Everyone’s excited. You show it to the business units. They want it in production.

Now you need MLOps. You need monitoring. You need retraining pipelines. You need model versioning. You need to think about drift detection, A/B testing frameworks, rollback procedures. You need security reviews. You need compliance sign-off.

What started as “let’s build a recommendation engine” turns into “we’ve accidentally recreated Netflix’s infrastructure but worse.”

And here’s the kicker — while you’re building all this, your competitors who bought something off the shelf are already three iterations ahead of you.

When Building Makes Sense

I’m not saying never build. I’m saying build strategically.

Building makes sense when you’ve got genuine differentiation that’s core to your competitive advantage. If you’re a retailer and your personalization engine is what sets you apart in market, yeah, build that. If you’re a logistics company and route optimization is your moat, invest in proprietary AI.

Building makes sense when you’ve got the team to sustain it long-term. Not just build it, sustain it. That means data engineers, ML engineers, DevOps people who understand ML systems, product managers who can translate between technical and business, and ongoing budget for infrastructure.

If you don’t have that — and most enterprises don’t — you’re just creating technical debt and a maintenance nightmare.

The Buy Reality

Buying AI capabilities means you’re accepting someone else’s approach to the problem. That’s actually fine for most use cases.

You don’t build your own CRM. You don’t build your own accounting software. You bought Salesforce and SAP and you configured them to your needs. Why should AI be different?

The counterargument I hear is “AI is strategic, those are just operational systems.” Maybe. But even strategic capabilities can be built on bought platforms. Nobody builds their own cloud infrastructure anymore. You use AWS or Azure or GCP and you build on top of it.

Same logic applies to AI. Buy the platform, the tooling, the foundation models. Build the specific applications that matter to your business.

The Third Option: Partner

Here’s what I’m seeing work more often — enterprises partnering with specialized firms rather than strictly building or buying.

You bring in people who’ve done this before. They help you figure out what’s actually valuable to build, what you should buy, what your data readiness looks like, what your team needs to learn. They might build an initial version, but more importantly, they transfer knowledge so your team can sustain it.

It’s not cheap. But it’s cheaper than hiring a full in-house team and watching them thrash for eighteen months. And it’s more likely to result in something production-ready.

I’ve been working with AI consultants in Melbourne on a couple of projects recently, and the value isn’t just the technical work. It’s the pattern recognition. They’ve seen what works and what doesn’t across dozens of organizations. They can tell you what’s worth building and what’s just going to create problems.

What You Actually Need to Decide

Stop thinking build versus buy. Start thinking about these questions instead.

First: What’s our actual AI maturity? If you’re still struggling with basic data quality or your analytics team is underwater, you’re not ready to build AI systems. Full stop.

Second: What’s core to our competitive advantage? If it’s genuinely differentiated and strategic, consider building. If it’s table stakes or operational efficiency, buy or partner.

Third: What’s our capacity to sustain this? Not build it, sustain it. Can we maintain models, retrain them, monitor for drift, handle the operational overhead? If not, you’re building a liability.

Fourth: What’s the opportunity cost? While you’re spending twelve months building a chatbot, what aren’t you doing? What’s the value of getting something working in six weeks versus building the perfect solution in a year?

The Hybrid Approach

Most successful enterprise AI strategies I’ve seen are hybrid. They buy platforms and tooling. They build specific applications that matter. They partner for expertise and acceleration.

They don’t treat it as a binary decision. They treat it as a portfolio approach, where different capabilities get different strategies based on strategic value, maturity level, and organizational capacity.

And they’re honest about their limitations. If you don’t have world-class ML engineering talent and you’re not going to attract it to your industry, don’t pretend you’re going to out-build the specialists. Buy from them, partner with them, focus your energy on what you’re actually good at.

The Uncomfortable Truth

Most enterprises should be buying or partnering far more than they’re building. I know that’s not what leadership wants to hear. But it’s reality.

You’re probably not Google. You’re probably not going to build better foundation models than OpenAI or Anthropic. You’re probably not going to out-engineer Databricks on ML infrastructure.

What you can do is be really good at applying these capabilities to your specific domain. That’s valuable. That’s differentiating. And it doesn’t require building everything from scratch.

The best AI strategies I’ve seen are pragmatic. They use the best tools available, they build only what’s truly strategic, and they’re honest about capability gaps.

That doesn’t make for exciting innovation theatre. But it produces actual business outcomes.

Which is sort of the point.