Build vs. buy in the age of AI: Why extensible platforms are the only way to avoid the “catch-up” trap

Introduction

Every technology leader has faced the build vs. buy software decision at some point in their career. The decision matrix looked something like this: 

• Buy when speed matters, the requirements are fairly standard or you lack the internal expertise needed. 
• Build when you want more control and differentiation is core to your business model.

For companies seeking to integrate AI, the build vs. buy debate feels familiar. We’ll cover why the build vs. buy framework has changed, and how companies can use extensibility to build a flexible infrastructure for the future.

Build vs. buy AI software: What’s better? 

According to a PwC survey, 90% of companies are actively implementing AI in their operations. Brands and retailers, in particular, are already leveraging AI for a wide range of use cases, from customer-facing journeys, such as personalized recommendations, to operational workflows, like inventory/demand forecasting. 

Very quickly, businesses that don’t integrate AI into these areas risk falling behind, as 97% of decision-makers agree that AI is critical to the future of commerce. But the barriers are steep: For consumer-facing companies, integration complexity is the biggest issue, cited by 59% of leaders. eCommerce companies need their AI tools to integrate with CRMs, supplier portals, marketing platforms and dynamic product catalogs with thousands or even millions of SKUs. 

Some leaders in the messy middle of the integration challenge are finding that their tech stack isn’t up to the challenge, and that the choices they made five years ago are failing them in this new era. 

Lessons learned from the homegrown eCommerce era

A decade ago, many CIOs faced a similar challenge with the advent of eCommerce: Either build a storefront on an all-in-one eCommerce platform and deal with its limitations, or build a custom eCommerce website internally.

While many retailers chose to buy, attracted by the ability to quickly create online storefronts, a large portion chose to build homegrown solutions because they wanted control, customization and ownership. And for a while, each of these approaches worked, one way or another. 

Then the market moved beyond desktop-based eCommerce. Mobile commerce exploded, and customer expectations for personalization, speed and seamless omnichannel experiences rose sharply. 

Early eCommerce adopters that chose to build online stores like Moonpig, Loomstate and BIC found themselves constrained by custom solutions built in-house, needing engineering support and lengthy development cycles for any meaningful store updates. 

In each case, the homegrown platform that once represented control and flexibility became the single biggest obstacle to growth. These companies found that their architecture couldn’t bend quickly enough, and they had to replatform from scratch — an expensive and resource-intensive endeavor. The problem is urgent: Around 30% of the eCommerce companies we’ve surveyed replatformed from a homegrown solution due to these challenges. And the vast majority of those who haven’t yet switched (88%) believe they need to do so within 12 months. 

Meanwhile, companies that relied on legacy eCommerce platforms weren’t able to easily scale across multiple channels, brands or regions. Adding new features could take months of preparation and implementation, while customer experiences deteriorated over time. Companies like L.L. Bean and Breville spent years running on outdated platforms that cost a fortune to maintain, let alone keep up to date. 

That being said, the lesson many enterprises took away from this wave of replatforming wasn’t that  “buy” is better than “build,” or vice versa. In practice, the rise of modular architectures demonstrated that the real breakthrough lies between those extremes: Combining differentiated in-house capabilities with flexible, interoperable third-party components. 

Those same lessons are becoming relevant again in the AI era. As enterprises race to integrate AI capabilities, the challenge isn’t about choosing between building everything internally or outsourcing innovation entirely, but designing architectures that can evolve continuously without becoming tomorrow’s bottleneck.

The options for enterprises in the AI era: Buy, build or blend in detail

Companies that want to integrate AI into their businesses face a landscape where technological evolution occurs in shorter cycles. A functional, modern eCommerce website launched in 2025 may not be able to accommodate must-have elements like agentic commerce in 2027. In the new framework, there are now three options for technology leaders: 

1. Buy

Buying AI capabilities means adopting commercially available tools, pre-built models or platform-native AI features. For most companies, this is the entry point. Individual teams or employees bring in an AI tool to test a use case, like a support chatbot or product recommendations. Many commercial AI solutions can reliably deliver commodity use cases at scale and at low cost. 

Pros: 

The advantages are immediate: No model training required, no infrastructure to support, no specialist hiring. Vendors handle the underlying model development, maintenance and improvement needed. The biggest benefit of the buy approach is speed, so companies can launch AI capabilities in weeks. 

Cons: 

The risk of a pure buy approach is vendor dependency and, more specifically, the extensibility ceiling. If the AI product you purchase is a black box that you can’t customize its behavior, connect it to your proprietary data or extend it to serve use cases the vendor hasn’t anticipated, then you've traded one form of rigidity for another. 

The bottom line: 

Buying AI capabilities helps companies test and launch quickly to compete. Buying works well when the use case is well-defined, but it will struggle to deliver value if it can’t be connected to your proprietary data or adapted to your unique needs over time. 

Ultimately, buying a simple out-of-the-box solution doesn’t create unique value for your business — the value comes from pairing your unique data and knowledge with LLMs and generative AI. The question to ask of any AI solution is not just “what does this do today?” but “How far can I extend it tomorrow, and what does that extension process look like?”

2. Build

Building your own AI capabilities means developing models, infrastructure and tools in-house. If the AI capability itself is a genuine competitive differentiator — a proprietary demand forecasting model trained on years of exclusive transaction data or a recommendation engine that encodes understanding of a specific customer segment — then building may be justified. 

Pros: 

Building your own AI capabilities means full control over the system’s behavior, no vendor dependency and the ability to train on proprietary data in ways a bought solution may not permit. Those who succeed will gain complete control over the tech, intellectual property and potential market share. They will also avoid huge token invoices that are stacking up for most AI adopters. 

Cons: 

Developing AI capabilities in-house is a significant investment. Building AI from scratch requires specialized talent: AI/ML engineers, data scientists, MLOps specialists and AI governance professionals. The infrastructure requires model training, fine-tuning, hosting, monitoring and continuous retraining, often taking months to the first deployment. Institutional knowledge often stays with the engineers who wrote the code, meaning it’s difficult to train new engineers. 

Most critically, what you build is frozen in time relative to where the broader AI market is heading. The open-source and commercial AI landscape is advancing at a pace that internal teams can’t keep up with. The model you train today may be outperformed by a commercially available model in six months, at a fraction of the ongoing cost. And the loss of agility can mean you miss out on revenue opportunities.

The bottom line: 

Most companies that attempt the build approach will be caught in the catch-up trap: What they invest six to twelve months building at great expense will be outdated in as little as a few months. In the eCommerce space, there is little to no competitive advantage in rebuilding AI capabilities like customer support agents or personalized journeys, as these capabilities are commercially available and will soon be table stakes for eCommerce.

3. Blend

In practice, the organizations that execute AI integration most effectively aren’t choosing between building and buying — they’re doing both.

This hybrid approach means buying the commodity foundation (the AI infrastructure, base models and out-of-the-box capabilities) while building or extending a differentiated layer on top. This differentiated layer may be custom business logic, proprietary data integrations or an experience layer that reflects how your specific business operates.

To take this approach, companies must have a composable infrastructure — a modular architecture capable of supporting new technology layers. In eCommerce, companies built on a homegrown platform or a too-limited eCommerce platform will find it difficult to retrofit AI tools to their systems. The integrations will be too complex and will either take too long to develop or deliver inaccurate information to customers. 

Importantly, this approach isn’t for every company — and that’s perfectly fine. Organizations with low differentiation needs and straightforward use cases will often do well with fully packaged solutions. 

The “blend” model tends to emerge in a different set of companies: 

• Teams without large engineering resources, but with a need to move quickly and stay competitive. 
• Fast-growing companies are already seeing AI reshape their shopping journeys and want to adopt capabilities without constantly tracking infrastructure changes themselves. 
• Complex, multi-brand or multi-region businesses that require deep customization but still need to ship fast. 

In these cases, the goal is to avoid rebuilding everything and instead focus internal effort only on the capabilities that truly drive differentiation.

Pros: 

Companies following the hybrid model get the best of both worlds: Speed to market and customization. An eCommerce company can launch today without the significant investment of building, refining and iterating their AI capabilities over time. 

Cons: 

Manufacturers need strong governance around integrations, product data quality and ownership across systems such as ERP, PIM, search and eCommerce platforms.

The bottom line: 

Buying a commodity foundation doesn’t mean giving up control. On an extensible platform, control is preserved exactly where it creates value: In the custom data models that reflect your product catalog, the business logic that encodes your service rules, and the AI-powered experiences that make your storefront unique. 

Why extensibility is the blend approach every CIO needs

At the core of the build vs. buy AI debate is extensibility, AKA the ability to extend the usability of your technology. Instead of building a platform or AI infrastructure that can never outperform what’s already on the market, consider buying a flexible, API-first infrastructure that will facilitate innovation and growth down the road. 

In eCommerce, this means building on a flexible architecture — a modular eCommerce infrastructure like commercetools that allows businesses to layer in the technology they need.

Extensible platforms are built to evolve. Rather than hard-coding functionality into a rigid core, they expose well-defined APIs, interfaces, hooks or extension points that allow both internal teams and external developers to safely add or adapt capabilities around the platform.

For instance, a business can integrate a new payment provider or connect an additional inventory system without altering the underlying order management logic already running in production. This separation of core functionality from extensions enables change without disruption, and innovation without rewriting what already works.

Taking the hybrid approach for speed and differentiation

At the end of the day, the build vs. buy debate comes down to one thing: Extending usability and adaptability over time.

The AI readiness challenges business leaders face today won’t be the same as those they face in five or even two years. A hybrid approach of buying AI software with planned extensibility means that companies are future-proof against short technology cycles. Instead of a custom build or rebuild every few years, you can focus on your unique differentiation using ready-to-deploy tools.

With a flexible, modular infrastructure and a buy-and-build mentality, you can match competitors at speed and cost while creating unique value. 

White Paper

Navigating the Agentic Commerce Era On Your Terms

The practical guide to help you approach agentic commerce strategically — not out of FOMO.

Download Now