Building vs. buying generative AI for marketing: What CMOs need to know

Key Points:

• CMOs implementing generative AI in creative workflows face a strategic “build vs. buy” decision: adopt off-the-shelf tools (e.g., Firefly) or build a middleware layer that integrates AI into existing processes.
• Buying delivers faster time-to-value and quick ROI, but can limit customization and become a long-term constraint as your workflow and needs evolve.
• Building (not training your own model) creates a custom orchestration layer that increases AI fluency, fits your workflow, and keeps underlying model providers swappable to reduce vendor lock-in.
• A “build” approach better prevents off-brand output by embedding governance, brand rules, and approval logic directly into the creative process.
• As the real economics of AI shift (token usage, pricing, and rising complexity), enterprises need flexibility to protect ROI and continuously choose the best tools/models.

At the 2025 ANA Masters of Marketing almost every CMO introduced their approach to embedding AI within their creative workflow and the impact it was having on personalization and productivity. We saw two distinct go-to-market approaches: either building a middleware layer to intermediate AI tools with the workflow, or using off-the-shelf solutions, e.g. Adobe’s Firefly, directly. When it comes to implementing AI in creative workflows, all CMOs are faced with the choice of build versus buy. Our belief is that build offers the greatest long-term opportunity for most Fortune 500 enterprises. It not only avoids locking the organization into any single vendor, but it also helps your own team become more AI savvy and protects your workflow as the true economic costs of AI move from investors to customers, inevitably impacting which models to use and where they will deliver a return on the cost to deploy.

What we mean by build vs. buy:

That AI offers an opportunity to drive increased productivity isn’t in doubt. For instance, in applying AI to creative personalization, a South American grocer was able to generate 100 alternative email introductions informed by consumers’ previous purchase behavior resulted in a meaningful uplift in promotion response. What it previously took a marketer to write two to three alternative introductions can be replaced by that marketer reviewing 30-50 times as many AI-generated alternatives. Similarly for imagery or video, AI can easily adapt content to better reflect the variety of customer contexts from showing a Chevy Silverado within an urban, coastal context to the big-sky landscapes of Montana. Customers relate to brands that reflect their values and ambitions. AI enables companies to deliver that at scale.

The larger question is how that adaptation at scale is realized. Most marketing organizations are stretched delivering the day-job – after all we’re only as good as yesterday’s acquisition numbers. There often isn’t the time or investment to experiment with new technologies. That’s why ‘buy’ is so seductive. Propositions such as Adobe Firefly’s offer a turnkey solution that can be trialed for free while providing enterprise scale with a custom solution and education of your team on how to get the most from the technology. At the ANA, Melanie Huet from Newell Brands, described the positive impact they were seeing from embedding bought-in Generative AI within their creative workflow. Turnkey solutions offer an almost immediate return on investment but, by their nature of needing to serve everyone, have limitations on how far they can be adapted to your unique workflow.

The alternative is to build a solution that’s tailored to your specific context. What we mean by this is to develop an AI UX for your creatives that’s designed to amplify every aspect of your creative process. Norm de Greve, formerly CMO and now Chief Growth Officer of GM exemplifies this approach with the work they’ve done to develop Metropolis, GM’s approach to embedding AI within their workflow. This isn’t about creating your own AI – the underlying models remain those from OpenAI and the like. What’s different is the middleware layer that manages how those models are applied. Developing your own middleware layer takes time, investment and developing a level of AI sophistication to identify where the greatest impact can be realized and what is required to make that happen. A strong partnership with your CTO/CIO will be a necessity.

At this point it’s no surprise that many are taking the ‘buy’ route. It’s easy to initiate, easy to scale and easy to show early impact. The business case is a straightforward one to make. Build offers a less certain path and greater upfront investment. Why then is it the path we recommend the majority to take?

While buy offers the easier path today, it has a higher risk of being a strategic dead-end. Here’s four reasons why:

AI isn’t just another piece of software like Excel. It’s more akin to the building blocks of DNA. How these AI building blocks are put together will be a critical source of competitive advantage. Relying on others to provide this expertise places any company at a disadvantage to those that develop the internal capabilities to applying AI to their specific market context. No one-size-fits-all solution can be sufficiently agile or adaptive to compete.

The evolution of AI is frenetic. It remains unclear if today’s leaders will be leaders of tomorrow. A report from Menlo Ventures already shows Anthropic pulling ahead of OpenAI for enterprise LLM API usage, a domain where OpenAI as recently as 2023 had a 50% share. By making a buy decision today, you risk locking into an expensive dead-end. The cultural, process and technology costs of switching AI providers will be high. In contrast, a build solution will be a simpler matter of unplugging APIs from one provider to be replaced by another. Yes, there will be some backend ‘re-plumbing’ but your build team will have that expertise. The benefit of middleware is that your creative team can maintain their workflow with much less disruption. It also offers the opportunity for a best-of-breed portfolio of AI solutions across the variety of applications that you have.

Every AI tool you use will start with biases that won’t fully align with your brand values and design principles. Whether build or buy, there’s work to be done to imprint your brand DNA on resulting output, for example some image systems carry a stylistic fingerprint that’s tough to adapt. Build provides a mechanism to better govern, embed tone, apply visual constraints and drive approval logic. Build also enables you to develop more sophisticated automation around the AI to better contour to the unique needs of your workflow. A key consideration here is to ensure that you have a voice in the room versus what we often see happen is the choice of AI tools lying with the CTO without the CMO having a voice in the decision, optimizing for capital cost & security without creativity or operating cost being considered.

Every AI OEM is losing money, it’s the investors funding the landgrab and subsidizing client access. That’s going to change within the next 24 months. While, the cost per token is falling dramatically, something like a 75-90% drop in just the past two years, the cost to train new models continues to rise and the number of tokens required to address more sophisticated efforts such as agentic reasoning and video creation is rising faster than the drop in individual token cost. When investors lose patience and require a return, expect costs to utilize AI to rise substantially. Brands will need to rapidly identify where in their creative workflow AI continues to deliver a ROI at a higher cost and be prepared to switch to more efficient models as the market adapts to the real economics. Your ability to do this under a buy model will be slower and more limited – be prepared for some difficult conversations with your CFO.

The decision to build can be an intimidating one, but it needn’t be. AI tools such as Claude Code have made development of middleware solutions faster and more accessible. We are increasingly helping companies integrate AI into their existing workflow in a matter of months while also building the skills among existing teams to manage ongoing support and development. As you make your decision between build and buy, be thoughtful about the lifetime strategic and financial implications of that decision. While buy provides an easy route to immediate impact and is likely the only option for smaller companies, for larger enterprises we believe that for the majority build offers the stronger path.

Use the statements below to assess which side of the buy vs. build debate your company should pursue: