In my discussions with executives, technical leaders, and product managers, one question consistently arises when it comes to adopting generative artificial intelligence within the company: should we develop our own solution ("make") or adopt an existing one on the market ("buy")? This "make or buy" dilemma has intensified with the rise of tools like ChatGPT, Midjourney, or DALL·E, which have transformed our ways of working, innovating, and interacting with our customers.

In reality, this choice is not as simple as it seems. Each option has its advantages and disadvantages, whether it's cost, time-to-market, level of customization, or the effort required to bring your teams into a new technological era. There is no universal answer, as each company is unique. The important thing is to have a clear framework and adapted strategies to make the best choice. At Atomly AI, I regularly assist companies in this reflection, to find the right balance between speed of execution, technological mastery, and user adoption.

In this article, I propose to dive deep into this dilemma. We will explore the key decision factors, the importance of involving teams, hybrid strategies that combine the best of both worlds, and the long-term challenges that this decision implies. My ambition is to offer you a clear, concrete, and pragmatic vision so that you can make an informed choice, supported by solid strategic thinking.

The "make or buy" dilemma in generative AI: a major strategic choice

The expression "make or buy" is a widely used concept in the business world. Applied to generative AI, it crystallizes around a fundamental choice:

Make: Develop your own generative AI solution in-house, relying on open-source models (like LLaMA, Falcon, etc.) or specialized frameworks, then train it on internal data.
Buy: Purchase an existing solution, provided by a major technology player (OpenAI, Microsoft, Google) or a specialized company, and integrate it directly into your ecosystem.

Why has this choice become critical today? Because generative AI is increasingly establishing itself as a strategic tool, capable of automating tasks, improving customer experience, providing personalized responses, and generating added value, whether in content, design, or even decision support. The challenge is to remain competitive in a context where the speed, flexibility, and relevance of technological solutions make the difference.

However, the challenges are numerous. Implementation costs can skyrocket, integration with existing tools is not always easy, and customizing the model to meet specific business needs requires time and expertise. At the same time, adoption by employees is not guaranteed: they may already be accustomed to ChatGPT or other public tools, making the transition to a potentially less intuitive internal solution more complex.

Key decision factors: costs, time-to-market, customization, and integration

To shed light on the dilemma, it is useful to look at some key decision factors.

Direct and indirect costs

In "buy," costs often materialize in the form of licenses, subscriptions, or costs related to the use of an API (for example, OpenAI's). This approach allows you to obtain an operational tool quickly, without massive investment in development. However, in the long term, this dependence on an external provider can lead to increased costs, especially if certain request volumes are exceeded.

In "make," initial costs are higher: infrastructure, recruitment of AI experts, tool development, maintenance. But ultimately, you gain autonomy and avoid continuously paying for an external service. Calculating the total cost of ownership (TCO) in the medium and long term is therefore a strategic element of the decision.

Time-to-market

Time-to-market is often a determining criterion. If your company is looking to quickly deploy a solution to meet an urgent demand (for example, improving customer relations during a period of strong growth), opting for a "buy" solution may make sense: you benefit from a turnkey tool, generally proven, and you can deploy it almost immediately.

Conversely, "make" is a marathon rather than a sprint. It takes more time, requires strong internal resources, and involves a longer development and testing cycle. But in the long run, this solution will be perfectly aligned with your business needs and internal processes, offering superior customization and scalability.

Customization

Customization is a major asset of the "make" strategy. By developing in-house, you have the opportunity to train the model on your data, integrate sectoral or business-specific features, and build a truly tailor-made solution. This is particularly relevant if your company operates in a sector with strong regulatory constraints, very specific terminologies, or unique processes.

However, more and more "buy" solution providers also offer customization options. For example, Microsoft Azure OpenAI or Google Cloud AI offer APIs and tools to adjust the model to your data. Certainly, the level of customization is often less deep than a fully internal solution, but it can suffice in most cases.

Scalability and integration

Generative AI does not live in isolation: it must integrate into an existing ecosystem (CRM, ERP, CMS, collaborative tools, etc.). With a "buy" solution, you often benefit from integrations already designed with Microsoft, Google, or other software suites. This facilitates implementation and limits technical friction.

In "make," you have full control over the architecture and interfaces, giving you total flexibility, but requiring more work from your technical teams. The advantage is that you are not limited by the features offered by a third-party provider.

Human and organizational challenges: beyond technology

Adopting generative AI is not just a matter of technology; it is also a human adventure. Users have their habits, preferences, and may be attached to pre-existing tools. For example, some employees already use ChatGPT or other AI-based applications and are accustomed to them. Imposing an internal solution that is less efficient or less familiar overnight may create resistance and hinder adoption.

Change management is a central element. It involves training teams, communicating the advantages of the new solution, involving key employees in the decision and development process. An AI, even ultra-efficient, will not bring all the expected value if the end users do not fully adopt it. User experience, ease of use, quality of results, and consistency with work methods are all factors that determine the final success.

Hybrid strategies: the best of both worlds

The good news is that the "make or buy" dilemma is not necessarily a binary choice. Many companies today opt for a hybrid approach, combining purchase and in-house development.

For example, you could buy a base model from a reputable provider, like OpenAI, then specialize it in-house with your data and expertise. This strategy allows you to have a robust and proven technological foundation while gaining in customization. It also offers the possibility of evolving gradually: you start with a "turnkey" solution, then gradually internalize certain parts of the tool, depending on your resources, skills, and ambitions.

Strategic partnerships are also a powerful lever. Collaborating with Microsoft, OpenAI, or other key players can provide technical support, dedicated training, regular updates, and a complete ecosystem of solutions and integrations. For example, I assisted a company in the tourism sector that started by integrating a purchased conversational model (on Azure OpenAI), then developed specialized modules in-house for marketing content generation, capitalizing on internal data on customer preferences.

This type of hybrid approach can be applied to many use cases:

Task automation: combine an external model for language understanding with an internal engine dedicated to business process management.
Content generation: rely on an existing platform for text production while integrating internal rules of style, tone, or regulatory compliance.
Industrial customization: import a pre-trained model in a generalist sector (for example, retail), then inject internal data specific to your environment, product catalog, or business language.

Thinking long-term: governance, skills, and sustainable competitive advantage

Beyond the immediate choice, it is important to have a long-term vision. Generative AI is a rapidly evolving field. Models, technologies, and best practices are constantly being renewed at an impressive speed. Therefore, this decision should be seen as a strategic investment, capable of generating a sustainable competitive advantage.

Governance: Establish solid governance, with processes for validation, updating, and model supervision. This is particularly true in regulated environments (banking, medical, legal sectors), where AI must operate in compliance with regulations.
Skills development strategy: Whether you choose to "buy" or "make," you will need to develop the necessary internal skills. This involves training teams, recruiting specialized profiles, and creating a company culture oriented towards innovation and experimentation.
Anticipating technological developments: Generative AI models are constantly improving. Anticipating developments, staying on the lookout for innovations, and being ready to adjust your approach are integral parts of a sustainable strategy. A solution purchased today may need to be replaced or adapted in a few months. An internally developed solution will need to be maintained and retrained regularly.

Conclusion: a decision matrix to guide your choice

The "make or buy" choice in generative AI is a complex dilemma, but not insurmountable. Each company has its priorities, constraints, and its own definition of value. The key is to rely on structured thinking, to test, learn, and move forward step by step.

Below, I propose a simplified decision matrix. It is not intended to be exhaustive but can serve as a starting point for your reflection:

Criteria	Advantage "Buy"	Advantage "Make"
Initial costs	Low (licenses/API)	High (infrastructure, R&D)
Long-term costs	Supplier dependence, recurring costs	Potential long-term savings
Time-to-market	Fast, immediate deployment	Slower, progressive development
Customization	Limited, depends on the supplier	Strong, aligned with your business needs
Integration	Simplified, plug-and-play	Custom, but more technical
Internal adoption	Facilitated by recognized tools	Requires training, change management
Scalability	Dependent on the supplier	Total, but requires internal skills

Practical recommendations

Start by evaluating your strategic objectives: What are the target uses of generative AI in your company? What are the priorities (cost reduction, speed, customization)?
Test existing solutions: Before launching complex development, test market tools. This will allow you to familiarize yourself with the technology, understand user expectations, and identify your internal differentiation points.
Do not hesitate to seek expert support: Whether from specialized AI firms (like Atomly AI) or partnerships with suppliers, expert support can help you avoid costly mistakes, accelerate your decision-making, and help you deploy a solution adapted to your needs.
Consider hybrid strategies: The "best of both worlds" exists. Start by buying to gain speed, then internalize certain parts to gain control, or vice versa. Flexibility is often the key to lasting success.

Ultimately, the "make or buy" in generative AI is not a simple opposition but an opportunity to build a tailor-made technological and organizational strategy. The goal is to maximize the impact of AI on your business, minimize risks, and, above all, ensure that your end users—your employees, your customers—fully adopt these tools. It is a path that requires reflection, experimentation, and a long-term vision. But it is also a great opportunity to strengthen the agility, innovation, and added value of your company. And at Atomly AI, this is exactly the kind of challenge I love to tackle alongside my clients.

Blog

Make or Buy: The Strategic Dilemma of Generative AI

Blog

Make or Buy: The Strategic Dilemma of Generative AI