Generative AI: commodity or core business?04 March 2024

With every new technological innovation comes the question of whether to buy a solution ‘off the shelf’ or develop it in-house. ERP, website, cloud, automation, IoT, Data Science, AI, etc. the question of “make or buy” is a fundamental part of strategy.

Generative AI is no exception. As in other cases, there is no single answer, but rather a range of answers depending on how critical the solution is to the company’s business.

For example, for applications that benefit from access to external data, the obvious solution is to use a public service. For example, for research, tools such as Perplexity or Consensus are very powerful, and (almost) nobody has any interest in developing their own solution. All you need to do is embed the practice by monitoring access, providing training and raising awareness of the risks.

For a service such as meeting summaries, document summaries or help with wording, public models are generally sufficient, but you will be looking for secure implementations to avoid any data leaks. Privatised” solutions such as Microsoft Copilot will generally meet your needs. For certain applications (technical summaries, for example, or sophisticated formulations), you may need to re-train a model, or give it access to databases of examples, which is considerably more complicated. Vertical solutions such as Nabla can offer something in between.

For applications that use your data and documents, you will need to turn to proprietary developments, using decision chains, APIs to your systems and a dedicated UX. The language model itself will generally be standard (proprietary or open source, another choice to be made…).

Finally, for core business applications, where you have a strong competitive advantage, or of course for applications that you wish to sell, the development will be particularly meticulous and secure, going as far as the in-house development of large models.

The answers can be very different from one company to another. So a company occasionally doing graphic design will be satisfied with a public version of Dall-E. An agency whose core business is this could go so far as to redevelop its own model.

At Eleven, we offer a structured approach to navigating this era of AI. Our methodology is based on the evaluation of the adequacy between the capabilities of the available AI models and the specificities of each profession. We help businesses identify inflection points where investment in custom AI models becomes not only viable but strategically advantageous, enabling true differentiation in a competitive market.

In conclusion, enterprise adoption of generative AI is not a binary choice between buying or developing in-house. Rather, it is a nuanced decision, influenced by the nature of the application, the criticality of the data, and the competitive advantage sought. In this context, a methodical approach adapted to each specific case is essential to maximize the benefits of generative AI while minimizing the risks.