Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.
AI. Buy or Build?AI. Buy or Build?
With the emergence of new AI platforms, businesses no longer need to make the outdated binary choice of build vs. buy.
3 Min Read
Image by Gerd Altmann from Pixabay
Sponsored by Lenovo
Custom or out of the box? It’s the question that every IT leader faces when looking at new GenAI use cases. Recent Lenovo research suggests that over 50% prefer to buy resources, services and capabilities rather than build them in house. However, a similar percentage (47%) find it “significantly challenging” to develop solutions tailored to their business.
Given this tension between the imperative to have something ready-made and the need to customize, it should come as welcome news that IT leaders no longer have to make a binary choice for GenAI solutions. The emergence of “AI platforms” in recent months means that foundation services can be delivered out of the box, with customizable capabilities on top that can be tailored to meet your requirements. No one needs to start from scratch any more.
The Best of Both Worlds
This marks a shift in the market. It offers the best of both worlds – more reliability and quicker deployment from a proven, tested AI platform, but with the ability to customize for your use case. It will also cut in half the time needed to get a GenAI use case up and running. There may still be instances that demand full custom solutions – organizations that need, say, large volumes of complex video processing may still need a “from scratch” AI system. But around 80% of businesses will be able to get the results they need by using an AI platform and modifying services on top of it.
AI Platform in Action
Lenovo Hybrid AI Advantage with NVIDIA, for example, offers an AI platform with a library of agentic assistants managed by a central control plane. These can be customized as needed, and range from sales assistants, product recommenders, and an enterprise knowledge base for product and services queries, to assistants for copywriting, employee onboarding and content inventory management. Because these assistants represent typical and consistent use cases, they can be deployed with customizations rather than having to be developed from the ground up.
Three Steps Before You Start
The platform approach has significant advantages. It’s quicker and easier than an in-house build and offers more custom capability than an off-the-shelf AI. However, like any AI solution, it should be implemented with care and thought. We recommend three considerations:
Make sure it will support your requirements. If they are highly specialized, you may be one of the 20% of businesses that needs a from-scratch solution.
Test extensively, and check for security. Ideally, start with internal solutions to build out your expertise and understanding, and to improve accuracy and reliability. Make tweaks until you’re in a place where the results are good enough for external exposure. Solutions need to be hardened – operational, secure, and reliable – before being moved into production.
Put governance and data cleansing in place. Monitor for model drift and bias and set up guardrails to make sure responses stay within your criteria. Guardrails might include restricting data sets. For example, in a test for one use case, I asked the AI who Mickey Mouse was – and it told me. It shouldn’t have known, because its responses should have been limited to a specific set of documents. Also, most data cleansing practices are designed for relational databases and need to be adapted for LLM datasets, which are largely document-based. My earlier post on AI pilots has more on these points.
Fast-Evolving Market
This platform-plus-customization approach to AI shows that the market is maturing fast. As to where it may evolve, we see out-of-the-box agent solutions that leverage multiple LLMs coming to the market before long. Currently all custom-only, these would have the capability to solve multi-step problems. Examples could be generating and compiling code, looking for errors and then correcting them; or generating an article on a specified topic, including finding suitable images.
It's an exciting future. And an exciting present, because the ability to leverage AI platforms is here, now.
To learn more about how Lenovo and NVIDIA technologies can accelerate your GenAI use case, contact [email protected].
About the Author
You May Also Like
More Insights
