Brecht Coghe
The AI landscape is changing fast. Foundation models have taken a huge leap forward, leading to more cost-effective AI solutions while unlocking increasingly more value in practically every business process. This has created a new reality in the AI landscape, with sometimes more questions than answers.
One that we regularly hear from those responsible for AI is: “Do we invest in custom AI solutions, or do we wait for an AI-as-a-service (AIaaS) ?”. Simply put: Should we buy or build? A great question! We’ve seen a number of companies navigating this dilemma, and I’d like to share some of their learnings.
Nokia exploiting the potential of custom AI
An interesting example is Nokia’s approach to the buy vs build dilemma. They’ve invested in their own Large Language Model, tailor-made for the telecom industry. The model has been specifically trained to support their customers. When deploying it, they reached a customer satisfaction of more than 95% while making the support process more efficient. Quite impressive! Now, looking from the perspective of the buy vs build choice, Nokia realized that it is impossible to get ahead of competitors by only introducing AIaaS in their organization. By creating their own custom AI solution, they have leveraged their knowledge and data gathered over the past decennial in the telecom industry. They’ve used it to develop a unique service for their customers and to create an advantage over their competitors.
AI-as-a-Service: A Convenient Yet Limited Solution
A fast way to value with AI
With the recent advancements in AI, a growing number of AI solutions will be productized and provided as AIaaS solutions. A great step forward. The question of whether to use those services becomes even more important. AIaaS has the advantage of being cheaper, being hosted and requiring less effort for integration. The latter is often a key aspect of the service itself. However, even though that is great, AIaaS won’t help you win.
One-size-fits-all
Almost by definition, an AIaaS solution is a product made to fit all uses within an AI use case, which leads to some considerations. First, as always, the business case should be priority number one. Based on how an organization works and what they want to achieve, requirements for an AI solution, independent of custom vs AIaaS, can be slightly or even wildly different. An AIaaS does not always provide that flexibility. As an example, let's consider GDPR. An AIaaS solution that does not provide a data processing location inside of Europe will very often not be usable. So even though at first it looks like the perfect solution for what needs to be done, it might turn out not to have that option or feature that is really needed.
➡️ Take-home message one: Before even looking into AI-as-a-Service solutions, list your functional and non-functional requirements and validate if the solution you are looking into provides what you need.
A second implication of the one product that should fit all is that the feature roadmap of the AIaaS is shared among all its clients, and thus, it is hard for an organization to get the exact features that are needed prioritized or even to get them done at all. However, naturally, both organizations and use cases will evolve over time due to a changing market, which might create a mismatch in the longer term. This is an expectation that often also needs to be set from the start when choosing AIaaS to avoid discussions with business to explain why 'the AI' can not just give the solution to this new nuance in the use case.
Choose AIaaS with care
To summarize, given that an AIaaS solution meets the requirement needed and can be integrated for a use case that is not expected to change over time. The advantage of it being cheap is for the taking, and it can start producing value for your organization quickly.
Custom AI Solutions: Differentiating Your Organization
Now, are custom AI solutions only useful when the above limitations of an AIaaS solution are a problem? No, not at all. As mentioned earlier, AIaaS is a product that fits all uses. If everyone has the same AI, no one is really getting ahead. The power of custom AI solutions is that they can differentiate your organization from its competitors. So, to be able to answer the buy vs build dilemma, the real questions to pose are: ”How do you want to make the difference?”, “What can make you stand out?” and “How do you want to stand out in the future?”. Once these questions have been answered, an AI roadmap can be drafted with custom AI solutions that align with the company's strategy to create a competitive edge. At our clients, we have experienced different ways of doing that:
1. Standing out by leveraging unique data
Brussels Airport Company, for example, has a unique cache of data on airport and general logistics. By investing in custom solutions to predict passenger flows or to predict airport materials needed, they made use of that data to create a strategic, competitive edge and to become an inspiration for many other companies in their industry. Data is the new gold, and that is exactly what is at play here. Organizations might have more and better-suited data than their competitors already, and AI is the perfect way to redeem that advantage. Having a custom AI solution trained on that unique data will automatically differentiate you from competitors and will create the competitive edge you are looking for.
2. Standing out by including exclusive insights from your experts
Competitive edges are not only present in the data. For one of our clients in the engineering space, we created a regression model to simulate wear over time of a valve based on sensor values of the machine. The first baseline results were mediocre at best. However, when diving deeper with the expert engineers, they introduced their intuition to the challenge. This unique expertise can be translated in terms of feature engineering and pre- and postprocessing of data, and in the end, impressive results were obtained. The learning here is that many organizations differentiate by having the best people in their field or by having gathered unique expertise. Introducing this in a custom hybrid AI set-up results in higher-quality services or products or in more efficient processes that, in turn, have an impact on the bottom line.
3. Building your unique selling point
Even when that unique set of data is not available yet, or when it is challenging to translate people's knowledge into the AI application, building a competitive edge is possible. Think of Tesla, who decided to invest in self-driving cars. Their strategy is first to create a large fleet of cars. Then, iteratively, they gather data and improve their custom AI solution to grow their advantage in the industry. Similar to Tesla’s data flywheel, organizations can invest in getting access to a unique set of data that is in line with their company strategy, f.e. by covering a specific niche that no other competitors offer. Once this access has been created, a custom AI solution is one of the main tools that will lead to that return on investment.
➡️ Take-home message two: When looking into AI, determine which competitive edges you want to build and then explore how custom AI solutions can contribute.
Organizations can out-scale AI-as-a-service
Now, the power of custom solutions to differentiate and achieve strategic goals is clear. However, there is one more reason to consider custom solutions. Investing in only custom AI solutions for smaller companies is often not feasible, and so choices need to be made based on their strategy. However, for big companies with huge volumes of AI usage, the pricing model of AI-as-a-Service (=Price per usage) can come under pressure. At some point, creating their own custom solutions might become cheaper and as added benefit they build their AI knowledge internally, and they can customize it towards their own requirements. Together with a global healthcare firm, we developed the “Meeting Reporter”, a tool that records meetings and generates automated reports and action lists. Even though a similar tool will soon be available in Teams, the solution already paid off significantly over the past couple of years, thanks to their scale and resources.
Buy vs Build: An Inevitable Decision in an AI-Dominated Future
The AI world is changing fast, and some companies are waiting for the AIaaS they need to pop up. But.. AI is 'the' technology of the future, and organizations that do not have it yet, will be falling behind soon. Having the solution, whether it is AIaaS or custom, is only step one. AI needs to be embedded everywhere in organizations, and the employees need to shift their mindsets. The sooner organizations get started with using AI, the earlier AI maturity will be reached and the earlier the edge over the competition will kick in.
➡️ Take-home message three: Start including AI in your strategy and operations as soon as possible to ensure your company is ready for the coming evolutions because I am 100% certain that we are in for quite some AI ride!
Conclusion
There is more to the question of “Buy vs build” than just: “Does an AI-as-a-Service exist?”. Keep in mind that the requirements for an AI use case should be captured first, and that it is not always straightforward to fit an AIaaS exactly the way that is needed. Moreover, to let organizations thrive, they need to investigate how custom AI solutions can enable them to get the strategic, competitive advantages that they are aiming for and how they can build the edge that is needed over their competitors.
How do you approach this challenge? How do you collect use cases and prioritize your AI roadmap? We are collaborating with a number of our clients to share these insights. Interested in the report or to share your views? Reach out to us or a colleague at Superlinear!
Contact Us