Just as with any choice of technology, we often get the question from our clients whether to roll your own, or to buy off the shelf. This is just as true when implementing an AI solution as it is when building a webpage or customer portal.
Just as with any choice of technology, we often get the question from our clients whether to roll your own, or to buy off the shelf. This is just as true when implementing an AI solution as it is when building a webpage or customer portal.
The answer that any skilled consultant will give is of course "it depends". In this article we will discuss what it depends on, and how you can safely experiment with the choices you make.
Just as most consultancies we see the world in shades of gray. It is never black or white. It is never fully one or the other choice. It is more of a scale.
There are of course combinations of the choices above that will help you achieve the results you are looking for.
The further right on the scale you move, the more important it will be to have access to knowledge in data science, access to infrastructure and knowledge about how to run and deploy models. As the world of AI keeps changing at an ever-faster pace, you will also need to spend time to keep your custom solution up to date.
At the other end of the scale are the pure SaaS solutions. Give the service a credit card and you are off to the races. Many solutions have specific use cases, few possibilities for configuration and a price that grows with usage. But most SaaS solutions will evolve over time, add new features and build on new development from the AI community. You don't have to worry about scaling, updating infrastructure and spending sleepless nights thinking about your datacenters. But you do have to worry about AI Act how your data is being used, where it is sent and who has access to it.
Before you dive headfirst into AI models, SaaS services and snakeoil sales, consider a few things.
What problem do you want the AI to solve for you? Do you really need an AI to solve that problem? If you need to transform data from one format to another you could use an AI, but this is a problem software developers have solved with basic tools long ago, and no AI is necessary.
Whether it is in-house or through a trusted partner - what skills and knowledge can you utilize for this solution? Do you have people with knowledge about data science, data engineering, infrastructure and software development? Where are they specialized? The more custom solutions you are building, the more knowledge you need access to. If you need to train custom models you need skilled data scientists, ML-ops capabilities, data engineers and infrastructure to support the whole system.
What partnerships do you already have in place? Do you already have software running on one of the big public cloud offerings? Do you already use a SaaS solution that might be able to solve your use case fully, or perhaps part of it? Many of the services used are run on one of the big cloud offerings from Amazon, Microsoft or Google. Good partnerships can lead to scaling benefits and discounts as well as synergy effects.
How much can the solution to the problem be allowed to cost? If you build a custom solution the initial price will be high (whether it is internal development or using consultants) but the maintenance costs are likely to be lower. Any SaaS solution is easy to get started with but will most likely cost you more in the long run. Consider how costs will develop over time. Most AI solutions charge by token, meaning how much information you send into the AI, and how much you get as a reply, affects the cost.
Consider what type of data you have, and what is needed for your solution to be compliant with regulations like GDPR and the AI act. Do you deal with sensitive data? Will the solution support sensitive decision making? Does your data have to stay within the EU or is it ok to send it to servers in the US or China. These constraints will affect the range of choices you have when buying off-the-shelf products. It is worth noting that many AI services only have servers in the US, making it problematic to be compliant with EU regulations if your data contains personal information.
How many users will be on the platform? Are they internal or external? Do you need guaranteed uptime, or will outages not make a big difference for your business? The answers to these questions will give you a hint of how hard and expensive it might be to build your own infrastructure. It will also give you a hint of what is required of the products you want to buy, or services you want to build your application on. In general, the faster you want updates, the closer to real time you want your data to be refreshed, the more expensive the solution.
Choosing which AI system to buy or build involves many considerations. We offer a Data and AI Blueprint approach that covers all important aspects of choosing the right AI system, thereby reducing risk and increasing business value.
The approach will ensure you:
With this in place your business will be able to make an informed decision about whether your business should buy or build when applying AI to your business.
For more resources on strategy and how to make conscious buy over build decisions, spend some time learning Wardley mapping or reading the excellent book Good Strategy, Bad Strategy.
Call or write VP of Engagement, Esben Højmark Gimbel.
