Are you wanting to explore the infinite possibilities relating to implementing AI? The first step is, naturally, to develop a solid AI strategy. We will guide you through five things you should consider before implementing AI in your business.
This blog post was originally published on Visma’s Norwegian blog. You can read the original post here.
TL;DRBuilding a credible AI strategy requires addressing five foundational areas before any technology is deployed: the competency question of whether to build or buy data science expertise, data quality and capture discipline, GDPR and privacy compliance, integration with existing systems, and a clear business case for each use case. Originally published on Visma's Norwegian blog, this post treats AI adoption as an organizational capability project, not just a technology purchase.
Are you wanting to explore the infinite possibilities relating to implementing AI? The first step is, naturally, to develop a solid AI strategy. We will guide you through five things you should consider before implementing AI in your business:
Implementing artificial intelligence requires expertise. You need to evaluate if that is a competence you want to buy or recruit in your company. The job function, Data Scientist, has been voted one of the most sought-after job functions of the 21st century. It can be a challenge to assemble the right Data Scientist team as their competencies are in high demand. Artificial intelligence does not live a life of its own and, in other words, must be put into production before results can be seen. At the same time, it can be a challenge to integrate the solution with the internal systems.
If you want to implement AI, you need to have available data that your systems can learn from. A lot of the time the data quality is lower than anticipated and in some cases so bad that it is not possible to properly use the AI.
Data must be obtained in a way that allows the system to learn from it at a later stage. In many cases, business processes must be changed to better structure data. However, changing business processes when acquiring data can be a long process. It is therefore something that must be taken into account when considering whether the expertise should be internal or external.
Many AI solutions are designed to make decisions that are ultimately about people. The consequences of that are that the data used for learning must be in accordance with the GDPR and that the project is continuously assessed. There must be a clear legal basis to be able to use the information.
Risks can be reduced by, for example, anonymizing the data set. However, be aware that this process is also seen as data processing and should therefore be approved and impact analyzed. Security includes more than just privacy, so as few people as possible should have access to the data and the server.
Ethical AI is about AI making choices that we humans would have made ourselves. If you are not aware of these conditions, you may end up building models that inadvertently discriminate against people in an automated decision-making process. If, for example, you have conflicting conditions in the data sets, you may end up in a situation where your digital bank advisors reject all loan applications from immigrants, regardless of the ability to pay.
In addition to this, it is important to think about whether you want artificial intelligence to be explainable and verifiable. Will your system always make the same decision using the same input data?
Finally, it is important to consider exactly what the AI will be used for. Will AI replace some of the tasks for human work? If so, what other tasks should they take on instead? Perhaps AI should be used to increase profits instead of lowering costs and thus preserve the number of jobs.
Before it makes sense to implement an AI strategy, the problem must be in place. It is not necessarily enough to have sufficient data that you want to “do something cool with” to create value. Look inside the company and find the problem that needs to be solved. The basis for a successful AI implementation is a sensible business case.
Frequently Asked Questions
What are the five things every AI strategy should address?
The post identifies: competencies (whether to hire or buy AI expertise), data capture (ensuring you have clean, structured data for the AI to learn from), privacy and security (GDPR compliance and legal basis for using personal data), integration with existing internal systems, and a phased implementation plan. Skipping any of these creates implementation risk.
Do we need to hire a data scientist to implement AI in our company?
Not always, but you do need to decide whether AI competency will be internal or external before starting. Data Scientist is one of the most in-demand roles of the 21st century, which makes hiring challenging. Many companies use external providers or buy AI-embedded software instead of building a team from scratch.
What data quality issues typically block AI implementation?
Data quality is frequently lower than anticipated, and in some cases poor enough that AI simply cannot function properly. The root cause is often that business processes were designed without structured data capture in mind. Fixing this sometimes requires changing how data is collected, which is a process change project before it becomes an AI project.
How does GDPR affect AI projects in finance or HR?
Many AI systems in finance and HR make decisions about people, which requires a clear legal basis for using their data. The AI must comply with the GDPR, and the project must be continuously assessed as it evolves. Anonymizing data sets can reduce risk, but this process must be handled carefully as anonymization can also reduce the usefulness of the data.
The CFO’s role is particularly challenging during the year-end process, as it involves reviewing...
This blog post is written based on an interview with Carsten Lillelund Pedersen, Thinking as a...
