The benefits of artificial intelligence (AI) for companies are undisputed. However, companies need to be well prepared in order to use AI effectively. The most important thing is to create the right technological conditions. In this article we illustrate what is important and provide suitable examples.
Challenges in the introduction of AI
The technological landscape is characterised by a high degree of diversity. While some companies are introducing e-mail as a replacement for fax, others are implementing the umpteenth content management system (CMS) for better website organisation.
However, almost all companies are currently asking themselves whether and how they can best utilise artificial intelligence – regardless of their other tech stack.
In order to fully utilise the enormous potential of AI, companies need to lay certain foundations. A holistic approach is required that takes into account not only the technological and data-related infrastructure, but also cultural, legal and economic factors. Only then can AI realise its full potential and help companies remain competitive and develop innovative solutions.
The first step on the way to a successful AI introduction is to deal with the technological legacy. This applies to data as well as infrastructure and software.
Data as the basis for AI
Quality and timeliness of data
Data is the fuel of AI, because the respective language model is trained using the data. But not all that glitters is gold and not every collection of information is suitable for training a model. Existing databases must be checked for quality and cleaned up if necessary. Data quality is of crucial importance, as incorrect or inaccurate AI results can be caused by poor data (see: Bias in AI).
The saying “s**t in – s**t out” applies here, too.
However, the effort is worth it, as the example of Uber shows:
The mobility provider had the problem that inconsistent and inaccurate data led to poor predictions and unreliable services. Initially, Uber tried to improve data quality through manual checks and ad hoc solutions, but this did not bring the desired success.
The mobility provider therefore introduced a comprehensive data governance strategy that includes regular data checks and cleansing. The measures introduced led to a significant improvement in data quality and consistency. Uber services scored with a significant increase in reliability and accuracy. This led to a significant improvement in AI predictions and thus the overall quality of service.
Keywording for data processing
Another important quality criterion is the assignment of tags to data. By tagging data appropriately, AI can find and process relevant information more quickly. This leads to a significant increase in the efficiency and accuracy of AI applications.
This effect can also be illustrated by a large data processor – the business network LinkedIn:
At LinkedIn, difficulties in quickly finding and utilising relevant data led to a loss of efficiency. Data management was insufficient to cope with the growing volume of data. This slowed down the machine learning algorithms.
To solve this problem, LinkedIn implemented advanced metadata management tools with automatic tagging and categorisation functionality. This change resulted in faster and more accurate data processing, which significantly improved the efficiency of the AI applications and the user experience. The implementation of these advanced tools also led to a significant increase in work performance and service quality.
Consistency of the data
It is not only the quantity and sorting of data that is important, but also the consistency of the data. In order to enable reliable analyses and predictions of AI models, data must be uniform and consistent. It is therefore essential that data sources are regularly checked to ensure they are up-to-date and accurate.
Salesforce was faced with the challenge of integrating data from different sources and keeping it consistent. The variability of the input often led to inconsistencies. However, manual processes and individual customisations for data integration were inefficient.
To solve this problem, the company introduced a data integration platform that enables automatic consistency checking and correction. The solution significantly improved data consistency and enabled more reliable analyses and forecasts through AI. The use of this platform in conjunction with regular audits enabled Salesforce to ensure data quality on an ongoing basis, thereby optimising decision-making within the company. Processes could also be organised more efficiently.
Digitalisation as the technological backbone of AI
End-to-end digitalisation is essential to prepare the ground for AI applications. AI can also be used to move robots. In most cases, however, it is more efficient for a chatbot to answer questions from PDFs instead of having the robot fetch the right folders from the cupboard. If you want to use the new technology efficiently, the conditions have to be right and everything has to be digital.
In many places, companies are still struggling with outdated systems and so-called tech debt. The term refers to the burden caused by the use of outdated technologies. These are often costly to maintain and hinder progress. In order to make room for new, flexible systems, these technical debts must be eliminated.
However, the problem does not only affect traditional companies. The giants of the internet also have to constantly scrutinise their technological status and pay their debts.
Streaming giant Netflix was faced with the problem that its monolithic software architecture could not keep pace with the speed of its expansion and the demands of its users. The monolith led to long development cycles and difficulties in scaling. This affected, among other things, the AI-driven recommendation algorithms for which Netflix is known. While individual aspects had to grow and scale quickly, others were not ready for this. However, the optimisations and enhancements made did not lead to the desired success, as the complexity of the system continued to increase and made it difficult to maintain an overview and efficient maintenance.
The only solution was a complete IT modernisation with the aim of switching from a monolithic to a microservice-based architecture. Netflix decided to switch to a microservice-based architecture, where each function runs as a separate service and can be updated independently. The switch allowed the streaming company greater flexibility, faster development cycles and improved scalability. This led to a better user experience and faster innovation.
To determine the tech debt in an organisation, managers should ask a series of targeted questions that cover different aspects of the existing IT infrastructure, processes and technologies. There are no one-size-fits-all textbook solutions on how to eliminate each aspect. Companies are too different for that. Nevertheless, it is worth taking a look at the solutions that other companies have chosen:
Modernisation of outdated technological infrastructure
The technological infrastructure of many companies is dominated by outdated core systems and software solutions. Often, only a comprehensive migration to a new technology stack can provide a remedy. Banks are particularly affected by this problem, as they often operate systems in programming languages such as COBOL, which are hardly mastered these days.
The Royal Bank of Scotland (RBS) also faced this challenge: Its old IT systems were prone to failures and caused high maintenance costs. Patches and emergency measures did not offer a long-term solution and led to further cost increases.
RBS therefore decided to modernise its IT infrastructure, switch to cloud technologies and introduce a DevOps culture. According to the bank, these changes led to a more stable IT infrastructure, reduced downtime and maintenance costs and faster and more efficient development of new applications.
Such structures are also the backbone of AI implementation and utilisation. Integrating them into old environments for which they were never intended and keeping them running would involve a great deal of work, which is not foreseeable.
Modernisation of outdated software and application technologies
Individual software and applications can also be problematic if they are based on outdated technologies and are not adequately supported.
Due to the boom in music streaming, Spotify was forced to react quickly to market changes and introduce new functions. These included AI algorithms that suggest music to users, create playlists and find podcasts. The old technologies were not suitable for this. Before the changeover, Spotify used traditional development cycles and programming languages that were not designed for rapid changes, which led to repeated delays.
The switch to modern programming languages such as Python and Go and the introduction of a CI/CD pipeline enabled Spotify to develop and provide new features faster and more efficiently. This significantly increased the company’s competitiveness.
This also helps with the further introduction of modern AI systems. Most AI models and frameworks already have Python APIs and can be used quickly. If you don’t want to provide and maintain servers yourself, existing infrastructures are helpful. Many providers such as AWS or Azure already have special offers for the use of AI.
Improving integration and compatibility
Another common problem is the integration and compatibility of different systems, which often leads to data inconsistencies.
PayPal was faced with the challenge of integrating various internal and external systems, which resulted in inefficient processes. However, speed is an indispensable factor for a financial services provider on the Internet. Before the changeover, PayPal tried to optimise the integration through individual adjustments and bridging solutions. However, these approaches involved a great deal of effort and often led to errors.
By implementing an API-first strategy and using microservices, PayPal was able to simplify integration, increase flexibility and reduce data inconsistencies.
Such a strategy should be considered before the introduction of AI. Regardless of which form of machine learning is to be used, API-first integration helps to speed up integration.
Reduction in maintenance costs and downtime
Maintenance and support are also areas in which high costs and failures frequently occur.
Before automating its maintenance processes, Google had difficulties managing its huge infrastructure efficiently. This led to inefficiency and frequent outages.
By implementing comprehensive monitoring tools and automation solutions, Google was able to significantly increase system availability and reduce maintenance costs.
However, monitoring the infrastructure in this way cannot solve all problems. In particular, duplicate solutions that still exist lead to increased maintenance costs. One example of redundancies in Google’s infrastructure is YouTube’s duplication of all systems for interacting with comments, which are available for both the user and creator interfaces.
However, it is essential if you do not want to risk high downtimes and maintenance times when introducing systems as complex and performance-intensive as AI systems often are.
Minimisation of security and compliance risks
Risks are particularly high with outdated security systems and unclear compliance requirements.
A data leak at Equifax in 2017 showed that their previous security protocols were inadequate.
Following the incident, the company introduced modern security measures such as advanced encryption techniques, two-factor authentication and regular security audits. The measures implemented led to a significant improvement in the security situation and played a key role in restoring customer confidence. Since then, the problem has not recurred.
The use of chatbots can quickly lead to comparable data leaks. A team of researchers at Northwestern University has already impressively demonstrated that chatbots are only too happy to pass on the (secret) information available to them if they are given the right prompts. Regular audits and well thought-out information management are therefore essential if company secrets are to remain protected.
Conclusion: AI introduction needs a strong foundation
In order to successfully implement AI in a company, the foundations must first be laid. Comprehensive digitalisation, reducing technical debt, harmonising the IT infrastructure and building a solid, modern tech stack are just as important as ensuring data quality, tagging and consistency.
The fulfilment of these requirements and the application of a holistic approach (including change management and employee training in particular) are prerequisites for AI to realise its full potential. This enables companies to maintain their competitiveness and develop innovative solutions.
