Is your business ready for AI?
Across all sectors, organisations are looking to leverage the use of artificial intelligence (AI) to improve efficiency and gain deeper insights. This trend has sparked numerous discussions about how to achieve AI readiness. The first step in laying the groundwork for robust AI use is data cleaning.
Data cleaning is the process of detecting and rectifying errors and inconsistencies within data sets to enhance their quality. It is critical that data used within AI has been evaluated to ensure that it’s fit for purpose.
Accurate data underpins the reliability of AI models, allowing them to make precise predictions and informed decisions. Moreover, clean data contributes to the efficiency of AI algorithms, enabling them to learn more swiftly and perform optimally, thus conserving time and computational resources. The reliability of AI outcomes also depends on the consistency of the underlying data, which is critical for building trust in AI systems.’
Data should be evaluated to ensure that it is:
-
Secure: Data must be protected from unauthorised access – this could be to prevent sensitive data from being presented to the users of AI, or from external breaches and cyberattacks. Implementing robust security measures is essential to maintain the integrity and confidentiality of data.
-
Data Quality: High-quality data is fundamental to producing accurate and consistent AI results. Data should be of high quality, having had missing values, errors, or outliers resolved before use.
-
Ethical: Ethical and privacy considerations should be considered for the data utilised. Any data used to train AI should be representative and free from bias.
An important thing to remember with AI is that it assumes that it is permitted to look at any data available in order to ‘learn’. Unless we have classified and protected any data that we don’t want to use (personal letters, end of year self-assessment forms etc) the AI will use it. Therefore, thinking about what data you don’t want to use can be just as important as thinking about the data you do wish to use.
Risks of poor data quality used to train AI
Bias
Large language models (LLMs) can require massive amounts of data to train and the most cost-effective way to do this is by scraping the internet. However, when we accept large amounts of web text as ‘representative’, we risk perpetuating dominant viewpoints, and inevitably perpetuating biases.
Data is primarily created by humans so carries inherent biases. It is therefore important to evaluate data for both accuracy and discriminatory biases that may affect the AI's perspective. Conduct parity tests during training to identify and address any biases.
Inaccurate representation
Representation errors are often a result of subjective training data. Additionally, accurate data labelling is crucial to avoid measurement errors. Without quality control, human-labelled data can introduce bias.
Outdated data
AI programmes may struggle with data quality aspects like timeliness and consistency. If trained on historical data, they can't account for changes in the 2020s, resulting in outputs that lack up-to-date and complete information.
Duplication errors
Using unchecked and duplicate data from multiple sources can cause errors. Moreover, unstructured data without metadata can create confusion, complicating analysis for the AI programme.
Our FDM Consultants are actively helping clients to become AI-ready, as well as supporting the identification of use cases and promoting adoption of AI throughout the user groups.
Use cases for AI
The first step in any organisation’s AI journey is to identify the right use cases that align with their business objectives.
A key challenge is that organisations that implement AI often have to stop because they don't have a data strategy in place. Or, if they do have an established data strategy in place, there are no Data Governance measures in place to ensure that data isn't mishandled, overshared and is of a good quality.
FDM’s AI offering
All our consultants are introduced to AI as a core capability. Our AI Engineers can support with Prompt Engineering to building custom AI solutions for your business. We are working with partners to encourage ethical use of AI and how to govern this through GRC (Governance, Risk and Compliance) tools.
The AI revolution is well and truly underway, but its success depends on how well-prepared organisations are to adopt it – with a combination of strategy and talent to implement it.
FDM is conducting a global survey of business leaders to determine how AI is being adopted and deployed. Please fill in this short 5-minute questionnaire. Answers will be treated anonymously, and your data and contact details will only be used for the purpose of this research. The results of the study will be published in a whitepaper and will be shared with those who responded via the email they used to respond.
Authors
Rod Flavell
Rod is the founder and Chief Executive Officer of FDM Group and has more than 40 years’ experience in the technology sector. He has been instrumental in the development of the Group into an international, award winning employer with a prestigious client base operating in multiple markets. Rod is a strong advocate of improving diversity in the technology industry, as demonstrated by the Group’s Women in Tech, Returners Programme, Ex-Forces and veteran career transition initiatives. In 2019, he was featured in the Management Today Agents of Change Power List for the second consecutive year. He was also featured in the Yahoo HERoes Top Advocate Executives of 2019 for his work promoting gender equality in the workplace.
