The 10 Biggest Mistakes Companies make when creating an AI Strategy
Introduction
Given all the noise surrounding it, you could be forgiven for assuming that generative AI is the only technology that matters right now. Rarely have we witnessed a technology that has captivated the business world as quickly and with such impact.
Despite being shrouded in uncertainties and conditions, AI is poised to fundamentally transform our daily lives and professional environments. According to Cisco, 84% of companies think AI will significantly impact their business operations. But only if it is implemented and scaled correctly.
In this article, we consider 10 mistakes companies can make when setting off on their AI journey.
The 10 Biggest Mistakes Companies make when creating an AI Strategy
Starting without clear AI goals, leading to misguided efforts. AI is not a panacea for a poor strategy, nor a bolt-on technology that just makes stuff work. The point of AI, data analytics, intelligent automation and other digital technologies is not about getting better at digital; it is about building new value in line with your business strategy. Start with the strategy and develop your AI goals accordingly.
Overlooking data quality and governance can undermine AI's effectiveness. ''While AI can benefit society, business, and economies, it also creates new challenges for customers, users, and other stakeholders" (Cisco). If you ignore the ethical implications of AI use, you risk losing trust. If your employees and customers don't trust your AI, you will not succeed. It's imperative to implement clear governance over how AI-based solutions are developed, deployed, and operated, with sophisticated approaches to securing quality data.
Skimping on skilled AI talent can hinder progress and innovation. As opposed to replacing jobs, AI can be a channel for huge productivity gains for the business, but only if companies invest in the right skills and talent training. A top priority for organisations must be to build a future-ready AI workforce.
Underestimating the need for scalable AI infrastructure can severely limit growth. Scaled Artificial Intelligence deployments require bandwidth and processing power (GPUs). Organizational infrastructure must support the tremendous amount of bandwidth AI models need to train and run new AI solutions, regardless of whether they are on-site or in the cloud running on AWS Bedrock.
Neglecting the importance of user-friendly AI tools, and ongoing AI training, reduces the amount of AI adoption. As Apple has proven time and time again, if you make technology intuitive and appealing, then people will rush to use it. Don't, and they won't. AI education must start at the top of the organisation and cascade across every layer below. Everyone can benefit from new AI learning even if they are not creating; they can ideate and consume AI through user-friendly AI tools, e.g., ChatGPT, Claude, Sora, or conversational AI platforms.
Not engaging and convincing, stakeholders early, can lead to resistance and a lack of support (change management). Without people, even the best technology has no impact. The willingness to adopt AI varies greatly depending on one's level in the organisation. Board and leadership teams learn and get excited about generative AI's impact early on. Employees and middle management get to hear about AI much later and feel it's is being done to them rather than for them. In addition, with the media suggesting AI will be a job killer, it is no wonder we see late adopters resisting AI. Instead, organisations must create a dynamic conversation around the possibilities of AI—and how every role plays a part—to pave the path for willingness and engagement.
Neglecting the importance of data diversity affectis AI model accuracy. Few, if anyone, set out to deliver biased, non-value-adding AI business models, but AI is a complex field that requires specialised skills. As a result, AI systems can inadvertently invade privacy or make decisions that appear unfair or biased. The best AI model data is quality data. It is representative of the audience you are trying to model, and it is sufficient in volume, timeliness, completeness, availability, and non-biasedness. Anything else is poor quality data, and any decisions made due to bad quality data can lead to inaccurate outputs, system errors, and, in worst-case scenarios, serious harm to individuals and organisational reputation and likely lead to legal problems. 'High-quality, ready-to-use data formatted so that people and systems across an organisation can easily access and apply it—can deliver new business use cases as much as 90 per cent faster and reduce the total cost of ownership by 30 per cent.'' (McKinsey). But remember, for quality data to remain effective, it needs to be tested and validated regularly. Never create an AI model and walk away.
Failing to establish clear metrics for AI success, hindering measurable outcomes. The adage, "if you can measure it - you can manage it'' is true for most things in life, and AI is no exception. If you cannot tell whether you are winning or losing when using AI, then you will unlikely get any value from your investment. After all, no one likes to play a basketball game with a broken scoreboard! So, create metrics that track AI's success, e.g., percentage improvement in customer retention from building an AI retention mode.
Cybersecurity. Failing to protect data can result in a significant risk of data breaches and misuse of organisational data. Organisations must implement robust security protocols to protect their sensitive information from unauthorised access and maintain the integrity and reliability of AI systems. As AI and generative AI technologies become increasingly integrated into business and society, safeguarding against potential threats is crucial to prevent malicious activities to keep user trust and competitive advantage. Moreover, comprehensive cybersecurity strategies enable organizations to comply with regulatory requirements, minimising legal risks and reinforcing their reputation for data protection.
Treating AI as a One and Done Project. Implementing an AI strategy is not a one-time task but an ongoing journey. It demands consistent upkeep, updates to and fine-tuning of data, and adjustments to keep pace with evolving circumstances. Organisations that view AI as a static project rather than a dynamic process typically see their systems fall into obsolescence or lose effectiveness. Embrace a philosophy of perpetual enhancement for your AI initiatives. Continually assess, refine, and update your AI solutions to ensure they remain effective and precise as conditions and data change.
An AI Strategy Needs Time, Money and Executive Energy.
AI is too important to fail. It offers unprecedented opportunities for organisations willing to invest. Organisations that grasp the AI nettle can generate huge value, but success comes from building a range of supporting capabilities.
For example, adequately managing, funding and investing in AI strategies that consider hiring and developing the right talent and AI infrastructure, as well as address ethics, privacy and security risks and scalability.
Over the past three years, McKinsey noted that the digital and AI maturity spread between leaders and laggards has increased by 60 per cent (McKinsey). AI comes with risks and challenges.
Success is neither easy nor guaranteed, but if your organisation is unwilling to invest in AI, your competition will. The prize is there for the taking, but is your organisation the one that will take it?