AI will continue to innovate at pace through this year and beyond, and we will also see the emergence of AI regulatory legislation. The EU AI act is leading this charge and will impact not just EU member states, but those that want to trade with them, too. As with a lot of the EU legislation, we should expect the AI Act to form the template for other countries to adopt, as we saw with the legislation around GDPR.
Traditional approaches to delivering organisational value will need to have more flexibility when adopting AI, supported by a solid foundation of ethical usage, governance, and security. There are many foundational steps that are needed to ensure a successful AI adoption path. These include establishing ethical and regulatory guardrails in place through establishing a set of AI governance policies and processes, increasing the level of AI literacy across the organisation both for AI consumers and AI solution developers, and critically ensuring that the data that will fuel the AI engine is available, trusted and secure.
However today we are focussing on the key early step of deciding how AI will be adopted in your organisation. At an elevated level there are four routes to AI adoption and an organisation can apply any or all of them to meet their AI needs:
Use of Open Access AI Tools
It is highly likely that anyone reading this post will have experimented with generative AI tools either personally or professionally. Individuals across organisations are reaching to these tools to aid their day-to-day activities. This can be a great driver of efficiency gains but can carry a level of risk if the adopters do not understand both how these tools work. All open access tools are only as good as the data they have been trained on and users need to understand their limitations as well as their strengths. There are also risks around IP usage and sharing of sensitive information, users need to understand how the tools store data from prompts and if it risks any regulatory challenges. It is crucial that an AI literacy programme is started to provide this understanding to mitigate these risks.
AI -Enabled Software Solutions
Many commercial off-the-shelf software solutions are becoming empowered by AI. This can equally apply to Software as a Service (SaaS) platforms as it can to traditional software solutions installed and managed in-house. For example, a SaaS Customer Relationship Management (CRM) package can adopt AI for better forecasting and demand planning into the production cycle.
These software solutions can range from cross industry capabilities for departments like HR, legal and finance to point solutions addressing domain specific use cases. One defining type of software solution will be the growing number of CoPilot applications being adopted. These will act as general assistants or even dedicated specialised software for specific roles or business processes.
These tools will create huge opportunities to equip your employees with capabilities that enable them to be more creative, and efficient whilst being part of the AI journey. However, as with all modern technologies, there is a need for care and education coupled with a real assessment of the data the tools will use and store to support their user base.
It is likely that adoption of these tools will become accepted good practice across industry, and it may become mandatory for organisations to adopt them to remain competitive.
Given the emerging AI regulatory legislation that holds organisations accountable for ethical use of AI, it is extremely important that the use of AI- enabled software is captured across and monitored. The ease of use of these technologies means they are prone to the risks of shadow IT adoption and again an AI literacy programme highlighting these risks paired with a pragmatic process to support rapid experimentation and adoption mitigates this.
Leveraging Open-Source AI and off the shelf AI tools
If you choose to develop your own AI solutions you can leverage open-source tools, libraries, and models. In addition, many hardware and cloud providers are creating tools and libraries that accelerate the development of AI solutions. This includes off the shelf models, use case specific playbooks, and tools around AI driven capabilities such as computer vision and natural language processing.
One important subset of this will be the use of foundational models. A foundational model is a pre-trained large language model (LLM) that can be extended with further protected training with your own data. Using this adoption path can provide the best of both worlds with the foundational model having a wealth of public, verified and trusted data points coupled with your ring-fenced data that provides competitive edge.
Care must be taken to ensure that the foundational model not only supports your needs for the solution but also meets your security and regulatory requirements. With AI regulation coming, the solutions will need to be explainable and traceable – effectively so they can show their working by explaining how they achieved their result.
If you choose this adoption path, you will need to invest in enabling the solution teams with both AI skills and the tools to develop paired with solid standards and processes for building and deploying their creations. Given the already challenging skills shortage in AI expertise it is critical that an AI enablement and literacy plan is put in place to build your skills in house.
Custom AI Model Development
If you do not wish to adopt commercially available models, you can build your own custom AI. There is a huge advantage in that your solutions will be bespoke to your organisation’s needs and will offer high levels of competitive advantage, potentially at the expense of slowing down your route to value whilst building a technical debt that can be hard to regulate and maintain.
In Conclusion
AI Adoption will be a critical route of value for many organisations and understanding both what the value cases are for your industry as well as what route is best suited for each of these opportunities is a critical step in forming your AI strategy and roadmap.
None of the aforementioned approaches discussed here are mutually exclusive and it is likely that, as your AI adoption progresses, you will leverage multiple approaches. This, coupled with the pace of AI technology evolution, means your adoption strategy will need regular review as will the available value cases for you to adopt.
There are many other building blocks needed for successful AI adoption, all of which will require a commitment to evolve the skills and capabilities of your people. You can immediately ease the journey by starting an AI literacy programme to both educate key stakeholders in the opportunities AI will bring as well as to drive good adoption practices to accelerate value and mitigate AI risks. This will enable you to adopt an iterative approach to AI value creation that will deliver a consistent pipeline of benefit realisation through AI.