I have been researching artificial intelligence tools within the Microsoft stack for a while and it will be of no surprise to anyone that a lot of my time has been spent focused on Copilots. There is a huge number of Copilots out there in all areas of the stack (see image below). There is even more AI capability within Azure AI Services with 1,600+ AI models available. These provide greater control and customisation for those looking to do code-first development.
I researched extensively the challenges around choosing the right tools and explored how to adopt and govern them. These aspects are crucial for getting value out of AI and last year, I spoke about AI adoption at conferences, user groups, and virtual sessions. You can find the deck here. Microsoft now also has great guidance on AI adoption, strategy and planning together with AI checklists you can find here.
This blog post provides my top 10 considerations for AI adoption aimed at getting value from introducing AI tools. Last year was about execution, this year is all about results…
1. With AI, start with WHY
Some organisations have suggested AI is their business strategy which I have always found a strange statement. Organisations should have had a business strategy and a reason for existing before AI came along. AI is not a business strategy, it is a catalyst for executing a business strategy quicker (and better!). So the first question has to be ‘what problem are you trying to solve with AI?’ – i.e. what are your use case(s) that AI can handle? Why would AI be a good tool to handle these use case(s) as opposed to other options? Microsoft has a great AI decision tree for its products for those needing guidance.
2. Data, Data, Data
AI is all about data and lots of it. The quality of AI outputs will be based on the quality of the data the model is trained on. Do you have enough good data to train the models on? Are you a data-driven organisation and have you considered how this will change in future (e.g. synthetic data)? Do not underestimate the sometimes hidden cost and effort involved in getting the data ready for AI.
3. Skills, and Decisions
Do the people who will use AI tools and solutions have the right skills? Do they have the knowledge to use them responsibly? Do those making decisions on AI tools understand enough about how they work and their limitations? Is it clear when it is better to build a custom solution vs buying an off-the-shelf solution? It is important to guarantee upskilling happens at all levels to make the right decisions that will lead to value. By 2028, over half of large AI models built from scratch will be abandoned. Rising costs, complexity, and technical debt will drive this issue according to Gartner. AI technology is evolving so quickly that ‘technical debt’ conversations are not far away.
End users of the technology also face challenges. They need to develop new habits and mindsets to remember to turn to AI when in need but old habits die hard. Every organisation will have a different maturity level with most of us in the implementing stage:
4. Ethical Considerations
If the organisation you work for stated they intended to use an AI model to decide on employee promotions or salary increases, what would your reaction be? It’s important to consider any ethical implications from using AI models. If the data they are based on is biased, the models could be biased too. AI models are constantly learning and are inherently unpredictable. Therefore, human review and accountability are key. Microsoft’s responsible AI principles cover a lot of key areas for consideration:
I would also add two more:
- Scalability – is the solution capable of flexing with the organisation’s changing needs? Or will it be short lived by design?
- Sustainability – AI models are power hungry. Consideration needs to be given to the vast infrastructure required to support them. We also need to assess whether the environmental impact is being considered.
5. Integration with Existing Systems
AI models, whether custom made or off-the-shelf, will be interacting with existing infrastructure and software. When evaluating options, it is important to consider their capability to integrate with the existing ecosystem. This ensures they are the right fit. Selecting the right tool is crucial for long-term compatibility.
6. Security and Privacy
AI models use data. This one goes without saying. If an AI model is free to use, ask what the model owner is doing with your data. It is important to know how your information is handled. If it is free, you are the product. Security and privacy management is crucial. We must ensure AI is implemented responsibly. Rogue AI, AI jailbreaks, and AI scams are on the rise. Will we soon see an AI security breach in the news? Microsoft have created two open-source packages aimed at privacy and security, SmartNoise and Counterfit, worth looking into. It’s important to remember that everyone can bring their own AI to work. Guidance must be provided on what is and isn’t allowed in that respect as well.
7. Scalability and Flexibility
The use cases for AI can change over time. AI usage can also increase exponentially. It is important the AI model or solution chosen is capable of both scaling with new or changing use cases. It must also be flexible enough to enable higher usage over time. If AI tool decisions are being made in isolation, by e.g. different departments not communicating, multiple tools might be introduced in the organisation. These tools may address similar use cases. These tools may then need to be abandoned. This happens because they cannot scale or flex with the organisation’s needs a few months later. Having oversight of the AI requirements across the organisation is crucial. Having an AI strategy can mitigate this risk in most cases.
8. Cost and ROI
It is no secret that AI tools/solutions and even the running of the models themselves can be extremely costly. It is important to consider the long term financial investment. One should focus on years rather than months when it comes to AI as the time to realise value out of an AI solution can vary. A clear investment plan needs to be in place. The use cases need to be prioritised, approved, and socialised within the organisation and there should be a clear approach to how the return on the investment will be measured. AI solutions focused on personal productivity can be particularly difficult for calculating ROI. Conversely, AI solutions aimed at organisational enhancement commonly have clear metrics and KPIs are usually defined from the start. Before implementation it is important to have governance and monitoring in place to allow ROI to be measured. It is also worth considering creating an AI Center of Excellence to link ROI back to business objectives too.
9. Change Management
Software that is not used by humans is of little value. We can introduce extremely complex, brilliant, and expensive AI solutions. They may solve no problem whatsoever. Consequently, adoption will be low, and the benefits will be none. Introducing AI tools is like any other new technology. Some people will be categorically against it and will not touch it. Others will be very excited about its capabilities. There will also be those in between the two. It is important to guide the end users of the AI solutions. They must understand why AI is being introduced. It’s crucial to explain what use case it is addressing. Users need to know what problem it is aimed at solving.
We then need to make it ok to fail and try again. AI is new and unpredictable. Not everything will work the first time. Not everything will be useful. Some of it will be just nonsense. Establishing a change management process from the beginning is crucial. This ensures adoption starts high and also remains high as AI models, updates, and new features are introduced. AI champions or super users in an organisation can have a massive impact on this. Identify them early and involve them as soon as possible!
10. Constant Feedback
Last but not least is the need for constant feedback – both from the human and the technical sides. End users of AI solutions need to have a process for providing feedback. They need to identify what isn’t working, what could be done better and what should be done next. Any AI strategy, governance and policies introduced also need frequent review to check they are still relevant. AI is changing the way we work. We can only adapt quickly enough when the feedback loop is in place. This loop identifies where AI is winning and losing the battle.
Finally some very useful links for those of you that stayed with it until the end! Check out:
- the Microsoft AI Learning Hub for a vast amount of training available based on your role
- the Microsoft Copilot Learning Hub to understand how to make the most of Copilots
- The Microsoft UK AI Hub for information and latest reports on how AI is being used specifically in the UK
And don’t forget to let me know how your AI adoption is going!
Areti Iles 