Five questions to ask before investing in AI
A practical guide to making better commercial choices about where, when and how to use AI in your business.
AI presents enormous opportunities. But before choosing a tool, commissioning a project or expanding something you're already doing, it is critical to be clear about what you're trying to achieve, what needs to be true for AI to deliver it, and what you could change, compromise or become dependent on along the way.
These five questions will help you work through those choices:
What are we actually trying to achieve?
What does AI need to work here?
What do we need to stay in control?
What needs to change around it?
What happens when things go wrong?
You don't need perfect answers to every question before you begin. But you should understand the opportunity, the dependencies and the risks well enough to know where the gaps are, which ones matter and what you need to address before you go further.
1. What are we actually trying to achieve?
Start with the business outcome, not the technology.
Before deciding how you might use AI, define the problem you're trying to solve or the opportunity you want to pursue. Be specific about where the value lies and what would need to improve for the investment to be worthwhile.
Ask:
What are we trying to improve, change or make possible?
Who will benefit, and how?
Are we trying to reduce cost, increase capacity, improve effectiveness or create something new?
How significant is the opportunity?
How will we know whether we've succeeded?
Will this create an advantage, help us keep pace or simply reduce cost?
Could we achieve the same outcome more simply without AI?
You should be able to say:
We want to [outcome] because [commercial reason]. We'll know it's working when [measure]. AI is worth exploring because [reason].
A useful test: If your answer still starts with “we need to use AI to…”, go back a step. You should be able to describe the outcome you want without mentioning AI at all.
2. What does AI need to work here?
Understand what your intended outcome depends on.
AI cannot function without data, but having data isn't the same as having data that's fit for the job. Before going further, understand what information, technology and processes your particular application will rely on, and where weaknesses could affect the outcome.
What data does this application need, and where is it held?
Is it accurate, complete and current enough for what you're asking AI to do?
Is it consistently defined and understood across the business?
Can the right systems and people access it appropriately?
Are there restrictions on how that data can be used?
What systems and processes will the AI need to connect with or work alongside?
What external technology or suppliers will it depend on?
Which gaps need addressing before you proceed, and which can reasonably be managed?
By the end of this step, you should be able to say:
For this to work, we depend on [data, systems, processes and suppliers]. We know where the important gaps are and what we need to address before going further.
A useful test: Could you explain what your AI application depends on to someone outside the project, including where the weaknesses are? If the answer is simply “our data”, you probably don't understand the dependencies well enough yet.
3. What do we need to stay in control?
Understand what introducing AI changes about your risk and responsibilities.
Using AI can change how information moves through your business, who has access to it, what you depend on and where accountability sits. And AI may enter the business through functionality added to technology you already use, not just through deliberate new investment.
What information will the AI have access to, and where will it go?
Does that include customer data, commercially sensitive information or intellectual property?
Do we understand how that information will be used, stored and shared?
Who is responsible for deciding how AI can and can't be used?
Who is accountable for its outputs or actions?
Which systems, models and suppliers are we becoming dependent on?
What happens if a supplier changes its technology, terms or service?
Do our existing policies, processes and governance cover what we're actually doing?
How will we maintain visibility as the technology and its use change?
By the end of this step, you should be able to say:
We understand what information, systems and suppliers we're relying on, where responsibility sits and what controls we need to manage the risks appropriately.
A useful test: If you were asked tomorrow where AI is being used across your business, what information it has access to and who is responsible for it, could you answer?
4. What needs to change around it?
Think about how AI changes the work, not just the technology.
Introducing AI can change roles, responsibilities, processes and the skills people need. It can also change the knowledge and judgement you need to retain within the business. Understanding those changes early will help you plan for adoption without inadvertently losing capability you still need.
Whose work will change, and how?
What will people stop, start or do differently?
Which existing processes or responsibilities need to change?
What new skills will people need?
What knowledge and judgement do we still need people to retain?
Where will human judgement or intervention remain important?
What could prevent people from using AI appropriately and effectively?
How will we know whether the change is delivering the outcome we intended?
By the end of this step, you should be able to say:
We understand how people's work will change, what needs to change around it, what capability we need to retain and how we'll support people through the change.
A useful test: Imagine the AI is working exactly as intended six months from now. Can you describe what's different about how people actually work – not just what the technology is doing?
5. What happens when things go wrong?
Plan for failure before you need to respond to it.
AI doesn't need to fail completely to create a problem. It can produce an incorrect answer, behave unexpectedly, become unavailable or be used in ways you hadn't anticipated. A supplier can fail, or data and information can be exposed or compromised. Think through the failures that matter and how you would recognise and respond to them.
What could realistically go wrong, and which failures would matter most?
How would we know something was wrong?
When does a person need to intervene?
Who needs to know, and who has responsibility for responding?
What happens if the AI, an underlying system or a supplier becomes unavailable?
Can we continue to operate without it?
What would we do if data or information were exposed or compromised?
How will we learn from problems and change what we do?
By the end of this step, you should be able to say:
We understand the failures that matter, how we'll identify them, who will respond and how we'll protect the business and continue operating when something goes wrong.
A useful test: If the AI stopped working, produced a plausible but wrong answer, or exposed information tomorrow, would the right people know what to do?
Make a deliberate choice
You don't need perfect answers to every question before moving forward. But you should understand where the gaps are, which ones matter and what you're prepared to accept.
Your answers may give you confidence to proceed. They may uncover something you need to address first. Or they may make you question whether AI is the right answer at all.
All three are useful outcomes.