If AI is the answer, then what is the question?
Unfortunately, this philosophical musing is how it feels many are approaching the opportunity of Artificial Intelligence today.
This isn’t just a fast-track to pointless or ineffective investment but it is also a way to ensure you are not fully achieving the potential that might be possible for your given business and problem.
As Einstein said (some version of):
"If I had only one hour to save the world, I would spend 55 minutes defining the problem, and only 5 minutes finding the solution"
So how can we turn the current approach on its head to increase our chances of real success?
By being truly critical and assessing our opportunity and plan from a variety of angles. I propose channelling your inner toddler and asking ALL the questions up front (and keep on asking). I’m going to lay out what those starting questions should be.
Why?
In the words of Simon Sinek: Start with Why
The biggest question you will ask and one I have spoken about before (check out my previous article: 5 ways you are sabotaging AI as a leader): Why AI?
What are you actually trying to achieve, beyond just telling your shareholders or customers that you used AI?
The "Why" has to come first, it can’t just be for AI’s sake. You need an objective, a goal, a north star to work towards.
Then, and only then, can you assess whether AI is the right tool to get you there.
This might be about Business growth, scale you can’t achieve without AI or user value you can’t deliver.
It might be about cost saving through efficiency or productivity.
There are lots of great reasons "Why AI" but you won’t be able to assess whether you are achieving them if you don’t define them AND align on them up front.
Once you do, you have something you can always come back to, to assess the success of your project regardless the complexity of implementation. Are we closer to our Why?
This should always, always, relate back to the existing Strategy or business goals. If it doesn’t, you need a very strong justification for why this diversion is worthwhile.
What?
Artificial Intelligence is the use of computer systems to perform tasks that previously required human intelligence.
Despite the attention that has been given to Generative AI in the last 2 years, it is not the only version of AI.
To understand the AI for your problem you need to get a really good view of the workflow or process that you are looking to augment or automate with computer intelligence.
You then need to understand the data that is available to build this AI on. Artificial intelligence is not born from thin air – the information that it learns from is data – and this comes in many forms.
You need to understand the end-to-end process to establish whether AI can do all of it or fit into just part. Maybe today the answer is part and in future it will be all, that’s ok, but be clear.
You also need to understand the reality of the data available and any limitations (inaccuracy, incompleteness, bias) that might skew the results of your algorithm once developed.
This is not a question solely for a technical engineer or data scientist. There is a part of this that business leaders and stakeholders should also be involved in as they may understand significantly more about the nuts and bolts of the "before" reality.
Consider a use case such as B2B sales interactions and workflow like client follow-ups. If a technical team choose to automate email interactions using AI to compose the email copy then they should of course speak to the existing client facing teams. This will be beneficial to understand any nuance about when clients are contacted, decision making factors to determine frequency or recipients and appropriate copy for the email. It is also important to speak to this team, or any others, involved in capturing the information we have about those clients that we might later use to generate proposed copy because they will know the process involved in logging this data and any red flags about directly translating to external messaging.
Who?
In terms of importance, I would usually like this to be higher in the list. But for intuitive flow I am putting it here so that conceptually you have already considered what it is you are planning to build.
Who is all about the people. All the people.
What person did this work before AI?
What person will work with this AI?
Who will engage with the outputs of this AI (employee/customer)?
Who might be impacted in an indirect way by this AI?
When thinking about the Who you can start small – who are the experts I need to ask to build this well? Who has the domain knowledge that I should include?
But you also need to bring empathy in very quickly. Who will feel threatened or disrupted by what I am building? Is someone going to lose their job (or fear that they will)? Am I reliant on someone else adopting what I build?
If I previously only rolled out reliable, digital products directly to customers – what is the worst version of how it will impact the experience, service or product they receive if we move to AI?
Is the answer to that question the same for all customer groups or does it change for subcohorts or, particularly, vulnerable groups?
This is something that can easily come up when our data is image based – either in input or output. You may have heard the stories about Google’s algorithm mis-categorising images of Black people as gorillas. I have to assume that at an overall level this algorithm passed aggregate tests for accuracy but clearly there were instances where it had very extreme wrong outputs.
In marketing, we might find that our messaging makes sense in aggregate if that is how we assess our models but understanding how that accuracy breaks down for smaller customer groups would help us to refine the messaging further or exclude populationss where we have concerns about our output reliability.
Finally, at scale, are there any societal implications of what we are building?
Some of these questions may sound dramatic, particularly if the AI that you have in mind is internal or has an audience of one, but they are muscles we all need to be building.
Consider the full 360 degree potential impact of what you are building because AI is not as predictable as other technologies and overall accuracy doesn’t tell the full story. You can read more about the latter point in my article Please Make this AI less Accurate.
When?
Logistics aren’t always the funnest part of a project but that doesn’t mean we get to ignore them outright.
As early as you possibly can you should be assessing your "readiness" for the AI solution you want to undertake.
This readiness includes data availability and permission – which could have a significant lag time if you aren’t already collecting one or the other.
It also includes organisational readiness in terms of available resources (teams, technology) and prioritisation.
There may very well be simpler iterations that you can start with that will either get you quite far along your needs without relying on AI, allow you to build out a robust business case for full prioritisation or set you up to build better AI in future (e.g. building a digital solution that collects the necessary data).
If this is the case then the answer to When is the answer to
"When will we be able to build the AI that is the best solution to our problem?"
Your shareholders will, more often than not, want the answer to be "tomorrow" so spending time getting the best possible estimate for this is key to managing expectations and getting the necessary supports and dependencies signed off.
If you have doubts about the suitability of your data to be the fuel for a large AI implementation then getting a better understanding of the current state and any fixes required should be your number one priority.
Where/How?
Finally, How is this going to look? Where does it fit into your organisation?
These are the final practical questions that will become most relevant at deployment but you would be prudent to ask before you begin as other dependencies may pop up when you do.
This may be about which teams or other technologies will interface with what you are building. It may be an operational question about what changes in processes or KPIs will be required once this rolls out. It may even bring into question the core advantages of your organisation.
What start as small questions can become bigger when you follow the necessary threads. Organisations are made up of people so just as individual people are complex, organisations are much more so. The more rocks you turn over the more you learn.
If what you are building is a customer facing product then you will need to consider the full user flow that exists today and what part of that this will replace or augment. Who will engage directly with it and what can we expect them to do?
If you are building an internal process then how does that change your expectations of what your teams currently do day-to-day, do their goals or productivity targets need to adapt? Have you considered every team and role that has even a secondary impact from this change?
I have worked with clients that have spent $300k-$500k on external data and modelling environments that have never been used because they continued to incentivise their internal teams from the outputs of the previous infrastructure.
Sketch out what this will look like at the end and work backwards from there. If you really know your use case (what) and your relevant stakeholders (who) then a lot of this work should already be covered but ensure you are asking how they will engage, who owns what and who monitors what once deployment takes place. Otherwise, your lovely well thought through AI project may never get used.
Whatever your role, be the curious toddler
There are deep technical versions of each of these questions and there are 10,000ft view vision versions too. My advice is that everyone involved in an AI project, from the C-level sponsor to the data scientist, should be asking a version of each. Everyone should have thought about each.
If you are a data scientist that doesn’t know how the model they are currently coding fits into your company strategy or your customer’s benefit then this is a moment to massively increase your career influence and effectiveness.
If you are a non-technical CEO and you want to capitalise on the latest technology then this is the moment to be "AI critical" and to deliver value that many of your peers are overlooking.
Everyone, embrace your inner toddler. Ask Why, What, Who, When and Where/How, AI now.
And then repeat throughout the project and implementation.
— – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – –
If your business could benefit from external support in assessing the reality of the AI opportunity today, I can help.
Check out my website https://kate-minogue.com/.
Through a unique combined focus on People, Strategy and Data I am available for a range of consulting and advisory engagements to support and enhance how you deliver on your strategy across Business, Data and Execution challenges and opportunities. Follow me here or on Linkedin to learn more.
