Artificial intelligence (AI) is not for the fainthearted. An AI project does not follow the calm predictable waters of software development. It is a step into the unknown, an adventure to find buried treasure. You may find the treasure, you may have to turn back or you may end up a long way from where you expected with an even bigger haul.
That's an exciting but daunting prospect for businesses, especially those used to well-defined projects with clear goals and milestones, as in software development.
There is no escaping that AI has a much higher risk-reward factor than most technology projects. But with the right approach, risk can be managed and rewards reaped.
Setting out on your AI journey
Like an adventure, most AI projects start with an idea of where you want to end up. But you must also accept that, at some point along the way, you might find a shortcut or a longer route that is ultimately more rewarding or even discover another destination that is even better than the one you are heading for.
Like any good adventure, these possibilities should be embraced. But you should also approach them with a framework for how to navigate the unexpected. Just as you should be able to distinguish the road to opportunity from the road to certain death, you should have the tools to spot an AI solution with huge potential from one that should be abandoned.
Keep your options open from the start
We talk about AI projects, but we should really talk about AI portfolios. All AI projects come with risk. There is no question that the best ones deliver a huge return on investment (ROI); project after project proves this. But many deliver very low value, cost more than they deliver or fail. Don’t head single-mindedly toward your destination. A good AI program explores lots of options before picking the best ones.
Companies should simultaneously run multiple AI projects aligned to their business goals. They should know what they are trying to achieve and set metrics for assessing success rather than specific milestones.
Take an AI project that predicts component failure. The fact that it is working is not enough; it must also deliver more in savings (e.g., avoiding breakdowns or reducing downtime from unnecessary maintenance) than it cost to build. If one route to your destination is showing promise but not enough value, abandon it.
By looking across the portfolio at key points, AI leads can assess how different approaches perform against metrics and drop the worst-performing ones, allowing them to shift their budgets to the more promising ones. If a project is showing twofold ROI and another is showing a tenfold, it’s a no-brainer where to focus resources.
As any good entrepreneur or adventurer knows, failure can be good. Projects not worth progressing may still hold valuable lessons for projects that are. But if you put all your eggs in one basket early on, you might be stuck heading for twofold returns when tenfold were just a decision away.
Don’t settle halfway through
Another risk on the AI adventure is getting too excited by the first hint of success.
Often the subject matter experts (the chemists or engineers or marketers) will have a clear idea of what they want to achieve and a firm belief in what data is driving what outcome. While the perspective of the subject matter expert is vital in framing the problem, sometimes they are too close to it to have a fully informed business perspective.
Time and again we find that stepping back leads to a better result. We often find that there are useful data sources that customers don’t think to tell us about because they do not fit their own initial models, overlooking the potential of more optimal approaches. Or they will find a correlation they expected to find (e.g., temperature change improves productivity) and throw all their resources into proving it rather than exploring whether the behavior is being driven by other variables.
Understanding potential pitfalls helps you make better decisions.
Have a decision making framework for your journey
Any adventure involves a willingness to take risks. But you come to it with prior understanding of how the world works, which you apply to new situations to help you make good decisions.
AI is not always intuitive, but best practices can be adopted and the application of tried and tested governance frameworks can tilt the odds in your favor. Governance frameworks must be agile, allow structured experimentation, include multiple review points that force teams to assess if their projects’ data and models are reliable and effective and, ultimately, determine which AI projects to take forward.
Make sure your decisions are driven by business goals, not technical perfection. Many models could be better, but do they need to be? Just because you can get an error rate down from 3% to 2% doesn’t mean you should. If 3% is more than good enough, end the project and start deploying the model.
In conclusion, treat your AI project as a structured journey. Create a portfolio of options. Keep an open and agile mindset. Regularly assess if your choices are delivering. Never be afraid to abandon projects and move on to more promising routes. Finally, always reflect on what you’ve learned.