By Tristan Kromer

We all know AI is here and it is reshaping the world in unpredictable ways. From fake movie trailers to AI-powered scammers, we are all being impacted by AI – just some faster than others. And we innovators are no exception – AI is eating our world as well.

AI eating the world

AI is a double-edged sword. On one hand, teams can move faster, validate ideas with precision, and make data-driven decisions.

On the other hand, AI can reinforce biases and stifle creativity. For VPs of Innovation, the stakes are clear: adopt AI thoughtfully or risk mediocrity or even outright failure.

We can use AI to:

  • Ideate
  • Validate
  • And Decide

Ideate

As the name suggests, GenAI is generative. It’s pretty good at coming up with ideas.

A person doing quality control for lightbulbs

You can increase or decrease the amount of randomness to get more or less creative answers and ideas. You can get better ideas by bribing the AI (“I’ll give you $100 for genius ideas.”) or threatening them (“Perform, or I’ll fire you.”).

There has been a consistent improvement in idea quality over time, and it will continue to improve – but right now, the output is still that of a hungover intern.

Why should we outsource creativity to an algorithm that, by design, gives us the average of what the internet thinks might be a good idea? That is mediocrity by design.

I will freely admit, I often use ChatGPT to break the ice by suggesting a few ideas. It is a massive procrastination buster.

However, at least for my own use cases, I almost never use any of the ideas outright. Most AI-generated ideas range from half-decent to mediocre to outright dull.

But this will certainly change as the technology improves and evidence of superiority becomes apparent.

How to Use AI for Ideation

Of course, we should use GenAI for ideation. But not at the expense of human creativity.

Use AI in parallel with human creativity.

Break out a pack of sticky notes and actively generate ideas. Then, look at the AI output and take that as input for your next round of ideation.

A bad or mediocre idea can inspire a good one – so AI can be of great value during ideation. The more ideas, the better.

Validate

Validating an innovative concept requires getting around the Build-Measure-Learn loop as quickly as possible – and AI offers some powerful tools here:

Rapid prototyping

For prototyping, AI is a potent accelerant. Copilots allow even a mediocre hacker like myself to whip up a completed prototype in a programming language I don’t even know within a few hours. AI represents a massive democratization of programming skills.

 

A minimum viable product allows you to collect the maximum amount of information with minimum effort.

Test Design

AI shows some weaknesses in designing tests to validate a given concept.

Some custom models have been especially tweaked and given sufficient expert material to draw from (RAG – Retrieval Augmented Generation). However, most off-the-shelf AI models today still act like unmotivated students fresh from their MBA class—all theory, no common sense – and even less creativity.

They will often suggest surveys where customer interviews are more appropriate and draw conclusions with the unfounded certainty of an overpaid consultant.

Hey look I'm a scientist and I'm carrying science things!

Synthetic personas

The most significant use of generative AI in validation has been creating synthetic personas – an AI that attempts to respond as your customer would. This approach is groundbreaking and gaining steam.

No more running out to interview users; just interview an AI!

Robot talking to human

Using AI to predict a customer segment’s responses and behavior represents a profound disruption of the innovation process for better or worse because it’s garbage in, garbage out.

If we create our synthetic personas with unvalidated assumptions – the synthetic interviews will output nothing more than unvalidated assumptions. This approach becomes the most efficient possible way of achieving confirmation bias.

You shouldn’t trust an MBA grad doing customer interviews if they have zero field experience, even if they say they’re from McKinsey. So don’t trust an AI who will never, ever “get out of the building.”

How to Use AI for Validation

Let AI be your co-pilot and help identify assumptions or risks. Let them suggest ideas for tests. And definitely let them help build your MVP.

But continue to keep well-trained and experienced humans in the loop and making the decisions. Do not let your team get lazy and just default to the AI suggestions.

A great innovator uses research to build up intuition. That intuition is what leads to breakthrough ideas that go beyond the “most likely” idea from AI.

Decide

As of this writing, the general-purpose AI models we’ve tested lack the ability to proactively evaluate risks, quantify uncertainty, and make effective decisions.

An innovation board should determine value based on cost vs risk.

They will calculate the margin of error for A/B tests but will not suggest or be able to create impact models using probabilities.

They will make the “most likely” decisions – which are not the interesting, disruptive decisions like re-segmenting a market to focus on a small, growing niche.

However, this will change very quickly.

The underlying technology of generative AI uses vast training sets of data to construct a probabilistic model of what the possible responses to any given query are.

Training a model to accurately estimate the range of possible outcomes based on quantitative and qualitative data is only a matter of time, money, and data.

The Large Language Models are not well calibrated for this use case but will either get there soon or alternative models will be created with the same technology.

How to Use AI for Decisions

As of this writing, just don’t do it – but watch this space closely.

Once AI models take over as bookies for sports – you’ll know they can evaluate risks just as well, if not better, than a human. Unfortunately, they may be trained on data and decisions made by a non-representative sample of biased humans so that these models may become just as biased as humans.

You should actively evaluate custom-built models for better decision-making on an ongoing basis.

In the meantime, you can outperform AI decision-making by using a diverse group of people in your decision-making bodies and specifying decision criteria before gathering data.

Conclusion

AI can improve ideation, validation, and support decision-making in the future. However, current implementations of AI within innovation teams can cause laziness and mediocrity.

We are at risk of outsourcing our key differentiators as humans: creativity and empathy.

If we let our mental muscles atrophy, our team and our AI models will overlook unconventional strategies that drive disruptive innovation. We will efficiently reinforce the status quo and stifle bold, forward-thinking decisions.

Innovators should look to the future and integrate AI into innovation processes as a partner, but not at the expense of building our own skills.

We should try out synthetic personas, but only in conjunction with our own validation interviews.

We must learn how to spot and compensate for AI biases just as we should our own.

We are spending billions of dollars to train the latest AI models. How much are you spending on training your team?

For now, innovators still have an edge – but only if they invest in themselves as equal partners.

Special Thanks to Megan Wilder, Peter LePiane, Ryan MacCarrigan, & Dan Toma for reviewing and providing feedback on this article.