Skip to content 
Sarah Askew, our AI Innovation Lead at STRAT7, interviews two of her colleagues to discuss the need-to-knows for companies setting out in AI product development, including pitfalls, opportunities and some surprises too.
Artificial intelligence (AI) continues to rapidly evolve. Businesses are increasingly looking to AI-powered products and services to gain a competitive edge and drive revenue. But with such fast-evolving yet complex technology, and an avalanche of competitor and supplier offerings, how can companies be successful in AI product development?
As a professional working in AI Innovation and product development, I have my own thoughts, but what wisdom can those working in related specialisms offer? I decided to sit down with two of my colleagues from STRAT7, Andrew Dare (CTO) and Hasdeep Sethi (Data Scientist) to discuss the need-to-knows for those setting out in AI product development.
Sarah:
Andrew, Hasdeep, thanks so much for joining me to talk about AI Product Development today. What advice would you give to companies who are looking to develop their own AI products?
Andrew:
For me, it’s important to avoid getting too caught up in the technology. Instead, you should focus on finding a way to get to something that you can do quickly. If you dive too deep into the technology, you’ll end up spending all your time polishing the technology itself, rather than using it to solve a real problem or address a real need. This can lead to ‘paralysis by analysis’.
So, my advice would be to very quickly get to your business value as soon as you can, even if you have to ride roughshod over the technology to an extent.
Sarah:
I agree with you. I think there’s a feeling of panic amongst lots of people at the moment. The possibilities are endless, and thinking of all the things you could go after can be quite paralysing. It’s about finding what’s right for you and your business, amongst all that noise.
Andrew:
And just making a start, I think.
Sarah:
Yes. What are your thoughts Hasdeep?
Hasdeep:
I think it’s about highlighting two to three things that your company can really enable through artificial intelligence. In a perfect world, everybody would be able to build their own stuff, their own technology. But realistically, most companies don’t have the budget. That’s why it’s important to be resourceful and use or adapt what’s already out there. There’s a lot of unknowns so it can be helpful to try existing tools out on a trial basis before you commit to anything.
Sarah:
It’s a great point. Often, the best way to understand our requirements is to play with existing technology and ask ourselves – ‘does this work?’ and ‘what do we need?’.
Coming from a research background, I tend to approach building a product roadmap with a researcher’s head on. To me, it’s just like a client project where you’re trying to answer a brief.
Speak to your customers – your internal users and your clients – and define what their needs are.
And as you both touch on, get a sense of the tech and competitor offerings. But don’t get bogged down in the details. Define what value means for you, not everyone else.
Andrew:
I know it’s a cliche, but the principle about failing fast is really important, right? So, if you get down the road of developing something and you realise it’s not going to be right for your business or it’s going to be too difficult to do, stop! Pivot in another direction. Once the technology is up and running, you can start to revisit the areas that were maybe potentially too difficult.
Sarah:
This actually segues nicely into one of my other questions which is what’s the one thing that might surprise people about developing products in this space? You talk about failing fast. For me, it’s the pace of change working with AI. It’s like agile on steroids! You can fully expect to completely tear up your plans pretty quickly from one week to the next and rewrite them. You have to be comfortable with that level of discomfort and uncertainty I think. There’s new AI launches every day at the moment.
What would you say has been the one thing that has surprised you, Hasdeep?
Hasdeep:
Aside from the pace of change, I think it’s that much of AI product development isn’t actually about data science and machine learning. As we said before, it’s about finding the applications that will help users. My own personal interest is in understanding how these things work. 90% of the conversations we have within STRAT7 are about how can we use it to either uncover insights that are impossible or much harder to unearth through human analysis alone, or they’re about making efficiencies. We don’t really talk about the underlying technology too much. It’s about use cases and prioritising those use cases.
Sarah:
True, it’s quite counter, isn’t it? You might think you’d start with the technology. But that’s a bit of a fool’s game because you can end up trying to reverse engineer it into the thing you actually need it to be.
Andrew:
The biggest surprise for me is the lack of understanding. AI is an enormously broad church, but if you talk to most people about AI, they’ll only think of it in terms of generative AI. As far as they’re concerned, that’s what AI is. Well, it’s not.
At some point we, as an industry, need to do some real work around making sure people understand what we’re talking about.
Sarah:
We’ve naturally touched on quite a few pitfalls. Are there any other pitfalls that you think people starting out in AI product development should watch out for?
Andrew:
Believing the hype. Look at the amount of hallucinations that happen in ChatGPT. If you think about the people you know, it’s incredible how many believe everything that comes out of generative AI. They don’t sense-check it. I’ve used ChatGPT to write Python code and it ran first time and was actually really good code. But that’s something that’s quite quantifiable, because there’s only so many ways you can write code.
Where you’re asking opinions, you’re in a very different world altogether there and, for me, unquestioningly trusting what comes out too much is a major pitfall.
Hasdeep:
They’ll always be this human subjectivity, and human creativity, that’s difficult to replicate. I know these models are really good at being creative in a text generating sense, but they don’t understand the human perspective and the context. If you simply copy and paste the output into Word document, it’s quite dangerous.
The other pitfall for companies like ourselves is client confidentiality and sensitivity. We’re obviously very careful and sensible about that. So, we’re taking a more cautious approach compared to a company which doesn’t have to worry about that, making sure that anything we do build is secure and definitely not going out on any of the cloud providers or big tech providers.
Andrew:
I think the final point really is cost. If you’re not careful, especially if you want to go down the road of training your own models, the cost for that could ramp up very, very, very quickly indeed. If you look at the amount of time it takes to train models and the amount of intensive computing required, it’s very expensive. So, I think that’s why you really need to start leveraging what’s out there.
Sarah:
My final pitfall relates to cost, but is more about value – coming back to the idea of real value. There are so many new releases coming out at the moment and the more cynical part of me wonders ‘in how many cases are these actually going to bring value to the businesses that are launching them?’. You’ve got to define your own sense of value. Does it enable processes or colleagues? Is it enabling efficiency? Is it bringing in more money? Does it offer clients value? Is it increasing client satisfaction? And I think without an idea of that and how you’re going to actually measure what you do, it can be very hard to understand whether what you’re doing is really worth it and evidence that back to your stakeholders and customers.
Hasdeep:
It’s easy to be caught up in all the news and the hype. But I think it’s really useful just to focus on how it’s going to enable your particular processes. Having that focus is important.
Sarah:
Thank you both for joining me, some really interesting and practical points for companies to consider.
