Artificial Intelligence (AI) has been transforming industries for years, with the availability of AI tools becoming more and more prolific, especially in the business world.
With more recent advances from OpenAI and rumblings of big announcements to come from Google, businesses of all shapes and sizes are starting to question how they can leverage more advanced AI and how they can start to integrate approaches such as generative AI to benefit their operations.
In this Q&A, we sit down with Jonathan Clarke, CEO of founded.ai and a PhD in Artificial Intelligence, to discuss how businesses can integrate AI into their existing infrastructures, and why they should take the time to do so.
Jonathan, you’re obviously a massive advocate for AI, but why do you think now is the time for businesses to sit up and take notice of it?
AI is no longer about just automating mundane tasks and increasing admin efficiencies. With the acceleration of AI technologies in recent years and the more ‘public’ availability of the code libraries that actually go into developing AI, the use cases for it are now much more wide-ranging. I think we’re truly entering the era of the Human Cyborg whereby humans come to rely increasingly on AI to do their jobs.
That doesn’t mean, however, that certain businesses are going to be locked out of adopting it because it’s so advanced. Quite the opposite – more and more businesses will be able to access AI for tasks they’d never even have thought of some five or ten years ago. In other words, startups and Enterprise businesses alike can really go to town on AI now.
What advantage do startups have when it comes to getting started with AI?
A lot of startups don’t have the issue of legacy systems in place, so they can set up new and ‘fit-for-purpose’ infrastructures and APIs quite quickly. Obviously this has a massive impact on their speed-to-market, but also that they won’t be constrained as their products or services develop.
And what disadvantages do they face?
Startups may lack the right processes to ensure that the right AI tech and set-up are chosen. We all know startups like to move fast (and break things!), but if they move too fast with AI, then they risk putting in place the wrong foundations for it to grow. This is why many startups find that as they grow and scale, they often have to rebuild their tech stacks and, of course, that’s not something investors like to hear just at the crucial growth moment!
Along a similar vein, proof-of-concept exercises involving third-party machine learning (ML) companies can be extremely expensive, and therefore financially impossible for some founders to take up.
How exactly does a startup get started with building an AI engine?
Firstly, make sure you have enough financial runway in place to support your build.
Then, start with your data!
The type and volume of data you have will determine what type of AI model you need to build – because there isn’t just one! You’ll also need to cross-reference the data with your end goal, i.e. what are you trying to achieve, what is the output you’re looking to generate for the end customer? Multiple models can be combined here to produce these outcomes so that you’re much closer to actual launch and market deployment.
Here’s a quick example. Say you’re a Legal Tech startup looking to develop a document processing AI. The AI algorithms here typically use Natural Language Processing (NLP) techniques to analyse and extract relevant information from the documents, interacting with the source data in several stages – document pre-processing, feature extraction, and classification.
You can see already from this simple overview that the ‘source data’ or documents are key to understanding what AI model is best, and how you need to configure it.
Now, there is a myth that you need oodles of source data to get started with AI. Whilst this is the ideal, it’s not the be-all and end-all. There are ways of working with low-level volumes of data in terms of enhancing it or configuring it in such a way to at least get the AI started.
What if I’m a startup founder just looking to build a prototype?
If you don’t have the capability in-house to do this, then you need to make sure that you choose a partner who will be upfront with you about costs, and about the trade-offs you should expect further on down the line if your prototype takes off.
That said, the code libraries and AI models now available mean you can take advantage of what are called ‘pre-built AI engines’ – something we specialise in at founded.ai. These are pre-built AI models that you can essentially pick and choose from ‘off the shelf’, but then customise according to your needs and requirements. They’re a really great way to get started and to check the feasibility of what you’re trying to do in the long run.
Why are Scaleups and Enterprise businesses unique when it comes to adopting AI?
Existing businesses and Enterprise organisations will undoubtedly have legacy systems in place and potentially lengthy processes to follow to get new systems set up and integrated.
And so they should – a certain amount of governance is key to ensuring the right AI is incorporated, and in the right way.
That said, these businesses do have a distinct advantage over earlier-stage ones in that they already have a tech team in place to handle the work. So, you’ll likely already have the necessary skills in your team, you just need to ensure you set out a clear business strategy first for them to follow.
Is integrating AI into an existing business a simple task?
To actually set up and integrate the AI is relatively simple, because of the abundance of APIs now available that work with a vast array of tech stacks. However, it’s the steps before that might take more time – some of these are to ensure proper buy-in and alignment with the rest of the company; some are to build a reliable and performant AI based on actual data; others are so that your tech team are 100% clear on what the goal is. Rest assured though that time spent upfront in getting clear on your strategy, your team, and even your potential blockers, is time well spent!
What steps should an Enterprise business follow to adopt AI?
Enterprise companies should always start with assigning a task force team to manage the project. This team should then be responsible for identifying those goals, obstacles, etc.
Most important though is the stage of analysis and requirements gathering that needs to take place early on. At founded.ai, we engage in what’s called a ‘Systems and Materials’ stage where we conduct exercises such as:
– Business analysis
– Systems & data review
– Integration plan
– Project plan
This all ensures that your next steps into evaluating and setting up your prototype are fit-for-purpose and that all potential eventualities have been considered in the configuration.
How do you work out which AI model you need to build?
This is very similar to Startups really – it’s a combination of what data is available and what your eventual outputs need to be. With Enterprise companies, there’s then the added factor of what tech stacks you have in place already and therefore what APIs are needed in order to facilitate tasks such as data ingestion.
For example, take a company in the tourism industry – like Airbnb or Booking.com that provide a number of different customer recommendations. Here, we’re looking at AI algorithms that collect data (via an API), extract certain features from it, build a predictive model from those features, and then generate personalised recommendations such as hotels, locations and even days out. Again, you can see the balancing of what data / tech stack are in place already and what eventual recommendations you want to create for the customer.
Many thanks Jonathan!
Do you have any closing thoughts for businesses looking to integrate AI into their existing infrastructures?
If it hasn’t already come across so far, I’d highlight again that there are a number of different AI models available – not just one!
For example, there’s lots of talk at the moment about generative AI, but don’t forget there’s also discriminative AI. While generative AI is focused on generating new data, discriminative AI is focused on classifying and making predictions based on existing data. Discriminative AI is actually what’s used in applications such as image classification, speech recognition, and sentiment analysis, and the use cases are huge – everything from fraud detection to personalised marketing.
In order to select the right AI model, or combination of models, you need to do that assessment and blueprinting work at the start.
Once this is established though, you can accelerate your dev time enormously by leveraging those ‘pre-built AI engines’ that I mentioned about.
Why is AI such a black box? And what can be done to fix this…
Where does the future of Enterprise AI lie? Prebuilt AI engines vs Low code apps…