AI or Artificial Intelligence. ChatGPT, Bard, X’s version, self-driving cars, visual AI. It feels like the birth of personal web pages in the 90s and websites built using Microsoft Frontpage or GeoCities. It’s been one of the most exciting technological steps for a long time. AI Integration in business is, in my experience, a significant and ongoing challenge for business leaders.
Businesses are scrambling to integrate AI in any capacity to say they offer their customers some kind of AI trying to outcompete in a 21st-century space race.
AI is creating opportunity and excitement but also a lot of fear. Before making moves, it’s wise to step back and look logically at where to integrate AI into your business, but before that, does it warrant it? Where do you implement it?
Before you begin
You’ve decided your business needs artificial intelligence; everybody is talking about it, and you want to market it as a product feature. First, is it artificial intelligence (AI) that you need?
If you’re thinking about how AI fits into your business, you’ve probably experimented with services like ChatGPT, and you know you can ask almost anything. You’ve probably seen videos and people selling tutorials and courses on it. You need to remember that it’s not all sunshine and joy.
What value are you adding?
The AI needs to fit your business and justify its existence. If you are trying to make AI work in your business, you will likely fail and end up with a solution that provides no real value and increases costs and technical debt.
If you are just integrating a GPT-based AI service via an API, your only natural addition is that you are automating something. Is that enough value? If your client uses ChatGPT, they can ask the same questions you are asking.
Who’s permission do you need?
You need to consider if you have permission to do it. Unless you obtain permission, you may face issues under privacy laws – consider GDPR implications.
Passing information to external organisations may break contract clauses and violate privacy laws. It could raise privacy and ethical issues. I wrote about ethics in a previous article and expanded on this point.
What is your tolerance for wrong?
Any AI will get things wrong at some point.
If you run a B2C business, you may have a higher tolerance for wrong. Midjourney, DALL-E, they get stuff wrong. ChatGPT does. But we accept it.
Suppose you are a business reviewing fraudulent transactions, deciding if somebody broke a contract (marketing, business, KYC, etc.), or in the news/media sectors, the margin for error is near 0.
Publishing a news article with the wrong figures can cost businesses significant money. Equally, if you are working in areas of compliance, your AI could wrongly cut somebody’s income stream off.
People sometimes refer to this as “false positives”. Work out what you can tolerate as a false positive, and see if the AI can perform under that threshold. Make sure you are not creating any extra liability.
AI Integration in business, but where is right?
If you try to find a way to make AI fit into your business, you will create a higher cost and a solution that doesn’t provide any value because you forced AI into a place it doesn’t need to be.
If using a translation feature is one idea, determine if your customers are asking for it. Translating some text to Gujarati is cool, but do you need it?
Instead, look at what your customers are telling you. Are they presenting you an issue where AI can help solve it? If you run a podcasting service, could you use AI to transcribe audio, which could draft a blog post? Do you need to provide any visual assistance? Do you need to scan invoices or convert written notes into text?
Have you asked your customers how they use ChatGPT in their day-to-day work? If they’re not, then do they need it in your product?
If your business doesn’t need AI, it’s okay to say so and watch from the sidelines. You shouldn’t use it if it’s not needed because the costs are not insignificant.
Do you need to use an AI Service to add AI?
I mentioned earlier that AI will get things wrong. AI systems, like ChatGPT, don’t naturally provide any method to validate the results of prompts. You don’t know the decision path or how AI got to the result that it did. We have seen AI give false citations, and its delay in knowing information holds it back.
Those limitations mean that you run a risk that a generative AI may give you and your clients false information or give you no information, depending on the timeline of your data. Can you tolerate that?
Other solutions may be cheaper and more reliable if you want to help understand context, topic, or language. Do those help you? AWS Translate might help with some language tasks. AWS Comprehend can help you with sentiment and language detection, a feature of ML (machine learning), a separate type of AI.
ChatGPT might tell you if an image contains a dog, but cheaper and better alternatives exist because of the details those challenges bring. Any AI integration in your business, must accept these risks.
Will your customers pay for it?
You can go to ChatGPT and ask, “In the style of a tabloid newspaper, write about vegan cheese. Give the article a witty headline with a controversial tone.“ and it will do it. If you can do that, your clients can also do that. If you plan on integrating with a generative API, you will charge a premium on something your customers can do themselves. Your only USP is that the user doesn’t need to ask ChatGPT themselves; you do it. Is that what your customers are willing to pay for?
Do you have Play-Doh?
At some point, every child had Play-Doh given to them. My grandmother used to make it for me. Play-Doh was a putty to clean coal soot off the walls of houses with coal burners. Over two decades of production as a cleaner, as coal sales declined, it was reborn into what we know as Play-Doh.
The history is fascinating; it was made in the same factory as the cleaner, using the same equipment; it had a slightly different recipe, went into child-focused packaging and smelt of almonds.
It’s a great example of finding new uses for something you already have.
Have you explored new ways to use AI that you’ve not previously thought of? Do you have volumes of statistical data which could create new models? Could you search different media for subjects?
You could explore generating synthetic data if you are a business with a large volume of transactional data. Synthetic data increases user privacy and allows for custom scenarios and testing of different conditions. Financial models could be fed synthetic data where a population has 10% more illness or fraudulent bank transactions among 18-25 year old increases 3%.
Do you have many documents tagged that could feed a recommendation engine? If you have many documents tagged by different users, you could use collaborative filtering to find relationships between documents.
AI will likely create a divergence in some business models into areas we have not seen before, just like Play-Doh.
Summary
AI Integration in your business needs careful consideration, proper thought and testing. Check in with your customers to see if they have any advice, but keep looking to see if you are uncovering something new.