The sins of AI adopters, part 2
The difficulty in identifying appropriate business use cases for AI projects remains one of the top barriers to AI implementation. Selecting an AI project, especially for the first time, can be challenging since it requires organizations to find the perfect balance between business and technology. Matching these two is at times tricky, but it has to be done. In the article about 12 challenges of AI adoption, I listed the lack of business alignment as one of the common challenges that companies face while implementing artificial intelligence. Why? Because identifying the business use case requires managers to have a deep understanding of AI technologies on top of their business expertise. What’s more, some companies start with AI too optimistically, without a strategic approach. And if you’ve read the previous article from the “Mistakes of AI adoption” series, you know that’s not the way to go.
Lack of business alignment in AI projects
The AI project you’re about to start needs to align with your overall business. It’s not a separate part that lives its own life. A data-driven company should make the business/tech alignment its priority. Adoption of any technology should be done with a strategy, and with the wider business goals in mind. So… what can go wrong this time?
Scenario 1: I want to be the next Amazon!
Don’t just copy the giants. Copying the exact solutions that Amazon uses won’t guarantee that you’ll reach the same level of success. You won’t become Amazon’s number 1 competitor simply because you implement the same technologies.
It’s important to remember that each business is different. Even the “smallest” differences matter. And often, these differences are not even that small. Your business model may be similar to that of Amazon in terms of online retail. So yes, you can use a variety of AI-powered solutions to drive results for your business. Like Amazon, you can implement recommender systems, dynamic pricing, or many variations of specialized predictive models. But what you want to predict – what you implement – is chosen based on your business goals, your needs, the available data, your budget… There’s a whole variety of factors influencing the selection of the appropriate AI tech that will support your business.
Scenario 2: So I want to build a model doing X
When organizations say they want to deploy a given model, there are 2 possibilities:
- They did their homework and they know that this model will bring the results they need.
- They heard or read about this solution and want to experience the benefits it promises, but they lack AI know-how to have a full understanding of such solution.
In the first case, it’s all good. The match between an actual business goal and the right technology is the way to go. However, in the latter case, problems may arise. For example, if I have an e-commerce store, I may be tempted to implement a recommender system. I know that many other companies, no matter what they sell, do that, so it seems like a reasonable idea. However, I am not aware of the limitations and challenges posed by such solutions, and because of that, I also don’t know whether I have all the right data to make this model work. So I assume it will work for me because it works for company ABC. And this way, we go back to scenario 1.
Scenario 3: I really don’t know what to do with AI
Believe it or not, but there still are organizations that want to implement AI just for the sake of having AI. This is not the best place to start, but it’s not hopeless. In such a case, the AI experts’ job is to discuss the business needs, problems, and processes to identify potential use cases. When implementing AI, it really is crucial that there is a good match between the business and the technology – and not just in a problem-solution way, but also between the domain specialists (people who will be using the model in their everyday work) and the data science team (who rock at building models but may be unaware of the exact needs it has to be adjusted to).
Scenario 4: Reinventing the wheel
Not every AI project has to be built, even if it’s a good idea. The truth is that many of the tools you want to use have probably been implemented in the services you may already be using, e.g. a CRM. So if what you’re planning to build is basically the same concept that is easily available in a SaaS format, you have to consider all the pros and cons. It’s important to note here that off-the-shelf solutions are often not flexible enough to provide results similar to those brought by custom made models. There are certain limitations to these off-the-shelf products, and as long as it works for you, it’s fine. Again: if you’re not certain which way to go with AI, get advice from an AI consultant, a data science team, or an AI development agency. Making these decisions can be difficult, it’s good to ask questions.
Selecting your AI project
AI projects, much like any other technology, should be chosen to strengthen your business and be feasible. But how do you know what to use when you’re constantly being bombarded with information about the new and wonderful ways AI improves any business?
The process of selecting your first AI project may be a part of creating your data strategy, or you might consider that even earlier on, just within your organization. This task requires both business and technical expertise, so even if you shortlist some projects, make sure you verify them before you invest any money in development.
There are some crucial questions you need to answer when considering AI projects.
What value does this project bring?
Does it reduce costs, increase revenue, enable new business opportunities? Create a thesis on how a given system will create value and what’s required for it to work. Many models may be useful in your industry, but you don’t want to go all-in – rather take it step by step and validate the use of AI in your organization first to mitigate the risk of wasting resources. Whatever project you’re considering, don’t ever lose sight of your business goals.
How does it work with my data?
Yes, data is a huge part of any AI system. Remember that the system will only be as good as the data that’s fed into it. However, you must consider the relationship between the value and the data. You may have lots of data about X but analyzing this data won’t bring real business value. And you may have less data about Y – but this is the Holy Grail you’re looking for and it’s worth a shot. Sometimes less data is enough when it’s good. Big Data is great, and always useful, but small and medium enterprises don’t have a chance to collect as vast amounts of data as Google or Amazon do. Instead, they can make very good use of the data they have, and selecting the right project is just about that.
Does it deliver a quick win?
The first AI project needs to bring results fast. Choose projects that can be conducted quickly (just so you know: “quickly” refers to up to 12 months; if you can do it under 6 months – great!) and have a high chance to succeed. The quick wins are important for a few reasons: they mitigate the risk of wasting money, they let you see how all this AI works, and they give you arguments to proceed – whether you need to persuade someone that it’s worth spending money on AI projects, or you need to prove that the model actually works and helps your business. When you have to wait long to see whether there’s any change, you can easily get discouraged. Don’t go long-term with AI right away, get to know it better first.
Do you have a team that will deliver?
A data science team (even a small one!) with good communication skills and the willingness to understand your business means the world to your AI project. Why? Because AI for business is still in its infancy, there aren’t a lot of highly skilled experts out there. They’re already working for someone else, or they’re terribly expensive. Even though team-building may not seem to be the priority at such an early stage, it’s important to decide what you even consider to be an option: an in-house team, outsourcing? Credible partners will help you accelerate your project. Starting off with an external partner is often a good choice, even if you’re planning to build your in-house team later. An external team brings valuable AI expertise right here right now, so you don’t have to wait a long time before you manage to build your own team and help them work efficiently together.
It’s the strategy again
Yes, I’m repeating myself. Have a strategy. Write it down, map it out, discuss it, adapt it, verify the results. It’s necessary. When you’re choosing the AI project you want to develop, the strategic approach will help you keep your expectations realistic and prepare well.
The first step to building your data strategy is to select possible use cases. Some of them will be better than others, or maybe you’ll end up with two amazing ideas. Is it good to start both at once? Since you’ve considered development time and quick wins, there shouldn’t be much risk related to having them developed at the same time. Sometimes, it may even increase the chances of building a really good model. Can you see how flexible it all is? There is no universal AI product for all companies, there is no “best” use case – it’s all very individual and should always (!) be adjusted to the company’s needs and requirements.