Before we jump into creating an AI proof of concept for our clients, we always ask them a series of questions so that they understand how ready (or not) their business is for this kind of technology to help them.
You might have determined that implementing AI in your business would be strategically viable and in line with any compliance or regulatory conditions of your industry, but it’s also important to consider both financial and technical viability of any such project.
In this article, we look at some of these considerations and then give you access to our free toolkit so that you can check your own readiness for AI, scored from 0 to 100.
Is there a clear business advantage to using AI to solve the problem at hand? Think about what a solution could offer the business financially. Here are some ways of considering the impact of taking action:
- Cost saving: could an AI solution produce direct capital or operational savings?
- Revenue generation: could an AI solution lead to higher revenues?
- Quality improvement: could an AI solution reduce failure rates or improve product quality so that the output is more reliable (saving money and potentially increasing your market share)?
- Customer satisfaction: could an AI solution help to improve the user experience, reduce complaints, and decrease customer churn?
Look at these factors together and then consider whether there is a clear ROI from using AI.
We always advise customers to focus their efforts – and hence ours – on the projects that have the greatest potential to provide the most bang for their buck.
When the introduction of AI could help the business save a lot of money or take a massive competitive stride, that’s when we get excited about digging into the tech work to make it happen.
Unless there’s a compelling business case, we’d tend to advise against moving forward with a technical audit (because we don’t want to burn your budget for no good reason).
Think about the specific technical goals of the project. What exactly are you trying to achieve? Unless this is really clear up front, it would be a waste of time to build a proof of concept to show that AI could help.
Whenever we’re engaged to work on an AI project, we always look to see whether any precedents have been set for AI to provide a similar solution before.
It’s risky to go ahead with an AI project that pushes the bounds of possibility. That’s not to say that a project couldn’t go ahead if it hadn’t been done elsewhere before – that would mean there’d never be any innovation – but it’s important to make such a decision with your eyes open.
If something hasn’t been done before, a safer play would be to use AI for a different purpose that would still provide a clear business benefit. That comes back to making the business case for using AI in a way that’s going to deliver the best results for the business.
Any AI system can only be as good as the data supplied to it. If you were to analyse that data manually, would you be able to understand it as it is? And does that data give you a complete picture of all you need to know to answer whichever problems you’re trying to solve?
Good data can be thought of as that which meets the following criteria:
- Clean & readable: the data must be understandable and not full of errors and exceptions.
- Accessible: the data must be stored in systems where it can easily be accessed and processed rather than in closed silos.
- Categorised & complete: the data must represent all scenarios to be covered.
- Unbiased: the data must be representative to avoid the risk of skewing the behaviour of AI processing.
Even with good data, it’s entirely possible that AI isn’t really needed. If your work is in an area where you see clear correlations in your data, then traditional statistical methods are likely to be sufficient to meet the needs of the business. Put another way, implementing AI for AI’s sake might place an avoidable financial or technical burden on your operation.
On the other hand, where there are a lot of inputs and correlations aren’t clear, AI is great at revealing and taking advantage of underlying patterns to make processes more efficient.
Even if all of the above are feasible, you’ll need to keep in mind that using AI means putting an ongoing technical demand on your systems.
So, do you have a sufficiently modern platform to handle the analytic load that an AI implementation would place on it? It’s possible that you’ll need to go through an integration modernisation step before you can really get the benefit of adding AI to your process.
Looking to learn more about machine learning and AI? Check out our AI and Machine Learning explainer page.