Artificial intelligence applications are numerous and constantly flourishing around the world. It can be complicated to assess different AI providers and select the right partner, here are a few hints to find a partner that will bring real added value to your business.
Grow the right mindset
AI is not made for everyone. It is not a magic spell that will automate your entire process and solve all your problems. AI has a tremendous impact with comprehensive, digitized and agile companies that understand its core components, capabilities and limitations. According to McKinsey, the chances of generating profits through the use of AI are 50% higher for companies with strong digitalization experience.
You cannot hope for the AI to tackle each and every aspect of your daily operations. Most AI companies have the capability to create models to classify data or leverage it to process some outputs. However, each AI project carrier has their core expertise and these providers are not corporations AI labs for all machine learning needs.
Successful partnerships are made of tight collaboration and short iterations. AI will optimize its impact when it is calibrated and in the right conditions, it might require a certain level of flexibility and open-mindedness.
A good way to succeed in a collaboration with an AI provider resides in educating your teams about how it works and how the technology can be leveraged with qualified data and adequate conditions. Also, one of the key aspects is to help them understand that it will not replace their jobs but it will only make their lives easier and let them focus on high added value tasks.
Prune the surface
Shiny websites or sales representatives’ armies can be a synonym for a healthy and sustainable business but does not necessarily mean a technological advance, especially in AI. One way of benchmarking your potential AI provider is to have a quick look at the size of the R&D and AI scientists team and the capacity to attract top talent.
Brightest minds out of their engineering schools or from R&D labs are hunted by dozens of corporations or startups for their talent in machine learning or software engineering. They are recruited by startup CEOs, VPs or CTOs who exposed them to their view of what’s under the hood. It’s a good sign if the provider has great talents, it’s even better if they constantly attract more.
Even large corporations struggle to launch real applications with enormous sums engaged and battalions of engineers hired. One other way of evaluating the credibility of a project is to look at research papers, patents or academic partnerships that your provider might have. AI is not magic and it is a long-haul work that requires a long-term view.
Eventually, one interesting angle to look at is the core activity of a company. Make the difference between a startup that has been created around machine learning from others that started with digitization and opportunistically included machine learning in their roadmap for branding purposes. AI-born companies tend to have higher chances of success due to their culture and original structure.
Look at proven use cases
Differentiate companies that display a logo after advanced discussions from those with real metrics and ROI. Early adopters tend to see AI as a revenue enhancement, whereas companies that have partially experimented with AI see it more as a cost reduction. Thus, it is important to understand that the more familiar a company becomes with artificial intelligence, the more it realizes its potential for growth and not just for cost reduction.
Real strength of AI companies is to adapt and integrate their models in different conditions with various partners. In the end, metrics and real life usage are the most important aspects to look at. You can identify that by looking at the media coverage, real customers feedback or by testing the product yourself through a demo or a trial for example.
A corporation going to the production phase with an AI provider is an excellent sign, as the provider probably had to be validated dozens of times by different decision makers and proven its economical value and ROI. Some companies might be early and not in production with anyone yet and that’s fine. If you have the right mindset and believe enough in the team, the vision and the technology of a young startup you can give it a try through a pilot phase to validate the proof of concept.
Test it correctly
Eventually, before concretizing your discussions to a pilot or implementation phase with an AI provider you might want to test it in real conditions. There are two aspects to look at, the overall performance and quality of the technology and the seriousness of the provider.
First, depending on the use case your potential provider will let you try their solution either through a demonstration, a live test or a trial period. As much as possible try the solution in real conditions with your own data or usage. It will allow you to envision the possibilities it can offer but also the speed, the user experience and the limits. You should try the solution with representative examples that you’ll encounter in your daily operations, there is no point in trying all possible edge cases.
Then, you might find out that some providers are pushing to send the results asynchronously, if you can, ask them to do it live. This way you will experience the real time results, the speed and you will surely avoid companies using Mechanical Turks (with more human behind than AI).
In the end, the test results will give you a glimpse of the current state of the provider’s AI. Be reminded that you want to find a long term solution and that AI is evolving. The performance at a certain point should not be your only point of comparison with other actors.
Conclusion
When choosing an AI provider, first make sure that you understand what you can expect and that you will be able to acculturate your colleagues about the usage of this kind of technology. Then you can start benchmarking the provider by checking a few details that might help you find the actors with state-of-the-art technology by looking at their teams, partnerships and validated business cases. When you tick all of the above boxes, you can launch a test in real condition and make up your mind with all of the elements.
The business models or deployment roadmaps are important aspects but not as essential as the core technology, the artificial intelligence model potential, the forward-looking vision and the fit with the teams whom you’ll collaborate with during weeks or months and even years.
