In our work with our clients we always see them struggle with this question – should they hire and develop data science teams to build AI solutions themselves, or should they work with AI vendors and buy those solutions? Over eight years, we have seen people succeed on both sides, and them making colossal mistakes on either side.

Here is that experience, synthesized in five bullet points.

1. Do you really have a vendor whose solution works?

Even before thinking about this decision, ensure whether you really have a choice in the matter. Is there a vendor whose solution fits your bill? Don’t go by how much money they have raised, instead focus on who else have they solved this problem for. Be very careful about the scale. Few AI solutions scale well.

2. Which option is cheaper, really?

We are often surprised at how surprised our customers are regarding our pricing. “We can hire an engineer for that amount and do it ourselves,” is, unfortunately, a common thought we have to deal with in our process.

The truth is a vendor may be cheaper by a huge margin. It appears expensive only because you do not fully appreciate all the costs involved in an AI projects – the dollars, the time and the political capital.

You can read more about costs of AI systems here: Seven Myths About the True Costs of AI Systems.

3. What about Security and Determinism? 

Normally vendor software, esp. SaaS, comes with data security risks. With AI there are serious concerns about determinism as well. We have spoken about this issue before.

You can access the presentation here: Why Silicon Valley does not understand Enterprise AI? This is the reason many vendors, including Coseer deploy in a single-tenant format behind the firewall.

Not only is the data completely secure, no update is released without proper tests and controls. You can learn more about our Keep-It-Safe approach, and hold your other vendors to the same standards to be assured.

4. Will your AI solution eventually work?

Consider a vendor like Coseer, which has spent eight years developing it’s path breaking algos. We admit unashamedly that we have made many many mistakes in this journey and have learned from each one.

When you start building an AI solution you are likely to make many of those mistakes, and many others specific to your situation. With a well-intentioned team you are very likely to make the solution work, but each of these mistakes will add to the dollar, time and political costs of your solution.

5. What level of customization do you need?

The need for customization is the biggest argument for building solutions rather than buying them. If the need for customization is too large for your vendors to accommodate, you may consider partnering with them to license their technology rather than their product.

Many vendors, including Coseer, offer a complete suite of APIs that integrate with your own IT systems.

Either case, this is something you need to think through carefully. Good luck!