How to buy Artificial Intelligence

So you’ve done your groundwork with AI. You’ve looked at the technologies. You’ve talked to the vendors. You’ve looked at your business to start seeing where the applications of the solutions could fit and could benefit.

Big question now is, how do you deploy artificial intelligence? What skills do you need? How much is it going to cost you? What vendors should you be talking to?

Here I’m going to look at the four different ways in which you can deploy AI into your business and give you some of the pros and cons of each.

1. Embedded

This means talking to your existing providers, whether it’s people who provide CRM systems, or workforce planning systems, or any other operational platform in your business.

They will have a programme around artificial intelligence, which I’m sure they’re happy to talk to you about.

The challenge is, you don’t know what functionality you’re going to get or when.

And you have to question their AI credentials.

Are they experts in this space? Are they going to deliver something which really will work well or it’s just about putting a badge on the platform?

You’ll probably pay a premium for it too and it’ll be a licenced product, which you’ll probably have to pay for for three years.

It’s low risk, it’s safe and it has some advantages, but does it really deliver the sort of functionality which you want for the future?

2.  Off the shelf

These are really point solutions developed by many startup businesses who have taken the machine learning idea and developed a product around it that tackles a particular challenge in your business.

Whether it’s understanding customer behaviour, understanding text and sentiment and other things like that, or predicting some next best activity. All sorts of different things.

These solutions mean that you don’t need to have too much more skill. You need to have perhaps a little bit more awareness of what data you’ve got and how to access that data so that you can feed that data into those models and work with the providers – but they will provide the hardcore expertise about how to get the best out of their model, assuming their model is good.

You can work quite fast with these providers. They tend to be small and fairly agile. You’ll soon see, quite quickly, what you’re going to get in terms of an output and if their model works for your data, the performance can be very good.

But, until you’ve tried it, you probably don’t know whether that’s the case or not.

The challenge that these guys have is that they probably haven’t got the sector experience, so they’re very broad in their approach to provide a solution that will work across many different sectors.

They don’t necessarily understand exactly what your business problem is or how to get the most out of it. So, for you, understanding contact centres, understanding customer experience, that may not be their forte, and therefore, that’s where you need to provide the expertise to guide them, to get the solution tuned up for you.

3. General Purpose

This is using a lot of the Cloud based tools from the likes of Amazon, Google, Microsoft, Watson and others.

They are complete, broad, generic solutions that allow you to work with text, images and lots of other machine learning models to get started very quickly.

They don’t cost very much. They have lots of training information online but, despite the training, it’s more training on how to use the tools then training in machine learning.

You need to have a certain level of expertise, inevitably, to know how to use these tools properly. They’re great as a test and learn environment.

They’re all pay per use, for most of their tools, which is great when you’re small but you’ve got to be careful when you start to scale up because those costs can mount.

Again, talk to the vendors if you starting to go live scale and I’m sure there’s deals to be had. They do provide very robust platforms at scale and they’re certainly worth looking at.

The challenge for you then is access to the right skills in order for you to be able to build these models in the first place.

4. Bespoke

This is when you’ve got a problem which really is quite unique to you or you want to keep the intellectual property around the solution in-house.

You’ll bring in a data science company who are used to building the bespoke models to understand your sector, understand what you’re trying to achieve, understand your data, and work with you very closely to actually create something which is unique to you.

Clearly you’re going to get high performance – though you’re going to need to put more effort into that and it’s going to cost you some effort and time in terms of capex as well.

It’s a great solution if you’ve got something that’s really special but probably, for most people starting out, try the other approaches first, see where that gets you, if it doesn’t work as well, then look to the bespoke approach.

Those are the four areas. If you want some more help on any of those or you just want to discuss them, please give me a shout.