The AA deploys an internal AI agent themselves using Databricks AI/BI

I read a fascinating post from Matt Sanderson this afternoon on LinkedIn. Matt is Head of Data Products for Channels (Digital & Customer Operations) at The AA ("the Automobile Association").

Only a year ago – or perhaps just a few months ago, you'd have had to phone one of the Conversational AI vendors, drop at least a few hundred thousand if not more, and get them to do this for you.

Today though?

Well, Matt and his team spent 2 months with members of their AI Centre of Excellence to make a chatbot to provide insights and answers on their Roadside digital performance.

How did they do this? Matt explains:

We utilised Databricks AI/BI Genie offering and integrated the solution into Microsoft Teams. This provides colleagues with a natural language interface to get the data they need instantly, at any time of day or night — ensuring a seamless user experience.

The chatbot is accessed via conversational interface in Microsoft Teams - so very familiar to all staff.

They've had great feedback already. Here's a sample:

My favourite comment so far came from someone in our trading team, who told me that the bot instantly answered a question that would have taken him at least 30 minutes to query himself. Win! Imagine the efficiency and productivity improvements we could realise if we scale this solution wider!

Absolutely brilliant. This is a fantastic example of what you can do with the technology.

If you've got any questions, you can find Matt on LinkedIn and his post is here.

Build or Buy?

As I alluded to above, a little while ago, Matt and team would have had no option but to out to market and buy this service from a Conversational AI vendor.

The rapid advancements in technology mean that he and his team were able to do this themselves. Yes, Databricks isn't free, but implementations such as these are really blurring the lines between what should be purchased vs what should be built internally.

Having been the guy on the inside of a large company having to content with dozens (if not hundreds) of 'craft workshop' teams all working on supporting all manner of semi reliable software implementations, I am always heavily biased toward buy. For example, I've been the guy spending a LOT of time and energy ripping out 15-year old Excel spreadsheets or Access databases that were taking so much time, cost and energy to maintain internally. Buy from a vendor and you get a service level. Buy, and you get a roadmap, new features and so on.

Build and you can end up with a lot of pain down the road.

The one challenge that I'm sure many vendors will point out is that Matt and team can quickly end up with a really compelling product ... that might not necessarily come with the service levels and support that it could well need in future.

The nightmare scenario for The AA CIO is Matt's chatbot becoming business critical over the next few months, but no one really noticing this... and then suddenly it stops working – at 2pm on Saturday, over a bank holiday weekend. Who gets the call? Or, more importantly, who picks up? Is anyone looking at their phone until Tuesday? Who's responsible? Is it the person who's maintaining the Databricks implementation, or the team who 'wrote' the bot? These are all issues to look at in due course.

The key here is that in 2 months – lightspeed in any sizeable organisation – this capability was brought into existence and is adding real value, right-away.

This should be deeply, deeply concerning to read if you're a Conversational AI vendor. See my opinion post, written live in McDonalds on Tuesday evening, for some more perspective on that.

Really, really good work Team AA! Nice work Matt!