Virtual assistants: FlamingoAI

Dr Catriona Wallace is the CEO and founder of Flamingo AI which is a listed company that has been around for a few years. It's still burning cash but it’s just about to really start stepping up the sales effort. 

In fact the other day, the shares jumped 13%, well it was actually just from 4.4 to 5 cents per share but still 13%, up to a market cap of $45 million after they did a deal with a US insurance distribution business called EXL. 

I think it’s a very interesting artificial intelligence play. What they do is they make artificial intelligence or machine learning virtual assistants to help companies deal with their customers online to take them through sales processes. It looks like it’s a very interesting proposition, I think it’s worth a look, certainly worth listening to or reading the interview. 

Here’s Dr Catriona Wallace, the CEO and founder of Flamingo AI.

Catriona, can you give us a bit of the background on Flamingo, when did it start and when did it become a listed business?

I founded the company in late 2014 and took the business to the US where we became a Delaware company, so we did a flip up, became a Delaware company, put the head office in New York but kept the data scientists and developers in Sydney.  We then decided that it would actually be a better strategy to build a global artificial intelligence business out of Australia rather than out of the US and so brought the business back in November 2016 and listed it at that time and so we have been in the capital markets just under 18 months now.

And just explain the product, how it works.

We’re in the artificial intelligence field, we’re a machine learning company.  Our point of differentiation is that we do unsupervised machine learning and this manifests as virtual assistants that we sell to large financial services companies in the US and in Australia.  Think of it like a chat bot on steroids, our customers or our clients use our virtual assistants to guide customers through their financial services product purchase or customer service experience.

Right, but do you sell it once off or a subscription and how much do you sell it for?

Yeah, so we deploy the virtual assistants, it’s typically around $100,000 to do an installation and set the virtual assistant up and get the machine learning brand trained.  Then we do a monthly subscription model and on average that sits around $300,000 per annum for one virtual assistant who is guiding customers through one particular process.  So, life insurance quotation, that may sit around a $300,000 a year, an annual fee, but we charge that at monthly subscription rates.

And does that depend on the number of seats in the client or it depends on the size of the company?

Right, not really.  The subscription is for one virtual assistant, so they pay a fee for one virtual assistant.  It’s easy to think of these as almost like employees, so they pay for one, this virtual assistant can handle 10,000 customer interactions during a month and we can charge a usage model, so usage might be number of conversations had by the virtual assistant, and we also offer a potential model around a revenue share model which would sit somewhere between 5% to 20% of the first year premium of whatever was sold, so quite flexible in the pricing model.  If these large clients get two, three or four virtual assistants then we can sell them an enterprise platform license, with them they can have multiple virtual assistants handling different customer interactions.

That’s very interesting.  So, how many did you say they can handle, how many customers they can handle, is it 10,000 or was it actually more than that?

Yeah, illustrative.  So, really the platform is built to scale globally around multiple clients, multiple interactions, so they can handle hundreds of thousands of interactions at a given time.

Right, so I’m just wondering where you’re at with the business model and growth now, I noticed in the last quarterly you made $200,000 so obviously it’s fairly early days still at this point.

We’re coming out of early stage revenue into a commercialisation strategy now.  So, the best way to think about it is we’ve built a unique product, all of the machine learning capability is our own built by ourselves, not built on the back of anyone else’s platform, so not on the back of Microsoft, not on the back of Amazon.  We’ve done that, we’ve tested, we have nine current clients who are all at various stages of implementation or go live.  We’ve seen very good product market fits.  We raised $15 million at the end December 2017 and we’re now moving into a commercialisation strategy.  The revenue to date has been really paid trial based revenue.  We do have one client nationwide whose monthly recurring revenue already and our focus now is putting on a sales team to start to take not only the business to scale but to really focus on the US market as our predominant market.

You’re likely to go further into cash burn.  I mean you’re going to now, as you scale up your sales force ahead of actual sales I guess you’re going to start burning cash at a big rate, aren’t you?

Not necessarily.  We have built this business on Lean Startup Principles, so that’s my background.  We’re actually very conscious of our burn.  So, currently the runway that we’ve got even regardless of revenue takes us well into 2019 so we’re very well-funded to achieve the milestones that we need to do and at this point in time not necessarily needing to raise any further capital based on the revenue targets and how we are doing commercialisation strategy, so very confident at this stage that we have a solid runway, incredible amount of interest in the sales pipeline.  We’re going from having no sales team to around five sales people across the US and Australia and are really focussed on conversion to monthly recurring revenue and new sales opportunities.

Talk to me a bit more about the nine – you say you’ve got nine customers at the moment, they’re basically testing the product, are they, for you?  They’re not actual paying customers, you’ve got one paying customer.  The nine who are using it, what’s the feedback you’re getting?

Yeah, so the feedback is very good.  The nine clients are all paid clients so we’re in formal contracted relationships with them in a software service licence where we do what’s called a production pilot.  So, a production pilot is a fully integrated virtual assistant that they run in a trial mode for a couple of months and if that’s then successful then we’ll look at moving them into the monthly subscription model.  It’s not an unpaid proof of concept, these are all paid engagements setting the platform up to then go into a further contractual relationship.

I’ll just get a sense of the market now.  In your recent presentation you talked about the $17.7 billion addressable market just in the US.  Paint us a picture of what that looks like.  How many businesses, are they all potential customers?

Yeah, so the addressable market is huge.  We’ve purposely focussed on the insurance sector because it’s highly commoditised and needing to increase their online sales revenues by 10 times in the next five years and it also needs to strip out costs.  We’re very focussed on the insurance vertical but in addition to that we’re now starting to look at banking and we’ve just done our first exploration into the telecommunications sector.  In itself in the US there are 6,000 insurance companies of some type, either top tier, we focus on Fortune 100 companies but second, third and fourth tier companies, 1,000 those companies that potentially could be our market.  Having said that banking sector is also entirely appropriate for us really as would be the telecommunications sector.  So, we’re looking for where is there higher customer interactions for sales or service which really means that these markets are huge for us not just in the US, also in the European market eventually which will probably be our next market after APAC and the US.

These nine customers you’ve got now, are they insurance companies?

They are predominantly insurance companies, we have one telecommunications company that we’re just doing early exploration in as well.

What exactly does your virtual assistant do for them?  Just run us through what the sort of day to day tasks that they’re performing?

Yeah, so I’ll give you an example.  One of our large US clients has our virtual assistant to guide customers through their auto insurance, so if a customer gets a quotation online then typically 95% of them will abandon or go elsewhere.  Our virtual assistant would appear at the time a customer gets a quotation and then guides them through their configuration of their quotation.  The virtual assistant, we call her Rosie, Rosie can guide customers to considering up to six cars, six drivers, can do all the vehicle checks, can issue the binding policy and take payment to do the binding policy for the auto insurance, so very sophisticated in what she can do.  From an end customer perspective it means that they can have a conversation with the virtual assistant can guide them through their full transaction so that they can complete what they set out to do.  So, that’s an example of the sales.

Is Rosie talking to them?  This is a voice activated thing, is it?

The first iteration that we do is digital, so this is all web based.  It’ll be a text chat so a customer would be doing it off their mobile phone or desktop or tablet and Rosie would text chat back with them.  However, to convert that into a voice application is the next step for us which is actually a very simple process.  We’re doing it because our clients want this to be a digital interaction but another way to do it would be using the voice interaction.

But it’s not doing that yet?

Not doing it yet.  We don’t have demand for that yet, it’s really that’s going to be next on our road map just reflective of where the market maturity is at.

Why does it need to be machine learning?  I’m just wondering because I mean I would have thought that these steps were pretty standard, they were always the same, what does the machine need to learn or am I missing something?

Well, possibly.  I’ll give you a quick example.  We did an installation for a large US insurer which was life insurance quotation and on day one – so, think about Rosie as a really smart employee.  On day one she will turn up and she’s very good at learning but she knows nothing about this particular insurer’s life insurance quotation. We can pre-feed her, so we work with another ASX listed company called Appen who is a machine learning training company.  We get Appen to go and crowd source all the questions and answers that an American consumer might ask in a life insurance quotation journey and we get Rosie to ingest that, she just sucks that up into her brain.  On day one typically then she’ll be able to answer maybe 25% of questions that a customer will have around their life insurance quotation.  By week two in an unsupervised learning environment she was able to answer 70% of all questions customers had.  By week seven it’s 90% plus and she was converting over 30% of customers who came to get a quotation, were converted into going on to do an application.  What she learns, Alan, is all the questions that customers will have to try and then complete their transaction online. 

If you compare this to a web form a web form there is no way that a customer can ask questions, reconfigure things, they just have to go through a static process.  So, there is the value.

Right, and so that’s Rosie, you’ve also got Maggie.  Is she different?

Yeah, so think about Rosie as narrow and deep, Rosie is going to know everything about life insurance quotation but she’s not going to know anything more, so you can’t ask her about a different product line, you can’t ask her around dental insurance or home and contents insurance, she just knows how to close sales for life insurance in this one particular company.  Of course, in another company she will know how to do auto insurance binding policy.  Think of Rosie as the closer, she’s narrow and deep.  Maggie is broad, so Maggie is more like a frequently asked questions bot where she is going to be able to answer many questions across all different product or inquiry types but not know the depth of product or customer transaction or journey that Rosie will know.  The way we put these two machines together is that Maggie is kind of the concierge and helps customers understand what they need to do and then Rosie is the closer.

You’ll sell them as a bundle, is that right, do you always sell them together or separately sometimes?

It can be separate or could go together.  Ideally they’d go together because Maggie could also plug into things like social media and potentially do lead generation for a company as well, so there’s a lot that we can do with Maggie as far as attracting customers to then buy from our client companies.  Really the best way to think about what we do is conversion optimisation using clever machine learning.

Right.  Just go back to the pricing for a minute, just to remind me it’s $100,000 to install and then $300,000 a year subscription I think you said.  Where did you get that pricing from, have you tested that that’s fine, that that’s an okay level of pricing?  Obviously you have.

We have, yeah.  As we do more installations and see the value that we’re creating we know that we’ll be able to move up on that pricing model.  Particularly we’re really encouraging our clients to use the usage or consumption based model so then we will be paid by the number of conversations and ideally for us as all good smart, young SaaS companies we would like to be in a position where we have a lot of our clients using the revenue share model.  It’d be a very low subscription rate and then we would take a variable percentage of what the platform generated.  We would like to move towards that and we’ll encourage our clients to move towards that which means we have to back ourselves that this product is very good and can sell on behalf of our customers and we will share in the upside of that.

You’ve just done a deal with a Nasdaq listed business called EXL Service Holdings, I gather that’s a go to market channel for you.  Is that a big deal?

Yeah, that’s a really big deal.  EXL Services is a 27,000 person, $2.4 billion market cap US based insure-tech company.  Their expertise is in data analytics.  They service eight of the top ten US insurers.  We are their chosen conversational AI partner organisation and so we’ll be doing joint go to market, they’ll be doing client introductions and ideally, they will also integrate our technology for us and also integrate it into their platform.  This is a significant partnership for Flamingo, so we’re very excited about it.

Right, how many did you say they’ve got, 27,000?

Employees, and over 800 clients.  Yeah, it’s great for a young Australian company to be travailing in that company.

I’m not sure it actually moved your share price, did it?

Yeah, about 13%.

13%, yeah well there you go, it did.  Tell us a bit about your own background.  What was the business you were in, you were a long time in a business called…

Fifth Quadrant.

Fifth Quadrant, yes, tell us about that.

Yeah, so that would be the last time I spoke to you would be Fifth Quadrant.

Yeah.

My background is in customer experience, design, human-centred design and market research around customer experience.  Very different to other AI companies which are really all around automation and efficiency.  I brought the customer experience meets machine learning into this business and this is a very different offering.  I’ve got a PhD in human technology interactions and I’ve been studying the role that computers play in replacing humans for some time and this business is kind of the converging of my areas of expertise.  We have, including myself, five PhDs in the Flamingo business, all experts in machine learning and natural language processing.  Based on my background I’ve been able to pull together the very best people in the world to now build this business.

Maybe you are a computer programmer or coder, how did you go about building it?  Obviously, you had the idea, what was involved in that?

Yeah, so I am not a coder, so I understand technology and I am very good at modelling out the role that technology plays, I’m not a coder.  The two key team members that I have in the business, one is Joe Waller.  Joe Waller is the ex-CTO of Bet Fair, so Bet Fair now being the world’s largest e-gaming company.  Joe was the second ever engineer at Bet Fair and scaled their exchange globally, so an incredibly talented technologist who did his time at Bet Fair then wanted to look at the next big disruptive thing and decided that was going to be Flamingo.  Then our Chief Data Scientist, Dr Jack Elliot.  Jack is most famous for using his PhD work in algorithms for discovering new works of Shakespeare, so he has an international reputation around that.  Jack’s area of expertise is in text, conversation and literary critique using algorithms and machine learning.  Jack is really the brain in the brain, is Jack.  Then we have Dr Han Xu who was previously in the Macquarie Bank team building their robo-adviser.  Han is also a PhD in machine learning and natural language processing.  We have Dr Yasaman Motazedi, again a PhD in the same area.  Dr Nick Marriot, PhD in the product side of it.

What I have done is been able to pull together some of the best brains in the world but also bring a very strong commercial acumen to it.  A recent hire for us on the board and as Chief Commercial Officer is Mark Keogh, Mark is previous co-founder of Grays Online, has been able to scale that business while he was there to now I think it’s a half billion dollar sales revenue.  Really, I think the secret to what we’ve done, Alan, is incredibly smart people in the field of machine learning, meet customer experience expertise and meet commercial orientation.  Those are the three things that I think have driven our success to date but really will drive it to be one of the top machine learning companies in the world.  

The place is bristling with PhDs, goodness.

Indeed.

And are there new Shakespeare works?  Goodness, I had no idea.

Yes, there is.  It’s easy to find this online.  I don’t recall the name of it but it was previously unknown work suspected to be Shakespeare but not proven.  So, Jack ran his algorithm, ran the work through his algorithms and it proved beyond any reasonable doubt that this would be a work of Shakespeare based on the language conventions and some of the other patterns in the data.  It’s now accepted by the powers that be in the UK literary world as being a work of Shakespeare but credited to Jack for finding it and proving it.

There you go.  Well, that’s a bit of a divergence.  Just to finish up it sounds like you’re on the threshold now of making some serious sales, would that be a fair comment? 

Yes, that’s right.  We really are now focussed on conversion to MRR and new sales opportunities, and that’s where all of our effort and the funding is going but we do that in a considered way.  I’m very conscious me as a CEO my role is to keep money in the bank, so to raise it, generate revenue or not spend it and that’s the approach that we have.  But, the opportunity is huge, Alan, so the inbound inquiry is almost unmanageable for us at the moment.  We’re doing some very clever go to market modelling around exactly the target market and how we will approach and segment the market, so we’re very focussed in order to get to revenue as soon as we can.

Do you have an exit plan?

At this stage our entire intention is to build this, I’d like this to be between years five and ten a billion dollar market cap Australian business.  I think that’s exactly the plan that it should be.  We’ve got no other plans around exiting to an acquisition, in fact we’ve had a number of companies approach for acquisition that we have not proceeded with.  We want to build this as an outstanding Australian machine learning artificial intelligence company, that’s the plan we’ve got.

Can you tell us what sized offers you’ve knocked back?

I cannot tell you but it was just too early for us.  Both were American companies, we have no interest in selling until we’ve proven out what this business needs to be and even then we think that we can realise the vision and the value for our clients, our shareholders, our staff by doing this ourselves.

You must be tempted to list it on Nasdaq.

I get asked all the time about that.  I think for a young high tech company it’s enough to be running this in an Australian listed environment let alone a Nasdaq.  Down the track it may be a sensible thing to do but no plans at this stage.

That was Dr Catriona Wallace, the CEO and founder of Flamingo AI.