Even before generative AI, Aussie startup Pendula was doing some amazing things with the way it could upsell your phone plan, your insurance policy or even your NDIS budget. Now it’s making a reported play to buy fallen pandemic ASX darling Whispir.
If you were paying close attention, maybe you could notice the power company only ever sent you texts trying to sell you solar panels on a sunny day. But you’d have to be really sharp to notice that – if they were using Pendula – your insurance company would not send a big upsell text on a day when there was heavy traffic jams in your neighbourhood – since it knew you might be stressed … and it didn’t arrive at the same time as your monthly home-loan repayment or rent was falling due when you might be feeling financial stress.
These are things Pendula was doing before generative AI, says founder and CEO Alex Colvin – using machine learning for clients that ranged from Origin Energy and AGL, to phone companies Amaysim, Coles mobile, Catch Connect, MATE and Kogan.
Pendula has raised more than $24 million since it was founded by Colvin in 2017. It announced in August it had raised $14.5 million through UK investor Octopus Ventures. It said the new money was going towards expansions into Europe via the UK and Asia via Singapore.
But it appears some of it is going towards local expansion.
Whispir had developed platforms for texting large numbers of customers. It listed on the ASX in 2019 in time to ride the COVID-19 wave of crisis communications by large companies and government departments. It’s share price rose from 90c at the beginning of the pandemic to almost $5 by July 2020. Since then it’s been largely downhill.
Whispir’s ASX Journey
It laid off 30% of its staff in November 2022. Whispir sank as low as 23c a share earlier in the year. Communications software company Soprano Design made a 48c-a-share offer for the company last month. On Tuesday, Whispir’s board recommended shareholders reject the offer and told them that Pendula was poised to make an offer. The share price was 52c at the time of writing.
While Whispir’s momentum has been all in the wrong direction, Pendula appears to be swinging the right way.
Alex Colvin explained to Forbes Australia that his company was thriving because it had the ability to take in information from outside the platform – like the weather, or the traffic or when you signed your lease – which made it more agile than the competition. And now AI was making it even more nimble.
When generative AI burst onto the scene in early 2023, quickly followed by reports of inexplicably aberrant answers to certain questions, two realisations leapt out at him and the Pendula team.
“One is that no one’s going to trust this thing to start generating content that would actually be sent to their customers, even if it could be trusted and even if the hallucinations aren’t that frequent,” says Colvin. “You can just imagine if you’re a telco, and you’ve got one and a half million customers and you’re generating 20 million, 30 million messages every year. That’s a lot of content that could go very wrong, very quickly.”
The second realisation was that they could still use AI to solve real-world problems without the need for complex code to be written.
He cites the example of when a telco customer does not click “auto-renew” on their prepaid mobile plan. “When that happens, normally you trigger, and you just automatically give someone the same offer. We’ll use other machine learning techniques to basically try and figure out how cheap a discount I can give you based on your own behaviour. That was machine learning. What GenAI gave us was the capability to do that individually. And we manually wrote this prompt, and it still works today, and it’s deployed over five customers now written in plain English: ‘Go out to the internet, search these competitors.
Tell me what the cheapest plan in market is, insert market, insert plan of this current customer. Tell me how much of a discount I’ll have to give this customer in order to be the cheapest in market. Then, based on this customer’s usage over the last three months – how many calls they make, how much data they use, and how many times they’ve called the call centre – calculate the yield that I would have based on these cost aspects. 0.01 cents per megabyte, $3 per call to the call centre. Then if the customer’s yield is greater than whatever the telco’s yield margin is, add 30%, then make the offer.’
“Now the cool thing about this is, you can pump a customer base of 14 million customers through that.
“The third thing that we can do is we can solve for all these hallucinations because the output of it is going to be predictable. It’s going to get the same answer every single time.”Alex Colvin, Pendula CEO
“We run it five times. As long as the answer is the same every time, we know this is an accurate representation. As a result of that, we can trust that the GenAI has been effective.
“The thing that unlocked for us was basically – as long as you can explain it in English, Pendula can now go and do it. And that was a huge unlock for us.”
The way it was being used in the National Disability Insurance Scheme (NDIS) was “a better use case from a moral perspective”, he said. “We are upselling participants on the NDIS schemes, but we’re not doing it in a way to extract more revenue from them.” NDIS participants would be allocated funding for their care but they tended to hoard it, Colvin said. “They get 50% of the way through the year, and they’ve only used 10% of their allocation because they’re too worried they’re going to run out.
“So using that same process, we say ‘what about some physio? What about some respite care?'”
Colvin gave another example that might have relevance in the context of last month’s Optus outage where an energy company cross-selling NBN services could look for where there had been NBN outages and know that its customers in those areas would be open to offers. Pendula was doing that before, but it required complicated coding. Now it’s done in plain English.
“It’s probably been the single biggest leap that we’ve had in our platform.” He was confident telling his competitors what they were doing, he said, because he was confident the competitors’ platforms would not be similarly malleable.
Colvin said they had recently landed a UK customer worth $250,000 annually but declined to disclose their name.
Pendula was poised to sign a significant “white label” deal, he said, whereby Pendula’s software would drive a large UK company’s customer engagement anonymously.
“It will push us back into triple-digit year-on-year growth. It’ll push us into profitability as well, instead of just cash-flow positivity, which is a big milestone for a business at our stage.”
He declined to talk about the Whispir negotiations.