How do you train and maintain a bot to get the most out of it? Former iProperty CEO and current chairman of iCar Asia Georg Chmiel gives us his tips.
Online classifieds executive Georg Chmiel has extensive knowledge about bots. Chmiel was the managing director and CEO of Asia’s iProperty Group and was instrumental in developing its chatbot Rebecca, the first of its kind within the Asian real estate sector back in 2016.
Currently, Chmiel is the chairman of iCar Asia Limited, an automotive classifieds portal which has been perfecting a new bot set to pioneer the industry in Asia.
Bots are there to support existing processes, it’s really that simple, Chmiel says.
Although there’s been debate concerning the relevance of bots in the future, Chmiel says there are already key industries - such as the medical and legal professions - that are using bots to transform how they serve customers.
“These professions are simply prime for bots – it’s anything to do with making decisions based on data,” he says.
“To give you an example, there’s a bot called ‘DoNotPay’ that provides legal advice, and that bot helped to overturn 160,000 parking fines.”
The complexities of bots can often stump those who are operating them, but Chmiel has 5 key pieces of advice he offers to those who want to operate and run a bot successfully.
“As always, when introducing new technology, start with simple questions and tasks; we’re not talking now about quantum computing yet” Chmiel says.
In the instance of the iCarAsia bot, he says one of the prime objectives of the bot was to make an existing process easier. Previously, potential customers had to look for a phone number and then send an SMS or call a car dealer if they wanted to book a test drive. Chmiel says this process was “inconvenient” for both the dealer and the agent.
“Disruption is always about making things easier,” he says. “The bot should be able to do simple tasks first, and that’s important.”
Chmiel says if bots are unattended they can be taught silly things, something he experienced first-hand while witnessing iProperty’s test bot Rebecca in action. In his opinion, one of the keys to having a successful bot is ensuring that it’s consistently monitored. He cites Microsoft’s bot Tay (released in 2016) as a cardinal example of what can happen when bots are left un-monitored. In the following example, we see Tay the Twitter bot turn into a racist after engaging in dialogue with followers.
Chmiel says repetitive tasks are prime for bots because they don’t sleep and are operational 24/7.
He says humans will often forget something, but a bot does not, and therefore is perfect for helping an intermediary to be more available more often.
The key thing is to really individualise the bot,” Chmiel says. “It must feel like a real person on the other side."
In the instance of IProperty’s bot Rebecca, Chmiel says he was often unable to tell if he was talking with a bot or not.
“My initial suspicion with Rebecca was that there was someone sitting in an office and responding because the responses were too good. This was also true for the iCar bot, at the beginning.”
The iCar Asia bot takes six months to become fully operational and to learn a language cohesively – much faster than the speed of a human. Chmiel says one of the secrets to its success is that it’s constantly being trained, and by the right people. Putting the right people in front of the bot, is a very important lesson.
“You wouldn’t put someone who is completely untrained in front of a person, so you wouldn’t do this with a bot either,” he says.
Although we’re yet to reach perfection within the bot world, Chmiel says we’re making steady progress, but the operation is a constant evolution.
“The key is really simple: invest in training of the bot and be clear how it fits into existing processes which people are using; do not create something new. Support existing simple processes. That’s what disruption is about, making things simpler.”