New words of wisdom from the AIIA Network Podcast
Future proofing, the cognitive inflexion point, change management first, being creative in structuring data and launching new customer experiences
As many of our readers are aware, the AI & Intelligent Automation
a new podcast episode
Here, we highlight top intelligent automation insights from episodes 16-20.
Our aim is to provide a voice for the business discipline of intelligent automation. Through personal conversation, we’ve already had the pleasure of interacting with some incredible practitioners in this field.
Episodes 16-20 feature case study anecdotes from Nordea, LSE, BNYMellon & Prudential. Within each of these conversations, we’ve garnered dozens of key learning points from each executive’s personal career.
Episode 16: Mattias Fras, Nordea
Launching new customer experiences
“Using the robot as the bridge between the app and the backend legacy system allows you to launch new customer experiences.”
Mattias shared this thought while explaining the bank has deployed bots to pre-screen customer data, searching the internet and filling in additional information about customers such as organizational participation, social activity and political activity.
Building the app to secure this information can be done in just a few weeks but the integration from the app to the back-end system is traditionally a comprehensive and burdensome process. Enter bot. Mattias and his team then simply need to imagine a new customer experience and they can launch it within a few weeks.
Episode 17: Julie Lotzer
Change management first
“Right now we're just focused on, "What in the business takes a long time for somebody to do? Is it a candidate for RPA?"
Julie works for a large-scale automotive manufacturer. At the time of the interview, she was focused on simply getting the organization to buy in to intelligent automation.
She’s aware of other approaches: “I know a lot of companies are doing due diligence with mining their processes to see what’s the best fit for RPA. We're actually taking any idea and ask, is it a candidate for RPA? We will probably go back and do those mining type activities, but right now we're just focused on, what in the business takes a long time for somebody to do? Is it a candidate for RPA?"
And so it was change management first.
Episode 18: Professor Leslie Wilcox, London School of Economics
The cognitive inflexion point
“In RPA, you configure the robot to do idiot work so that you can focus people on human work. AI is about moving into cognitive areas of work—judgment, reasoning, understanding context, and deriving meaning. This is much more difficult thing to replicate.”
And so, to the inflexion point we go with Professor Wilcox…
He notes that it’s difficult to ‘robotize’ the human in the above respects but that taking data from RPA and using it to do analytics and provide insight is a good use of cognitive solutions.
But we’re nowhere near 90 per cent of the industry engaging in cognitive solutions. Based on his experience, he shares, “it takes 8 to 28 years for something to optimize 90 per cent of the use.”
So, if you’re not employing cognitive solutions you’re still not behind.
But tick tock my friend.
Episode 19: Jon Theurekauf
“The decisions that we make, those that will change things in our workforce, that change things in our process structure, that change things in our technological structures, all those decisions that will have a future and allow the bank for the future to be more future proof.”
Future proof? Jon of course means transformation. The work being done now in intelligent automation is the digital strategy you crave, it’s the transformation strategy you need and it’s the redefinition of what your organization is based on the fact that you need to redefine who you are today to be alive tomorrow.
Change the people, change the process, change the technology and keep the business.
Episode 20: Tobias Unger, Siemens
Being creative in structuring data
“To be successful in AI, you need a stakeholder who believes you can actually help and has budget to deploy. You need labeled data to predict certain events. If you can, get your users to label your data.”
You need as much structured data as possible right now. Take Facebook’s lead, “you get them to put photos online and then you say this is my friend Seth with me having a Martini, here we go again, at the beach. You have a whole bunch of labels so you know that there's a person there. You know that there's a beach in the background. You even give a name to the person, so it's not just face detection. You can start to think about face recognition and that's beautifully labeled data.”
That’s all well and good, but Michael urges that you should look at the ‘the next step down’ too: “you need to find a way to label that data. You need to find a way to go through, build some training datasets and then build models based on those training tests. Another way to go about this—again, they are not mutually exclusive—is to go and try and find a way to encourage your users to do the labeling. Then you're back into the sort of Amazon and Facebook type approach.”
When I point out that not everyone runs a social media or ‘retail’ company, Michael shares, “so then the question doesn't just become so how do you get closer to the customer, but also how do you get closer to your intermediary? How can you give the right kind of recommendations for next best actions or whatever it is to your intermediary and not just sell directly to your customer? So there's nothing that changes fundamentally. It makes things a little bit more complex. It will probably throw up a whole bunch of data availability issues, which you will always have in an enterprise world.”
Yes, it’s problem after problem. But Michael notes it’s all about finding one small solution for your data, which should lead to one giant step solution for your company down the line.
“Our data will always be unavailable, in silos, by business unit, by geography. It will be dirty, out of date. It will be incomplete. We will always have these problems. I think the key is to find ways to overcome these problems and some of the data labeling tricks that we just spoke about are one element of that.”