AI will hurt banking without a ground-up approach

Injured Piggy Bank WIth Crutches

Image Source

The last twelve months have seen AI fever reach new peaks. Every technology giant we could name have got a huge AI story to tell. Starting with IBM Watson, Google Allo, Facebook AI Research, Amazon Echo have all had major traction. This is also reflected in investments being made into the AI industry. Industry leaders magazine predict 2017 would be the year of AI investments. It is predicted that about $37 Billion would be invested into AI by 2025. There have been similar “This time it is different” stories on AI for almost 50 year. Well, I believe it is definitely different this time, mainly because AI was preceded by the Big Data revolution this time. I believe it is different for firms and industries with access to quality data to leverage for AI. I believe that the technology firms mentioned above have got their data right. And I believe banks are a joke when it comes to quality data. AI will hurt banks without quality data.


Internal legacy tech challenges and data quality problems aside, new regulation in the form of PSD2 is going to have massive impact on a bank’s strategy to customer facing AI use cases. With PSD2, customers will be more in control of their transaction data. Third party providers (TPP) will now have access to customers’ transaction data, in the process becoming an abstraction layer between banks and customers. So, unless banks completely rethink their customer interfacing model, nimble players who can create clever AI applications on customer’s financial information, will make banks just a utility provider, hence hurting their margins.

Banks have two challenges to resolve at the same time, internal data quality issues to make AI work for operational intelligence, and external data ownership issues to make AI work for dealing with customers.race_for_ai_new_1-featured.png

Image Source

Data is the 21st Century Oil

There have been some success stories with Fintech firms doing AI and there are new ones emerging every day. The world is moving to a place where there will be an AI app (and an app store) for most activities. When we were evaluating IBM Watson at PwC, one phrase the IBM team used was “Watson going to school”. This effectively meant training IBM Watson on a particular topic. Let’s assume IBM Watson had to be trained in MIFID 2 regulations, it involved loading the regulation text into Watson and then an SME spending a few weeks with Watson asking questions on MIFID 2. The SME would then provide feedback on Watson’s answers, and Watson would use this data to provide better answers next time. This needed to be done until Watson became an expert in MIFID 2 before the capability could be launched commercially. We also learnt from the exercise, how understanding legal/regulatory language was different from natural language. That exercise showed how critical data, data structures and the taxonomy of data was for AI applications to work. On that note, news is that IBM just acquired a Promontory Financial Group to help improve Watson capabilities for banks’ back office functions.

'And then Dmitri noticed something that would have a profound effect on the human/robot wars.'

Regulations for data have failed?

Most successful AI platforms have access to high quality data, and in huge volumes. They also get to see regular transactions across different streams of data that they can then learn from. This is the case with Fintech firms that use AI at the heart of their proposition. What about our dear banks? Banks do have data, but the quality, integrity and accuracy of data stored digitally is generally appalling. A few years ago BCBS 239 emerged as a regulation focused on fixing data in banks, however compliance to that is mostly being treated as a check box exercise costing the banks millions. The point being, banks are years away from processes and infrastructure that provides quality data. If AI is introduced into this landscape, it would be more detrimental to existing processes, as there would be more hands involved in confirming results suggested by AI, and the costs of using AI would outweigh its benefits. Is there hope?

Banks need to get back to the drawing board to make the most of AI. Here is a simple approach I can think of,

  1. Banks need to have a function to deliver Intelligence capability.
  2. This function needs its own budgets and operating model.
  3. Very similar to data governance models, the Intelligence organisation needs to be federated across the firm.
  4. This federated model needs to cover parts of the bank that benefits most from AI, but also where data and processes are more matured
  5. Implementation of AI in these parts of the bank could spark viral uptakes across the firm
  6. There needs to be standardised ways for AI applications within banks to interface with each other. Without this, there will be AI applications developed across the bank for every single process and there wouldn’t be any integration possible. This is where governance will help.

Well, in order for the above to work, data still needs to be of good quality. Top down models are not always lean in their approach. However, it is possible to achieve a top down model that can be lean, if the priorities are based on benefit realization rather than empire building to get to an MD promotion.

Now, what about external challenges?

PSD2 could be a great opportunity for banks, and of course a huge challenge too. Let’s focus on the opportunity first. Say a customer banks with Barclays for his salaried account, has a mortgage from Halifax, and a credit card from HSBC. If Barclays was willing to be an Account Information Service Provider (AISP), it can effectively source the customer’s transaction (with their permission) from Halifax and HSBC and could offer Personal Finance Management (PFM) services. Imagine having access to transaction data across all products that your customers have. Barclays with access to mortgage transactions from Halifax could create a credit product for the customer for home renovation. While this is just one example, PSD2 could create new revenues streams for banks if they were willing to target other points in the payments value chain. That is a whole topic for discussion by itself though.

AISP in action

Image Source

While this sounds good from the banks’ perspective, the more likely outcome would be Fintech firms, and possibly tech giants (Google, IBM, Microsoft, Amazon) making clever use of customer transaction data as they are light years ahead of banks in their AI capabilities. While we have already discussed Tink in DailyFintech, Qapital and more recently Klarna are also jumping on the PSD2 bandwagon.

How can AI-Fintech firms chip into the story above? Well, they don’t have much to lose (unlike banks). AI firms focused on banking use cases, should focus on small problems and solve them really well. And the AI industry focused on Banking will need to identify the importance of standardisation of interfaces for data interactions across banking applications. While this exercise will be made easier for customer facing use cases (thanks to the PSD2 wave), operational AI internal to banks will be a much harder nut to crack. If achieved, it will allow for multiple processes and activities within banks to be replaced by AI in one go. I wouldn’t be surprised if banks, warmed up by Blockchain revolution, form consortiums to drive AI revolution. And if this happens, I believe AI penetration within banks could be faster and deeper than Blockchain managed.

Arunkumar Krishnakumar is a Fintech thought-leader and an investor. 

Get fresh daily insights from an amazing team of Fintech thought leaders around the world. Ride the Fintech wave by reading us daily in your email.

ChatBots in Consumer Banking


Image courtesy of SuitPossum

This is Day 5 of ChatBot Week on Daily Fintech. You can see the intro and index here.

How Much?

ChatBots are great when customers want answers to various “how much” type questions, such as how much….

  • Have I got left?
  • Will this cost me in interest/fees?
  • Will I earn/save by doing this?

Sure I could go online or check an App, but just texting those questions is easier.

Finding this innovation in the wild

MyKai from Kasisto launched on June 28. They must have had the back end ready and had the aha moment when they saw the launch of the Facebook Messenger App Store.

Bits don’t stop at borders but money has to show its passport. I wanted to try out Kasisto, but stopped at this Stop Sign (I am based in Switzerland):

“Right now, I am available in the U.S. If you live in another country, you can sign up and I’ll let you know when I’ve arrived.”

Do I have enough to…
“Eyeballing” physical cash in your leather wallet is how millions budget. We need an equivalent for the digital age.  MyKai shows shows how this can work by integrating with Venmo (check your balance before paying). Imagine that linked to your budgeting rules (Budget Nanny says no to that Latte).
Cousin of Siri

Kasisto which created MyKai originated from SRI International, the company that came out of Stanford University and created Siri.


MyKai anonymizes the data and does not have full access to your banking passwords. My take is “be careful” and this will be a big hurdle with most consumers, which brings us to the next point.

Public Proof Of Concept
MyKai is like a Public Proof Of Concept for a service that Kasisto sells to Banks. That is how they aim to make money. They claim Royal Bank of Canada and DBS Bank as early customers and Wells Fargo Startup Accelerator is an investor.
The future is already here….in Denmark

LunarWay, a Danish full stack startup bank (now tagged “neobank”, aka “challenger bank” in UK) offers LunarWay Bot today (in Beta with a few thousand users). It was developed in-house

…And in Africa

ABSA (Barclays Africa) offers ChatBots via both Twitter and Facebook.

Conclusion from innovation in the wild

ChatBot looks like a natural UI for a mobile first Bank. It will become an increasingly standard feature of both  startup and incumbent banks. Oh…and make tons of money for Facebook (showing us again how valuable being an ID provider is).

Messaging channels

Facebook is certainly the big one in America where Kasisto plays. When Kasisto hit a tech glitch they used SMS and Slack as fallback channels. Messaging itself is clearly a commodity. The money is at the intersection of ecommerce and payments and that is where Account Linking is the key innovation from Facebook (which we identified in the intro to ChatBot Week).

PSD2 and other utility regulation

Account Linking can be used a) by Banks to offer a better service or b) by startups who innovate on top of bank account data and aim to commoditise  banks into being low margin utilities. Actually that is not the objective of startups. They just want to offer a better service to consumers and to do that they need access to account data; turning banks into commoditised utilities is simply a by-product consequence.

PSD2 and other regulation is key to this. When banks have to open up their data, then they will.

It is a level playing field. Banks could win. Methinks we will see a lot of ChatBots from Banks soon.

level playing field

Daily Fintech Advisers provides strategic consulting to organizations with business and investment interests in Fintech & operates the Fintech Genome P2P Knowledge Network. Bernard Lunn is a Fintech thought-leader.

ChatBots in InsurTech From Offshoring to Automation to New Experience


This is Day 4 of ChatBot Week on Daily Fintech. You can see the intro and index here.

Unusually we are seeing more adoption from Incumbents than upstarts . This is the reverse of the cycle of innovation in banking (startups show the way and then banks “fast follow” via mix of copy and acquire). This post explains why it is being done differently when it comes to ChatBots in Insurance and then looks at a few early stage startups that may point to the future.

The action is from Incumbents

Look at this list of ChatBots in Insurance compiled by @rickhuckstep who tracks InsurTech very well. You can see a long list of incumbents and one startup.

You can also see what Halifax is doing on which does Service as well as Sales (Service is the key to retention) and is big in Insurance is clearly looking at this.

First, turn humans into robotic workers

Cutting payroll cost has been a favourite enterprise project since the dawn of computing. Insurance has a ton of routine customer support work. The product is complex and critical to the customer. Good customer support is a) essential b) expensive. This is the kind of work that was first offshored and is now being automated. Once a job has been made totally process driven, eliminating individual judgement, it is quite a simple step to automate that job. An offshoring project works by breaking tasks down into repeatable, fungible “units of work” that are given to humans to do as if they are robots. If those humans leave, they are easy to replace. It is a short step from robotic workers to actual robots and that is what AI enables.

This is of course not good news for the $28 billion BPO business because about 5-6% of those jobs are low end as per this report (ie low hanging fruit for automation)

From Hal to Her

When we search for images of AI, the robot appears. The disembodied voice of Hal in Stanley Kubrick’s brilliant 2001 a Space Odyssey (released in 1968) was a bleak  view of our robot infused world.  In another brilliant science fiction film, Her by Spike Jonze (released in 2013), the robot has the gorgeous voice of Scarlett Johansson and so the protagonist naturally falls in love with the robot.

The point is we don’t see the robots who are serving our admin jobs. We see their texts or hear their voice. Maybe holograms will be the next attempt to humanise them.

Humanising robots is just marketing

Robots in factories making stuff for us are invisible to us, so they don’t need to be humanised. Robots taking care of our every day admin needs have to be humanised to be acceptable. It is Marketing 101.

This started in offshoring where the rule was that call center workers working in the Rest imitated the accents, names and behaviour of their customers in the West. Krishna, Pranav and Deepika become Bob, Ted and Alice.

The ChatBots found by Rick Huckstep are called Magda, Allie, Mia, Arbie, Nienke, Marc and Hanna. Pick a name that fits the culture of the market you are serving, it costs nothing.

From makeup to makeover

Mobile Apps followed the usual trajectory of disruptive innovation from talking heads on TV (Radio to TV) to something that is native to the new media. Early Mobile Apps were simply bad web sites on a small screen. ChatBots will follow that new trajectory.

And now for something different

Spixii is not a cute name. OK maybe it is – cuteness is in the eye of the beholder. Apparently it means Blue Parrot. I guess was not available.

This is bleeding edge.  Spixii got €15k funding at the end of 2015 as per Crunchbase.

SPIXII looks like AI with a ChatBot UI – a RoboBroker if you like. As they put it:

“The first development of SPIXII consists in selling insurance using a chatbot.”

Their TLD is .ai They clearly aim to be AI with a ChatBot UI.

You can watch their Demo Day pitch here.

This feels like SimpleBank – changing the User Experience but not as a full stack regulated Bank/Insurance company. SimpleBank was a success – founded in 2009, acquired by BBVA in 2014. SPIXII may follow the same trajectory.

Insurgram from Germany is “Insurance by Chat” and they look a bit more mature than Spixii and have a partnership with Ergo Direct (think Geico if in America).

HeyBrolly is also aiming to change the whole Insurance experience and describe themselves as a concierge rather than a broker (so chat is just a natural UI layer to a reengineered experience).

Early days

Think of the app store in 2008  – that was 8 years ago. Current examples of ChatBots are fairly crude and occasionally embarrassing and funny but this is the usual story with new tech and the economics of admin service job automation using AI + Chat is compelling, so I am sure they will get better.

Daily Fintech Advisers provides strategic consulting to organizations with business and investment interests in Fintech & operates the Fintech Genome P2P Knowledge Network. Bernard Lunn is a Fintech thought-leader.

Announcing ChatBots Week on Daily Fintech


Image courtesy Admirable Design

This week is ChatBot all week on Daily Fintech. Today is the Tech in FinTech day when I attempt to summarize why so many people are excited by ChatBots and describe who is active at the platform level. Then we look at the use cases in WealthTech, Small Business Finance, InsurTech and Consumer Banking.

If you are new to ChatBots, here is a good intro (yes, there is now a ChatBot Magazine).

First we look at ChatBots through three different prisms:

Prism # 1. ChatBots = User Interface (UI) 4.0

Prism # 2. ChatBots = native UI for AI

Prism # 3. FB Messenger BotStore is like the Apple AppStore in 2008.

Then we look at what the Internet Platforms in America (GAFAM) and China (BAT) are doing with ChatBots.

Finally, to bring it home to Fintech leaders, we explain the relevance of one specific feature called Account Linking.

There will soon be a Fintech version of the Turing test. For example, is that my friendly private banker recommending an asset allocation change? Or is it a robot programmed to act like a friendly private banker?

Prism # 1. ChatBots = User Interface 4.0 

1.0 = Text command line eg DOS & “green screen” mainframe terminals.

2.0 = Windows Icon Menu Pointer (WIMP) in PCs and Macs.

3.0 = Smartphone touch

Yes, just when we all agreed on mobile first and mastered those skills….the game changes again. Apps have a problem because people are reluctant to install apps (it is the app tax friction).

The reason why ChatBots are UI 4.0 becomes clearer when we look at the next Prism.

Prism # 2. ChatBots = native UI for AI

How do we interact with those machines who appear human like in intelligence? One way is the same way that we spend a lot of time interacting with flesh and blood humans – via text messages. It is the perfect UI to hide all that complexity.

For our AI Week please click here.

Prism # 3. FB Messenger BotStore is like the Apple AppStore in 2008.

When the Apple AppStore launched in 2008, I caught a tweet by a friend that said in effect “this changes everything”. I did not get it at once so I called him and he explained and he was right. Thanks @alexiskold

The launch of Facebook Messenger platform with ChatBots on 12 April 2016 had the same inflection point significance.

6 weeks later VentureBeat was reporting 11,000 chatbots that you could try. Talk about Cambrian Explosion!

What are GAFAM doing with ChatBots?

  • Facebook (already covered above)
  • Amazon does it their own way as usual. Jeff Bezos always surprises. Their Alexa device is an AI machine in a home device. You don’t talk to Alexa with text messages. You talk with your voice – how cute and retro! It’s more VoiceBot than ChatBot, but conceptually both are UI for AI. Amazon has opened this up to developers.

Apart from the giants, all major messaging services from SnapChat to Telegram (which has a good intro for developers) have initiatives.

What are BAT doing with ChatBots?

  • Baidu has Duer, a digital concierge service.
  • Alibaba has not made a move yet as far as we can see, but given their big ambitions in FinTech, we can expect it to be highly relevant when they do.
  • TenCent has WeChat, currently the world’s biggest messaging platform with 700 million users, so they were early with ChatBots and have a digital concierge service called WeSecretary that helps users buy products, book restaurants, pay bills and other internet-based admin activities. It sounds more practical than SIRI.

ChatBot Account Linking

Facebook has an opt-in feature called account linking that allows you to, for example, link your Facebook account with a retailer. The ChatBot can then suggest new products based on a recommendation system. Watch out Amazon.

We expect this intersection of payments and e-commerce to be very active.

An example from Telecoms shows how this could work. You can chat to your Telecoms assistant to find out how much bandwith you have left or how much an international call will cost. That clearly has to link to your Telecoms account. Now imagine opening up to your banking accounts using PSD2.

Calendar for the rest of the week

Tuesday          ChatBots in WealthTech

Wednesday    ChatBots in Small Business Finance

Thursday        ChatBots in InsurTech

Friday               ChatBots in Consumer Banking

Please tell us about any use cases in those areas that you know about. We may already be researching them, but a heads up is always appreciated.

Daily Fintech Advisers provides strategic consulting to organizations with business and investment interests in Fintech & operates the Fintech Genome P2P Knowledge Network. Bernard Lunn is a Fintech thought-leader.

Shift Technology using AI to battle Insurance Fraud #insuretech

fraud detection

When I first spotted Shift Technology with their focus on fraud detection for insurance, I assumed I would find a venture in Israel (which is known for smarts in finding the bad guys in cyberspace, as we outlined when we went to Israel on our Fintech global tour). So I was surprised to find that Shift Technology is a Paris based venture. There is a lot more tech innovation in France than the image of economic sclerosis would lead you to assume. The next thing that jumps out at you is that they recently closed a $10m Series A round in a tough market from a top tier VC (Accel Partners). So they must be doing something right.

Pattern matching in the data

Call it AI or Machine Learning or Big Data or Data Science, the buzzwords accumulate, but this is about recognising the patterns of behaviour that indicate that bad actors are involved. Credit Card networks have been using this kind of technology for a long time. So have cybersecurity vendors and intelligence agencies. What we are witnessing now is this technology becoming more mainstream and being applied to new use cases. We looked at some of these use cases during our AI Week on Daily Fintech.

Fraud detection is the key and not just for Insurance

When you look at breakthrough innovation in e-commerce and payments, you often find fraud detection tech at the core. This is the story behind Klarna.

Fraud detection is also key to P2P/Marketplace Lending and Crowdfunding.

Shift Technology is currently laser focussed on the Insurance use case, as is evident from their home page where they have a real time clock showing the number of insurance claims analysed to date (over 54m at time of writing).

SAAS Model Could be for Incumbents or for Upstarts

Their business model is simple – they license via a SAAS model. One secret to venture success is to innovate on only one front at a time. You can be quite conventional on the business model while innovating on the technology or vice versa. Fraud detection  is a problem that incumbent Insurance companies will pay to solve. One can also easily imagine this as part of the solution  stack for an upstart P2P Insurer.

Data Aggregation & Sharing based on shared interest

The key to their business is the fact that clients are motivated for Shift Technology to aggregate and share the data. Writing smart fraud detection algos can be replicated by anybody with the budget to hire good developers. It’s the data that matters. In fraud detection, customers have a motivation to share. So this business has network effects and barriers to entry.

Daily Fintech Advisers provide strategic consulting to organizations with business and investment interests in Fintech. Bernard Lunn is a Fintech thought-leader.

Could #InsurTech AI machines replace Insurance Actuaries?

This is the fourth in our AI in Fintech Week series. You can see the intro post describing the current state of the art in AI here. Today we look at a job that very few people understand. It is a job that requires an aptitude for math and statistics plus knowledge of complex domains such as life expectancy, healthcare, accidents, weather, wars & terrorism. Fundamentally it is a job that requires math and statistics; our AI friend Hal is heard to say, “I am good at math and statistics, give us a job”. It is possible that the flood of new real time data into Insurance will change the Actuary job beyond recognition or even replace it with AI machines. That is the theory that we explore in this research note.

The Actuary job

Actuaries do statistical modeling to assess risk. However unlike using statistical modeling to assess financial risk, the job is not open to anybody with skills in math, statistics and computers. The Actuary job is a profession with rigorous professional certifications.

This site has the facts.

  • It requires a big investment in education. “It could take from 6-10 years to pass all of the exams, but you can begin a career as an actuary by passing the first two exams, and then taking subsequent exams while working as an actuarial assistant.”
  • It is a well-paid job. “Experienced Fellows have the potential to earn from $150,000 to $250,000 annually, and many actuaries earn more than that.”
  • There are not that many Actuaries. Estimates vary but the number is around 10,000 in the USA. An upper bound would be 3x globally ie about 30,000. That is small compared to other professions/jobs.

Data is the game-changer

There four main domains for Actuaries to model

  • Property & Casualty for Cars.
  • Property & Casualty for Houses.
  • Life Insurance.
  • Health Insurance

The data points, and models are quite different in these 4 domains, but data is the key to all of them and that is where disruption is coming from. The big game-changer in Insurance is the data flowing from new data sources such as telematics in home and in car for Property & Casualty and wearables on the body for Life and Health.

One company to watch is Atidot, which is a classic Israeli startup (founders ex military intelligence, steeped in cyber security, data science and software development).

Atitdot are moving the actuarial process from classic statistical analysis using static models to predictive analytics using dynamic and non-linear models. The real world works on real time, dynamic models. Reality does not obey the dictates of a static model. The problem is that our organizational processes are based on static models.

When will this person die?

That is a classic question that actuaries need to answer, specific to Life. In Health, the question might be “what is the chance of this person getting diabetes?” In Cars it might be ““what is the chance of this person having an accident?” Lets stick with the Life question as it is the simplest. A simple statistical model might have the following variables:

  • Age
  • Sex
  • Job
  • Location
  • Health risks such as tobacco, alcohol and medical diseases such as diabetes.
  • Average life expectancy

Occasionally the model is changed because of something global changes such as average life expectancy. Sometimes you can refine based on changes in risk profile of a job or a location. These changes affect so much in the Insurance business that they need to be carefully considered and go through committees and long established approval workflows.

That is a description of the Static Batch Enterprise, but the world is dynamic and real time.

Atidot applies many more variables simultaneously to process much larger data sets. The new data sources – telematics, wearables, social media, weather, news – are inherently real time and dynamic.

Some other ventures in this space worth tracking are:

Quantemplate focused on Reinsurance

Analyze Re focused on Reinsurance

FitSense focused on wearables data for Life & Health.

Wunelli focused on in-car telematics data for car insurance

Other AI Use Cases in Insurance

There are many possible use cases for AI for Insurance. Many of the use cases cited are really Big Data with some relatively simple algos. These are valuable but do not fall into the reasoning machines that we call AI. There are use cases where a robot replies via SMS to a custom query. These are real AI and they can to pass the Turing test. However they are not specific to insurance and can be applied to all CRM and Support use cases in any industry.

A huge pain point specific to Insurance is changing models related to weather due to climate change.

The company to track here is Meteo Protect.

AI can be used in claims processing to detect fraudulent claims. This is where human experience can detect patterns.

Replace or Augment?

Our thesis is that Actuaries will be augmented rather than replaced. AI machines will do the work that Actuaries do today. However human Actuaries will move up the value chain for two reasons:

  • The Actuary profession is full of smart, well-educated people and many of them are in positions of power. They will adapt to change.
  • There is far more value to an Insurance company from accuracy than cost-cutting. An experienced Actuary empowered by real time data and non-linear models will be hugely valuable. The Actuary defines how much to charge for a Premium and that is as core to an Insurance company as you can get.


In some jurisdictions, Insurance companies have to communicate to consumers how rating factors such as gender, age, zip code are used in pricing. That privacy issue will get more complex as we get into in-car telematics, smart home sensors and wearables. It is hard to imagine an Insurance company being allowed to do surge pricing like the transportation business just because the AI machine tells the Insurance company that risk has spiked.

This will be an issue that regulators will have to grapple with as InsurTech startups and Insurance incumbents battle for customers.


Daily Fintech Advisers provide strategic consulting to organizations with business and investment interests in Fintech. Bernard Lunn is a Fintech thought-leader.