datamaster summit 2020

How Banks Can Use Machine Learning To Enhance Customer Experience

 

Dan Waldner

Principal Strategist, Financial Services at Traction on Demand

From providing stellar customer service to complying with regulations, 360-degree customer views are a must for banks. Oftentimes, financial institutions struggle with creating these accurate views. Data silos and traditional approaches to master data management lead to limited customer views, slowing lending decisions, hindering marketing efforts and impending effort to meet KYC requirements.

But quickly and efficiently developing up-to-date customer records is possible. Banks are using machine learning to create unique profiles of every customer, paving the way for regulatory compliance and reporting efficiency. By mastering customer data, financial institutions can make more informed credit decisions, approve loans faster loan approval and readily identify cross-sell opportunities.

Leveraging this machine learning in a cloud-native environment further empowers the banks by providing them with the infrastructure to master customer data at scale.

Transcript

00:02 – 00:21

Speaker 1

The Consumer Bankers Association is pleased to welcome you to today’s webinar, how banks can use machine learning to enhance customer experience by Palmer. My name is Victoria and it is my pleasure to facilitate today’s event. Thank you for joining. Please note we are recording and all participant lines from you.

00:22 – 01:01

Speaker 1

If you have any trouble, please email conferences at consumer bankers dot com or send a message in the queue in a box. This presentation will last up to 60 minutes and will include question and answer opportunities. ABM You may submit a question at any time by entering the questions into the queue in a box on the bottom of your screen. As a reminder, the views expressed in this webinar are those of the presenters and do not represent the views of its members. It is now ready to introduce our speakers. Dan Waldner and Robbie Halsey, Dan and Robbie welcome.

56:17 – 56:35

Speaker 1

Thank you, and which bathroom must conclude today’s program? The session has been recorded and will be available within three to five business days. On behalf of the Consumer Bankers Association, thank you to our speakers. And of course, all of today’s participants. Have a good afternoon. You may now disconnect.

01:03 – 01:52

Ravi Hulasi

Thank you, Victoria, and good afternoon, welcome everybody. My name’s Ravi Hulasi. I’m Tamr’s chief cloud evangelist. Some of you may have been expecting my colleague Joe Schaefer to be joining us today. Unfortunately, as many of you in Northeast know this, quite some heavy storms going on and Joe recently lost power. So I’m stepping in for him today and hope Joe stay in safe and dry out there as we all are. So my background is I have been a data quality data management space for about 20 years now, working with a number of different customers in the financial services industry on common problems such as BCBS, 239 ethics problems and dealing with clients reference data. So before we get into our conversation today, I’d love to introduce my partner here, Time Warner.. Dan, Welcome.

02:53 – 03:24

Ravi Hulasi

Wonderful, thank you, Daryn. Before we get into our discussion, just got a question for the audience, our first poll question of this session as we’re talking about customer data, well, let’s let’s hear from you. How dirty or decentralized is your customer data? So you’ll see a poll pop up on screen. And please, we look forward to hearing your input on that. And of course, if you have any questions from a session, please feel free to add them for the Q&A panel and we’ll have some time at the end go free values as well.

03:24 – 03:45

Ravi Hulasi

But let’s start talking about that customer experience and the need for high quality customer data. So then can you tell me a bit in your experience? How how is customer data looking across the financial services industry right now with all these technologies coming in all these initiatives? What’s been the results of where things are looking?

13:53 – 14:51

Ravi Hulasi

Having done, even from my experience, just hearing about some of the challenges, they really resonates with me. For example, regulatory reporting with data warehouses where people just don’t want to touch them will add additional sources because they know the processes that are in place are so tightly coupled to the data that any change could just cause it to fall apart. So it often restricts the scope of improving and extending those systems. And just looking at the poll results here, I think I can see that our audience also has similar problems. Sixty six percent of the audience said their data was somewhat dirty and in need of a spring cleaning, and 33 percent said it was dirty. Notice in here that no one was either super proud of that either side completely understood and fit for purpose, and no one said that it was absolutely abhorrent. So it sounds like people have taken on some of these challenges around them, but there’s still a lot more to go and to improve.

14:51 – 15:07

Ravi Hulasi

So moving on to how how do we start if we’re talking about customer data here? So why do we want to start with customer data first when there’s so many different entities within an organization which warrants this sort of treatment?

19:11 – 19:54

Ravi Hulasi

I think we’re also seeing that as well when we see customers now, many of whom in the past have been maybe more averse to technology such as the cloud and the benefits it offers. Now when we talked to customers after, my first question is how how you deploy them a cloud, do you have a cloud native strategy? So clearly that urgency is driving customers to start to look for more modern technologies and approaches to achieve their goals? And so thinking about just that customer data a little bit more how of abuse are available is, is that 360 degree view something that you see a lot of or even going beyond that 720 and and household and view how much how far has it been achieved so far in the space so far?

22:25 – 22:47

Ravi Hulasi

And so he didn’t just think about that view, its main banks have had customers for a long period of time. They’ve got a lot of data about these customers, but they’re still struggling to build that Customer 360 view, let alone a 07:20 view. So why is this still a problem? One the pain points that people are really running into as they attempt to solve this.

26:55 – 27:50

Ravi Hulasi

I mean, clearly, it seems that centralizing data is is a key challenge, and that brings us on to our second poll question today. Have you tried and failed at centralizing your data so I can be honest that we’d love to hear from you? Also seen a lot of questions coming in on the Q&A panel as well. We’ll get to those a little bit later on. Please keep them coming. So I mean, before before we continue to talk more about technology, I think it’s important to look at how do you build a compelling business case? Often we hear customers and prospects struggling to understand in all ROI. How is that expressed? How is that set up? But can you tell us a bit in your experience? What were some of the benefits of going through that process to define and articulate the ROI before embarking on such essential before on centralizing projects?

33:04 – 33:19

Ravi Hulasi

And so I think many of us have experienced that poor experience. We’ve we’ve we’ve interactions where customer data is just not flowing across systems, across channels, but what other business impacts are they just not having a good grasp of who your customers are?

33:51 – 34:16

Ravi Hulasi

And I think I think the audience agrees, I mean, the the whole theme of centralizing data, looking at the poll results, here are 50 percent are saying yes. They have tried so far to centralize their data and have been successful. Congratulations on that. And following up with that, the remaining 50 percent are trying, but they haven’t really gone to that state of success yet, either due to failure or just maybe not enough time.

34:16 – 34:36

Ravi Hulasi

So let’s switch gears a little bit and talk about how we do this then and how can we actually go about being successful in centralizing our data? Well, what techniques? What approaches are you seeing that when people are successful, that can be attributed to those babies? Our approach is being taken right.

40:40 – 41:10

Ravi Hulasi

Even in our experience, as well as customers have tried this before, but it doesn’t have to be wasted or thrown away. Certainly if people have gone through that process of building that customer view with a limited number of sources that can also be valuable inputs into the system as well, where people have gone through those in terms, both data stewards have maybe attacked a fraction of that data that’s available. Let’s use that as an input into the system as well. So all is not lost data now and actually.

43:22 – 43:44

Ravi Hulasi

And I completely agree. And I think by way, if you’re looking for a restaurant recommendation and collect and think about this, now, what this means for the broader picture, how this involved an improved initiative like KYC and how much more do people have to do to really support such initiatives?

45:49 – 46:03

Ravi Hulasi

Absolutely. I think even along the whole KYC lifecycle with many touchpoints, many interactions from the on boarding free to the day to day execution, which really warrant this sort of Customer 360 view both internal and external to the organization.

46:04 – 46:46

Ravi Hulasi

So exactly looking at the time, now it’s time to go off some questions how if you come in first question, let’s see here I work for nCino, which is a bank operating platform, retail treasury, commercial deposits, etc. all in one platform. If we have a platform similar to what you are describing as the dream state, how do you recommend we take it to the next level? And so just before I pass to you, Dana, I know one one of Tamr’s largest fintech clients has been deployed innCino environments. And so this, I think, lends itself naturally to that sort of data mastering challenge that Tamr solves, but dynamic thoughts and advice to our audience here.

49:18 – 49:58

Ravi Hulasi

Look, our second question here, then I’m trying to merge multiple customer instances from a recent acquisition. What advice do you have for me? And from my perspective on experience, I think the acquisition, the merging of the data starts prior to the actual acquisition finalized. And often we’ve seen clients try and get that directional view of customer overlap. And that’s often seen as a throwaway piece of work, whereas I think there’s a real opportunity to put in place a process that can rapidly get results. Again, it’s directional results can be iterated on post acquisition when the real consolidation happens. But Dan, any advice that you can offer our audience here?

52:25 – 52:37

Ravi Hulasi

And I think that’s that’s fantastic advice that he leads on to a final question for today. Do you recommend that companies merge sales force instances together into the existing instance or create an entirely new one?

54:21 – 54:46

Ravi Hulasi

And having been involved, have just Salesforce projects, just with a handful of instances before to trying to consolidate not only with data, but the application logic for each of those just is this cause for concern and nightmare as well. So I think it’s important to separate those particular views, make people focus on the application. And there’s a whole data challenge here that needs to be considered and treated as a first class citizen.

55:43 – 56:16

Ravi Hulasi

On the force, I think we’re our time here, so with that down, I’d like to thank you so much for your insights, reports today and advice, and I’m sure our audience today, whether they’re starting that centralising journey, try to build that first Customer 360 or along the way and are looking for advice and tips on how to improve that. I’m sure they’ll all have plenty to draw from today. And with that, I’d like to thank the audience for attending. Thank you, Dan. and wish you a happy journey with your data. Thank you all.

01:52 – 02:29

Dan Waldner

Thanks, Robert. My name is Dan Waldner. I am principal strategist. Attraction on demand in the financial services space. I have spent almost 28 years in technology with about half of that time in the financial services sector. Most of that time in service was spent with Scotiabank, Canada’s third largest lender, where I owned at various times the entirety of the corporate real estate dataset and the architecture patterns there. I owned the entire Salesforce instance for the entire enterprise from corporate banking, commercial banking, capital markets and then finally wrapping up my time.

02:29 – 02:52

Dan Waldner

There I was the head of the customer data program, which was tasked with centralizing all of our customer data across all of our various capital markets and global banking markets, lines of business. So customer data and specifically how to leverage customer data to gain efficiencies and to maximize revenues is something that’s a passion of mine and something that I look forward to talking to you about today.

03:46 – 04:05

Dan Waldner

Yeah, it so as as the entire industry as a whole, it’s about as chaotic as the data quality itself. There’s a lot of problems with customer data right now in financial services. And for the purposes of this discussion, we kind of gravitated some of them towards the three big data challenges. But there’s many, many others out there. So we’re just going to focus on the big ones right now.

04:06 – 04:29

Dan Waldner

The first. First and foremost, is just the sheer volume of data that’s out there right now. When you think about, you know, the size of firms in general, even mid-sized firms want banks or credit unions that have, you know, 20 different locations, let’s say, a regional bank, they’re still creating terabytes of data per day.

04:29 – 05:25

Dan Waldner

When you think about, as an example, the volume of transactions on the stock market, for instance, if you think if you take a specific subset, there’s five point eight billion daily transactions on the NYSE alone, just one stock exchange. And then for each transaction that occurs as part of that, you have settlement instructions. You have attacks often affects transactions that have to occur. You have customer records that are entangled with that. You have audit records and swift messages and email messages to confirm every step of the transaction. Every transaction alone makes hundreds of rows in various platforms and applications throughout a financial institution. And that’s just one little corner. I shouldn’t say little one corner of your firm. That is a microcosm of the capital market space alone.

05:25 – 05:49

Dan Waldner

So if you expand that out to debt capital markets to additional additional exchanges, to additional lines of business like lending and retail, commercial banking, corporate banking, investment banking, the amount of data that gets produced is just staggering and the different complexities and types it’s it’s enough to choke any good strategy.

05:50 – 06:26

Dan Waldner

Second, if you look at how your your internal ecosystem is aligned within a financial services sector, it’s often incredibly siloed as well. Each line of business traditionally has its own constellation of applications that it uses now. My my recent history has been in the capital markets and we’ll make it market space where there’s often one one application for every product or product grouping. You have a precious metal system, you have an effect system, you have an equity system, you have a settlement system and so on. And many times they’re all self-contained.

06:26 – 07:03

Dan Waldner

So there’s lot of siloing of duplicate data across your organization, some lines of business share systems in a shared services model. So you might have one onboarding system across numerous lines of business, but often even things like onboarding is duplicated across big chunks of of corporate functions as well. So you don’t have agreement across your entire organization around doing the same business processes and capturing them in the same in the same way. So your data is often duplicate, scattered, siloed throughout your entire organization.

07:03 – 08:01

Dan Waldner

And while it is becoming more en vogue to to have an aggregation layer like a data warehouse or a data lake to be able to compile all this information, it’s often very narrowly use case. And what I mean by that is it’ll be a regulatory requirement to to produce these types of reports. And that’s the narrow focus for this warehouse. So they’ll collect all of the information that they need to be able to produce reports for the Fed or produce produce reports for local jurisdictions or whatnot because they realized that the actual process for creating a fulsome data warehouse or fulsome data lake is often overwhelming multiple years that cost a lot of money. So your your ecosystem, not only is it siloed, but even when it tries to aggregate in the hopes of of normal. Losing the information across the board, it often does it doesn’t have the necessary scope or funding to be able to make that a fully throated solution.

08:02 – 08:44

Dan Waldner

And then finally, you’ve got heterogeneity, right? And and what’s really meant by that is you’ve got every flavor of system you can possibly imagine. You’ve got on premise systems and cloud systems and mainframes and dot net and Java and S400 and ruby on rails and visual basic new systems and old systems, ones with good data, ones with bad data and really every possible combination of permutation in between. Coming up with a good strategy to allow all these systems to talk with one another and to compare at the data level is a staggeringly difficult problem to do with all of these different variables that are going on at the same time.

08:45 – 09:27

Dan Waldner

So how do we actually look at solving some of these problems? And there are three groupings. Everything happens in threes. It seems there’s three easy groupings for four, how do we solve this? The first is let’s go hire a whole bunch of interns to go and take a look at all of our bad data, and we’ll hire 50 people. All of the college students need something to do in the summer and we’ll put them to work, you know, hammering out the correct information where we had bad data. And that doesn’t really work for for the low hanging fruit it does. You know, if you’re missing addresses that you can easily be looked up and things like that that sure, it addresses it temporarily.

09:27 – 10:11

Dan Waldner

But data has a shelf life itself. And oftentimes, especially when you’re looking at organizational or institutional information, the shelf life is between six months and a year when that information is no longer relevant. Companies have moved. Companies have merged. There’s been a restructuring. There’s new holding companies and whatnot. So just taking a bunch of students and throwing it at the problem, it might remedy it for six months. But eventually all of that problem, the problems come back and you have to do this on a regular basis. And it’s not a really sustainable solution, especially when we’ve already spoken to the fact that data is huge and it continues to grow. Right. So you have to hire more students and do more manual effort. It’s just not a repeatable or reproducible solution.

10:13 – 10:36

Dan Waldner

Then the other way that often gets parodied around is let’s let’s go out and get one of these traditional MDM systems, which we’ve been promised you can put on top of our functioning data warehouse, and they’re going to write a bunch of rules and it’s going to produce an output that is relatively clean and sustainable for us, and that works to a degree.

10:37 – 11:34

Dan Waldner

And what I mean by that is when you look at the math required to be able to build out a rule set to connect the dots between system and System B and System B and System C, System C and D, et cetera, you only can get about eight to 10 systems into such a rule a rule based approach before you have millions and millions of rules required to keep these systems in alignment. The system ends up getting brittle and ends up getting broken when their schema changes. You’re constantly having to fix this, and you end up plugging more holes than the solution actually solves. I’ve seen it too many times where there’s too many systems into an aggregation MDM, and it just ends up breaking the system, so that only works to a degree. But if you’re talking about, you know, large scale organizations that have dozens or hundreds of systems like capital markets with their individual product systems, you have to take a different approach. It’s not going to work.

11:35 – 11:54

Dan Waldner

And then finally, you know, let’s let’s forget about rationalizing all that information. Let’s just get an aggregation point like a data warehouse. Let’s just throw all of the information in there. And when the time comes, we have to produce a report. We’ll just pluck the information out ad hoc. And that’ll be that’ll that’ll be our solution, right?

11:56 – 12:40

Dan Waldner

Doesn’t really work that way because you have a lot of problems with contextual information. And what I mean by that is, let’s say you have 10 systems that are in your data warehouse and they all use the term closed eight for what is closed. Eight means well in the equities system. For instance, the closed eight would be the data that the trade actually closes, the hits, the the exchange and you receive the equity. That’s when the closed date is. But for the settlement system, that might the closed date might be the date that the reconciliation occurs and the closed date for the customer system might be the date that the equity appears in the online banking system and all these dates, they’re all called closed date, but they all mean something completely different based on the context.

12:40 – 13:20

Dan Waldner

So when you’re producing these reports, if you’re not crystal clear about what context you’re using for closed date and this particular report needs to have that field in there, you could end up reporting on bad data, and that leads to a number of terrible. So it leads to systemic issues if you don’t catalog your information in a correct way so that you’re all in agreement on what the terminology uses. And frankly, I’ve seen more data warehouses and data lakes turn into swamplands or dumping grounds, or, as I call it, the junk drawer effect where data easily goes in. But nothing of use comes out, and that tends to be one of the problems with that temporary solution.

13:22 – 13:51

Dan Waldner

So, you know, the I’ve dealt with hundreds of firms, and I’ve never seen one that has said that their data is in a good state and hopefully many of you watching this now find at least one of these core data challenges, probably all of them that apply to your your firm. You’ve probably tried one or all of these temporary solutions. And hopefully this webinars can be able to paint your picture of what to do something that isn’t on this slide that addresses these problems.

15:08 – 15:32

Dan Waldner

Yeah, it’s a great question. Customer data is the bedrock of everything that we should be doing as financial services firms from a from a conceptual level, customer data is the easiest to wrap your head around, right? We all conceptually understand the individual, the Dan Waldner. Dan Waldner lives at this address. This is his SSN number, you know, the phone number or fax email.

15:32 – 16:11

Dan Waldner

email. of the traditional data points are relatively easy to define, relatively unambiguous across the various lines of business from a B2C perspective, from a B2B perspective, institutional corporate type information. It’s also relatively easy to standardize, right? So all lines of business generally treat organizations with the same type of information registered address, country of incorporation, state or province of incorporation, what they regulate or what their tax IDs are. If they’re capturing glyph information LII information. Do you capture any external identifiers like Duns ID or Definitive IP, right?

16:11 – 16:35

Dan Waldner

So we can come to a relatively quick agreement around the critical data elements that go into customer data? And then we can also easily define those in a in an enterprise way and then get people to easily map those those data points from their target systems into kind of a relatively easy, straightforward beginning.

16:35 – 17:15

Dan Waldner

Enterprise data dictionary, right? So ease of use, ease of use is is number one, right? It also aligns well with existing onboarding and KYC processes, so you can make a pretty easy case for obtaining funding. Traditionally, under what we termed defensive benefits, and I’ll get to that in a little bit. But mainly, you know, most financial firms have a relatively robust budget for compliance and regulatory purposes, as well as anti-money laundering and terrorist financing. So being able to say that we can, we can improve these processes by aggregating our customer data into one such repository generally easiest to obtain that type of funding.

17:16 – 18:02

Dan Waldner

And then it sets the table for offensive benefits. And the chart that that’s up on the screen right now kind of addresses that offensive benefit. You look right now there’s a schism in the financial services world, one where you have digital natives, which are your fintechs and your your technology firms that are that are migrating their way into the financial services space that are from the terminology are they have no technical debt or very little technical debt. They’ve been able to blue sky out their data architecture as well as their technical ecosystem from scratch. So there doesn’t have that burdens that some of the non digital incumbents have right in the non-digital incumbents.

18:02 – 18:28

Dan Waldner

Of course, being your your enterprise banks and kind of your your older credit unions and regional banks that are trying to play catch up with the data driven nature of these digital natives, right? So by setting the customer as that stepping stone the first point within your data architecture, you can start to build out from there looking next at account level data and then after that transaction level data to be able to slowly bridge that gap.

18:29 – 19:09

Dan Waldner

What you see on this chart is the pandemic accelerant has really highlighted. It’s increased the gap between the digital natives and the non digital natives because the non digital natives spent a good chunk of last year trying to adjust from a brick and mortar based working environment for meeting customers in their branch networks. What not to go fully virtual, where the digital natives? It was status quo. There was no difference between the pre-COVID and post-COVID world. So you’ve been able to see the difference that date has been able to make and growing their bottom lines through just the reprioritisation of things.

19:55 – 20:32

Dan Waldner

No. And I don’t think it’s going to come as a surprise to most people on this call that because data is so siloed in the organizations, the perspectives of the customer are equally siloed. Everybody has tried to do a Customer 360 with their existing infrastructure, and it’s turned into a very manual effort. Each line of business within most organizations has a pretty reasonable sense of their view of the customer. So your wealth management firm knows its wealth customers well and your commercial banking knows its customers well and corporate banking and capital markets and retail and whatnot.

20:32 – 21:22

Dan Waldner

But when you start to think about, OK, let’s take this perspective of Dan Waldner. What type of holdings does Dan Waldner have as the individual? What does Dan Waldner do as as a am I? What’s my my job? Do I have that? You mentioned Customer 720. That’s the that’s the dual concept of Customer 360 from a B2C perspective and then Customer 360 from a B2B perspective, because Dan Waldner with as the person that lives at this address with this mortgage and his credit card and his checking account also has responsibility. And I’m a decision maker for a small business, right? So understanding the duality of his personal and his business level relationships is something that’s so far beyond most data architectures at most firms.

21:22 – 21:57

Dan Waldner

Now that it’s actually viewed as the holy grail to be able to fully realize the full weight of a relationship between an individual, both from their personal and their end, their business level influences and the financial firms so they can adequately gauge the value of that relationship. Many firms are a decade away from being able to do that. Some of the some of the more the digital native firms and some of the more cutting edge larger firms are able, they’re approaching that ability, right?

21:57 – 22:23

Dan Waldner

But there’s so much technical debt out there that it makes it very difficult to kind of bring those silos together. It’s the customer 720 is a term that I’ve kind of I like to to to use because I think it adequately represents kind of that that dual full circle picture that allows us to to see all of the entanglements of the individual.

22:48 – 23:30

Dan Waldner

Well, that’s a great question. When when you look at the the previous slide showed a picture, a graphical nature of what it looks like from just one line of business. Right. If you could quickly go back to slide six, if you could. Katie V. Here we go. So what we’re seeing from this graphical representation is the perspective of, let’s say, for in this case, we’ve got a full service bank, right? This is what it looks like from the retail banking side of the business because all retail has a perspective of IS for the purpose of that, just the D-Day presence for Freya Gibson, right?

23:30 – 24:07

Dan Waldner

So this one system has their checking account, and that’s all I can see, because the way that the information has been siloed in this organization, the wealth management side of the businesses is separate. You’ve got insurance, which is separate. You’ve got Freya’s professional responsibilities as CFO of a mid-sized firm in the Midwest that’s not seen in here as well. All you can see from Freya’s experience is that she has a checking account because she happens to do most of her banking with that with another bank, but she’s got a small presence with your bank, but she also has a very large presence in those other. Those are the other lines of business that I mentioned, right?

24:08 – 25:05

Dan Waldner

So if you skip ahead to the to the next slide, when we take all of those different lines of businesses information, and let’s say we’re trying to do a data centralization initiative where we’re starting to piece together all of these different perspectives of Freya into one collective perspective. You get something that looks like this, right? And so Freya Gibson in one system is kind of connected to Freya G, to Freya Gibson, to F Gibson with two ends. You’ve got these different perspectives of the data that all look like they’re kind of related to what it should be, but not quite so. Writing rules in these situations are going to be less than effective, or you’re going to have hundreds of rules to try and fuzzy match on different things. And then you’ve got these different concepts like Andre Gibson and Charlie Gibson, Guy Gibson. They look like they belong in Freya’s orbit, but not really sure how right.

25:06 – 25:49

Dan Waldner

So if we skip ahead to the next slide? From the perspective of having a full customer 720 perspective, this is what Freya’s actual relationship with the bank looks like, but it’s not something that the decentralized data would lend itself to to allow it to be seen. If you’re in, say, that retail banking division like, imagine pulling up Freya’s information now as a retail banking officer and seeing all this information right, so that seeing that she’s actually got a checking and savings account, but she doesn’t have a credit card or mortgage or insurance. Those are opportunities for cross-sell that we have within this, and we can see that in her extended household.

25:49 – 26:10

Dan Waldner

You’ve got a Lannon, Carly and Andre and Guy, which are children, parents, spouse, ex-spouse, different relationships that fall within that household model, depending on how this firm measures its housebuilding to be able to provide that now holistic perspective of Friday’s world.

26:10 – 26:53

Dan Waldner

And so now you’re going to be able to do all of the more interesting things or as we turn to the offensive side of the coin, which is growing market share and being able to market newer and better products to Freya because at the end of the day, that as much as we like to look at. Profitability and revenue and the financials to be able to to make more money, right? We need to have the ability to provide our customers with the best financial services products at the right time for them. And if we’re only getting to see a fraction of the information at any given time, we are not able to be as effective advocates for our customers as we would if we have all of this information in one place.

27:50 – 28:32

Dan Waldner

Absolutely. And so I just touched upon it a little bit. I touched on two different concepts, right? So the defensive versus the offensive benefits of of of data, right? And so just so that I’m clear on the definition, I view defensive benefits as things that allow us to keep the lights on in our organization or be more effective than what we do. And so operational efficiency is just one regulatory and compliance is to KYC, AML, all those fun processes that nobody really likes. But we want to find a way to do the better so that we don’t get fined by the the regulators or we’re able to do our business and we’re able to do it more effectively, right? That’s the defensive perks in my mind.

28:32 – 29:02

Dan Waldner

And then the offensive side of the coin is really about all the the sexy stuff that we’ve heard over the last few years about all the cool things that analytics can do, right? So growing your bottom line whitespace analysis next best action being able to drive growth of revenue through data arbitrage, right? Knowing knowing things that our competitors don’t know. So defensive offensive. But we’re going to get into some of those now as we as we go through this right.

29:02 – 29:59

Dan Waldner

So the first benefit to to being able to build a compelling business case is making the case that that data makes our customer stickier, right? One of the biggest problems that financial institutions have right now is that banking is largely going the way of a transactional nature. I mean, if we think just on the call all of us, if you think about the last time that you got a mortgage, did you go to the bank that had the best relationship with you? Or did you go to the place that offered you the tenth of a percent lower than every other place? Right. We’re finding more and more customers are going towards that lowest price for lending or highest interest rate for investing wins the day. And that race to the bottom isn’t isn’t benefiting anybody. What we want to do is make customer stickier so that we can generally be able to serve them better. But then also drive higher profitability and higher revenues, right?

29:59 – 30:43

Dan Waldner

So as I talked about before, the goal of every effort should be to focus on timely and customized financial services products to meet the customer’s needs, right? And so you can’t possibly do that if you don’t have the full perspective of what a customer has with you in one place. You can’t necessarily go out and offer them insurance products if you don’t know that you’ve got insurance with them. If the them in this case is the retail banking segment, or you can’t go and make compelling personal personal offers if you only have the B2B side of the relationship in front of you, right? So it leads to very siloed customer interactions and customers that really don’t have a compelling experience with your firm on that.

30:43 – 31:04

Dan Waldner

A little bit on that point, though, because that begs itself if you if you can now start to see all of the things that your customers have with your firm, you can very easily, you know that the low hanging fruit are the cross-sell opportunities that that you have within your life. And again, I’m using the example of a full service firm.

31:05 – 31:45

Dan Waldner

But imagine if every time that your auto loan department received an application that that became a lead within your auto insurance department or, for instance, a mortgage and home insurance right? Or if there’s a wealth event like a child entering college, which your wealth managers should be well aware of. Imagine if that became an opportunity to remarket to that child for student loans of credit or for opportunities for for their first credit card, right? Simple cross sells like that can can dramatically grow the bottom line and often get missed without having all of the data in one place, right? So that can go into your your business case.

31:46 – 32:35

Dan Waldner

And also, you know, how many times have we had bad experiences as customers? When you call and call centers and you have to repeat yourself, you know, time and time again. You know, you put in your phone number on the keypad because it tells you to and then the first thing the agent asks you for is your phone number, which is infuriating, but smoothing out that customer experience to grow the bottom line. That’s that’s key as well for for being able to draw out how we’re going to make customers tick, how we’re going to bring more customers into into the firm as a result of having our data in one place. And that’s not even looking at some of the more traditional quantitative operational efficiencies and regulatory stuff that we touched upon on the defensive side of the coin as I started out earlier in this conversation.

32:36 – 33:03

Dan Waldner

You know, it’s relatively straightforward from a customer data perspective to get compliance or regulatory or EML funding from most firms. So if you’re looking at just building out the business case just with let’s get seed funding to be able to show how this stuff is possible, that often lead lends itself nicely with that initial defensive side of the coin. Let’s focus on that first. Let’s just streamline efficiencies and make our regulatory reporting well. And that will transition to those offensive sides of the coin that I mentioned.

33:21 – 33:50

Dan Waldner

I, you name it, across every line of business. So decreased revenue growth, decreased opportunities, no whitespace analysis. I can guarantee if you don’t have a good, holistic view of your customer, there are regulatory gaps or AML gaps. There are problems that your you are encountering as a result of this that you’re just not aware of. You’re essentially driving at night headlights off, and it’s only a matter of time until you hit something right.

34:36 – 34:58

Dan Waldner

And I think the how is the secret sauce in this conversation because I think many of us on the call said, Yeah, I agree. This is the state of our data right now. And yeah, I agree that centralized data is a slam dunk makes all the sense in the world. I figured out how to justify it a while ago, but it’s the how I just don’t know how to get from where I’m at now, the beginning of this conversation to where I want to be right?

34:58 – 35:55

Dan Waldner

And if you look at the way that it’s been done traditionally right and I use done with with air quotes, the traditional way to do this is then using a rules based approach, right? And the way if you want to think about a conceptual, you’ve got like a Venn diagram, right? You’ve got three circles and they overlap. And the goal is to connect circle a circle B and you do that using a rule. Maybe you do Social Security number and A equals Social Security number and B and examining the B to B to C example, because it’s probably the most straightforward. And then let’s say, if I want to connect the dots from B to C, I’m going to use last name birthdate and zip code, right? So that connects up because for for the sake of argument, C doesn’t have a set number, right? So now we’ve got a rule set for A to B ruleset, for B to C, we need a rule set for A to C, so we’re going to do the same thing birthdate last name zip code. So that ties those three things together.

35:55 – 36:53

Dan Waldner

So we’ve got three data sets and three rules, right? When we throw in a fourth rule now, you need more rules to to tie into this. You need A to B to C, B to C and then A to B, B to DC. And you see ninety rules for one additional set, right? And then the more sets that you layer on top of it, the numbers get big pretty quickly. By the time you get to, I think it’s eight records, eight sets. You need a thousand rules. By the time you get to 10 records, you need a million rules. By the time you get to 20, it becomes several million records or sorry, seven million rules in order to be able to tie all these things together. So it gets really complex to be able to tie these things together so that we’re able to express the individual from a unique identity perspective in one one way in one place, right, that those rules just continue to grow exponentially and eventually it all just breaks down.

36:54 – 37:24

Dan Waldner

And then we’ve discovered this over time that it rules based approach works really well. If you’ve only got a small number of systems, but then it grows too big and there’s no business context as part of it, right? Like, you’re getting a human to write the rules for this, and it’s dumb. It follows the rules. It’s true or false. That’s it. There’s no heuristics as part of it. It just it is what it is from a machine learning perspective, from an alternative approach to this, and it’s one that I’ve taken and implemented myself in my career.

37:25 – 37:53

Dan Waldner

Machine learning gets better. The more data you throw at, the more data sets that you throw at it, the more it’s able to interpolate the information and people to improve and refine its approach. And the way that machine learning works right is that out of the box, it’s got some matching algorithms that are that are kind of baked into the suite of products, and it will go and examine all of the data based on the information that’s as normalized as you can give it.

37:53 – 38:15

Dan Waldner

So if we’re taking the customer data perspective, we’re going to go to every one of our contributing systems and say, OK, so give me first name, last name, birthdate, address city, province or state country as much information you’ve got. But let’s normalize this and let’s say 30 fields across, and let’s try to name and have the same context in all of these fields.

38:16 – 38:54

Dan Waldner

And then let’s process it through the machine, right? And so Tamr will do its best to be able to go and figure out which fields line up, but it’s going to come back and ask a human a set of questions. It’s like, Do these two things line up and. You’ll go through and say, yes, these two things match, you know, they don’t win the first time comes out of the box. It’s kind of stupid and you’ll rerun it. And the second time you run it, it’s getting better. So 50 percent of it will be able to provide, you know, 50 percent of the things will match and 50 percent will fail. And then you rerun it again, and now it’s at like a 90 percent confidence level.

38:54 – 39:38

Dan Waldner

So now now most of the stuff that’s coming out is actually true and it’s getting really good and you have to train it usually about one more time and the whole process of training this this this machine to able to do this usually takes about a week more or less, depending on how frequently your data gets updated and how frequently your data changes as part of this. And by the time you fully trained this thing, it actually replicates the efforts of like that, that what we mentioned in the beginning that that group of 100 interns that go and, you know, fix the data for you, that you cost you a ton of money and you do it once a year when college is out for the summer. But now you’re able to process this thing on the fly and the system continues to get better the more data you throw it.

39:38 – 39:58

Dan Waldner

So the next time that you throw a new dataset in there because you’ve added maybe you’ve got a merger and acquisition where you’ve acquired another company, you process all the information in there. You’re able to process this thing through at an incredibly rapid rate to the point where this now becomes an effective golden source of the customer across your entire organization.

39:59 – 40:38

Dan Waldner

And it’s very good at keeping track of all of the contributing systems so that now you have this kind of barium test to be able to tell you where all your bad data is by virtue of the fact that it’s different or you’ve got this template of what the good data looks like. And if you can go around and change the outputs in all of your systems to fix the data in one centralized way. Now. It’s not to make it sound like it’s a miracle, but it certainly gets you up to an incredibly high percentage rate of accuracy in a very short window of time that gets better as data goes. Yes, this data gets into it, not worse.

41:10 – 41:32

Dan Waldner

And so in the way that I’ve done this in the past, we’ve included third party data sources like Refinitiv, like FactSet, like LexisNexis and Grief and all these external repositories that generally have very good information. And we leverage that to help purify the information as a as an alternative contributing system, right?

41:32 – 42:10

Dan Waldner

So the machine learning model is smart enough to say, OK, so this this information that we know on the street, which is generally considered accurate about an organization, let’s say we’re going to align that to all the different perspectives of the same organization based on what we can identify within your equities trading system, your ex, your commodity system and whatnot. And then we’re going to compress that down and produce this output based on the quality information that we’ve gotten from Wall Street, let’s say. And not necessarily have to rely upon people who have typed in the information from our middle office team, right?

42:10 – 43:00

Dan Waldner

So now you’re you’re this is this sounds too good to be true, but you’re able to put in generally dirty data into the system by humans who weren’t paying much attention and magically, you’re able to get purified good information out of the system. It almost defies the laws of physics, but the machine learning model is good enough to be able to glean those similar characteristics across the entire dataset and align it to a cluster of good, high quality data to say, Oh, you meant this, and I’ll use it, I’ll use a similar analogy How many times have we all been on our cell phones and somewhat distracted and we typed in just complete trash like nothing close to what we wanted?

43:00 – 43:21

Dan Waldner

And then you hit search and Google said, Oh, did you mean restaurant recommendations for the Cambridge, Massachusetts, area? And I have no idea how you were able to get what I was intending based on what I put in. But you’re right, that’s exactly what I wanted. It’s using the same similar algorithms of machine learning just in a different, different context.

43:46 – 44:22

Dan Waldner

And that’s a great question. So from a KYC perspective, we all know that KYC expenditures in in most financial services firms are usually significant portions of the of the budget. And KYC is also one of the most difficult to really hammer down in terms of the the use cases of anti-money laundering, like traditional KYC of asking questions around where you work and how you operate and stuff like that. Those are relatively easy business processes that that kind of are not to be not too difficult to solve correctly.

44:22 – 44:51

Dan Waldner

But then how do we take those KYC and how do we spot the people who we shouldn’t be doing business with under that AML scope? And so there’s a lot of really cool tools out there like Mantis and Hot Scan that that kind of try to do what what machine learning approaches do, but they never quite pull it off quite right. And it usually takes three and four and five pulls of their system to be able to to to get similar behaviors and similar data points, right?

44:52 – 45:48

Dan Waldner

This type of machine learning approach from reconciling individual customer information, presuming business processes capture identification documents accurately and those identification documents are generally considered to be accurate. They can line up, do not do business with individuals significantly faster and in a shorter period of time to the point where you can catch it before you actually forward money or funds or actually go down the garden path as opposed to current processes where you may have already opened up facilities to these do not do not deal entities and have to scramble to cover your regulatory obligations. When you think about what that means to be able to capture potential AML use cases before you’ve let the cat out of the bag. That’s that is significant in many financial services firms.

46:47 – 47:26

Dan Waldner

Yeah. In my in my current role as principal strategist for for traction on demand ideal largely within nCino and I deal largely with financial services firms that implement Salesforce’s FSC platform, right? And so taking it to the next level, the way that I interpret that from from nCino perspective and from Salesforce’s perspective is the elimination of duplicated information in the platform. Every CRM and every lending platform, every loss has the problem of duplicate or dirty data in the system, making it difficult or a laborious process to go through that loan origination process.

47:26 – 48:01

Dan Waldner

So being able to take all of our data that we have as a firm and to have that integrated into your salesforce environment with like an API or similarly into the nCino side of of the Salesforce platform to be able to easily access KYC documentation and similar loan information that may have occurred outside of nCino to be able to draw on the entire ecosystem of data around the customer as a consumer for nCino takes in platform to the next level.

48:01 – 48:38

Dan Waldner

So now I’m not having to ask customers for duplicated information when they’re doing their loan origination that it’s a far more streamlined approach that doesn’t have to ask the customer so many questions, and that the time that the cradle to trade, so to speak for for these loans is significantly shorter. So now where it took 15 days on board a loan, now it’s taken two from beginning to end because all of the the information has been all rationalized, it’s far more streamlined. Similarly, as as a contributor, nCino was able to submit all of the loan origination data into one centralized place and to make that accessible to the broader ecosystem as well.

48:38 – 49:16

Dan Waldner

So that now let’s take, for instance, the nCino is is is coordinating an auto loan on behalf of one of our customers that will go and notify the auto insurance department that there’s an incoming loan. Hey, this might be an opportunity for us to do a cross-sell and be able to to offer them some auto insurance as part of the part of due course. That’s kind of taking it to the next level and growing that bottom line to an activity in a relatively siloed environment, which nCino can be depending on how it’s implemented, bringing that holistically to a different department as an opportunity to grow that revenue. That, to me, is the next level.

49:59 – 50:41

Dan Waldner

I’ve never seen an M&A activity where the consolidation of data and the feasibility of the consolidation of data has ever really occurred before the deal has been signed. It’s always let’s just give this dataset over to technology and guys go and figure it out right from the perspective of let’s, let’s consider the two overlays that we provided before it, right? So you’ve got your traditional rules based approach and then you get your machine learning approach, right? So in the rules based approach, now you’ve got you take let’s say you’re a relatively mid-sized regional bank and you’ve taken over, you merge with another relatively mid-sized regional bank. So now you’re essentially duplicating your entire ecosystem.

50:41 – 51:19

Dan Waldner

And let’s say for the sake of argument that there’s no common systems across the board that there’s no easy way of folding them all together in a rules based approach. Let’s say you’ve got 20 systems as as your firm and 20 systems with the new firm going from 20 to 40 systems in a rules based approach. That’s not two times the rules. That is. Several times, several hundred times more rules as a result of exponentially grows, right? Whereas if you’re looking at it from a machine approach, it’s it’s actually going to make everything far more accurate because now the machine has more data to build its heuristic model on top of it, right?

51:19 – 51:45

Dan Waldner

So in situation A. Everything breaks and you have to you have to find a way to reconcile everything. The data normally takes 18 months to two years to try and reconcile, and it almost always occurs with certain systems will be the victims and certain will be the survivors. We’re going to migrate everybody from one system over to another. And that’s how we’re going to try to reconcile that, which causes millions of dollars and lots of problems.

51:45 – 52:23

Dan Waldner

The secondary way of doing it through using machine learning is we’re going to throw all of this data into one central repository and then we’re going to provide each system in this in this broader now ecosystem, a golden source of all the customers. And it’s actually going to get more accurate because now we have more data to build a more accurate model. And the difference is years and millions of dollars and not going to work too well and lots of projects versus literally throw it all into a blender and somehow a unicorn pops out. And I know that sounds crazy to think that that’s real, but I’ve seen it. I’ve seen it work.

52:38 – 53:10

Dan Waldner

Oh man, I could spend an hour talking about that. There are very good cases to being able to do both. It is. It is a best situation, a best case situation, in my opinion, to keep Salesforce instances separate so that there’s we don’t lose velocity of improvements and growth of each platform while still having a centralized back end to keep all the perspectives of the customer centralized.

53:10 – 53:40

Dan Waldner

If you if you have siloed Salesforce instances, then you have the problem that no one customer record can be seen across all of them. So it’s very difficult to keep track of it all. If you have one master salesforce instance where you have 13 lines of business in this one, yes, you could see everything. But then every improvement that you want to make to the instance has to be vetted with 13 groups to make sure you don’t break something into the complexity of that organization shoots through the roof.

53:40 – 54:20

Dan Waldner

There is that happy medium happy medium is let’s let’s come up with a way to keep all of these orgs in sync so that they can all share the same perspective of the customer and and be able to see the 360 view amongst themselves, but not slow down individual progress. And leveraging an MDM like I would advocate a machine learning based MDM to be able to tie together those perspectives of the customer and then reintegrate it back into each one of the Salesforce. And I think that’s the happy medium, the Goldilocks zone for this to be able to really drive a high rely on on your Salesforce instances.

54:46 – 55:42

Dan Waldner

Absolutely, and and the one thing that many firms don’t invest well in is integration layers, right? Because as if you have a really robust data integration layer, then it doesn’t matter how many Salesforce you have because the data control of that flow freely between them. Right. And so you can go and get a perspective of the Customer 360 without without confining yourself to one or two or five or 50 different works. And I’ve seen I’ve seen the entire spectrum. I’ve seen the really complex organizations that have very good reasons for keeping things very separate and very distinct. And I’ve seen that the one massively complex that nobody can make any different changes to it, and you have a pile of technical debt on this one at work, right? And there’s no there’s no silver bullet other than make sure that your data can move and flow like the liquid it is and not the solid it shouldn’t be.