Latest / Agentic Conversations (formally mlops.community) / Data Platforms in MLOps: Translating Business Goals into Product Decisions // Richa Sachdev // #184

Transcript

Data Platforms in MLOps: Translating Business Goals into Product Decisions // Richa Sachdev // #184 · Agentic Conversations (formally mlops.community)

• 0:00
  My name is Richard Suche, right? Work for JP Morgan Chase and I
• 0:06
  need data reporting and process automation team.
• 0:09
  I love my coffee plain black first thing in the morning and
• 0:13
  as the day progresses I gravitate towards a latte or or
• 0:19
  shaken espresso with oat milk. Welcome back to another.
• 0:24
  ML OBS community. Podcast I am your host Dimitrios
• 0:28
  and today what a wonderful conversation with Richa if
• 0:31
  you've been following along. I am currently on the island of
• 0:35
  Crete and I have been saying there are some funny Greek
• 0:39
  things that happen when you are in Greece and one of them I will
• 0:44
  bring up today before we talk about the excellent conversation
• 0:48
  that I had with Richa is all about why buildings?
• 0:54
  Just get abandoned and then they don't get finished being built.
• 0:59
  And so you've got these concrete structures in the middle of such
• 1:03
  a beautiful scenery, and it's like what happened there.
• 1:06
  They're just going to leave it like that.
• 1:08
  Or a building is almost finished and the rooftop has a bunch of
• 1:13
  like what are they called rebar coming out of the top of it.
• 1:17
  I don't understand. I do not understand what's going
• 1:19
  on here. I heard it is because of
• 1:22
  property taxes if you never finish.
• 1:24
  The house then you don't have to pay property taxes.
• 1:27
  So people say they're going to build another level to the house
• 1:30
  and then they don't pay property taxes.
• 1:32
  But I would love to know for those Greek people that are
• 1:34
  listening or they are out there, I know there's a tonne of Greeks
• 1:37
  in the ML world. What is going on here?
• 1:40
  I do not get it at all. And don't tell me it's because
• 1:43
  of the austerity measures, because I was here in the 90s
• 1:47
  too. And there was still a bunch.
• 1:50
  Of randomly half. Built concrete structures.
• 1:54
  OK, so. Now.
• 1:55
  Let's turn the page, finish that part of my Greek question marks
• 2:01
  that I have in my head and let's go to the conversation that I
• 2:05
  had with Richard. Which we're about to get into.
• 2:08
  So one thing that I really appreciate about Richa and the
• 2:12
  way that she sees things is that she came from the software
• 2:16
  engineering background and she is now very heavily immersed,
• 2:22
  steeped in business. And so she understands and she
• 2:26
  walks me through. This whole entire podcast was
• 2:29
  basically her showing me how knowing the tech is very
• 2:34
  valuable. But the other piece?
• 2:37
  That is more valuable now at her current job in the financial
• 2:42
  sector is knowing the business side of things.
• 2:46
  And so she has to understand very deeply the business goals
• 2:52
  and metrics that they're hitting.
• 2:54
  And then she can translate it into the.
• 2:57
  Product design decisions and justice.
• 2:59
  The technology decisions OI appreciate her talking to me
• 3:04
  about that and how she looks at it and how she.
• 3:06
  Makes decisions when she is. Confronted with them, she spoke
• 3:11
  to us about data platforms. And this is something that.
• 3:18
  It got popular of course when Andrew Yang was talking about it
• 3:21
  a few years ago and I think it almost has been swept under the
• 3:25
  rug a little bit because of LMS. And I get the feeling more
• 3:30
  people will be talking about it and it will become a huge part
• 3:34
  of the conversation soon enough. Maybe by the time this podcast
• 3:37
  is out because. Big Data platform is your
• 3:42
  foundation, and without a strong data platform you can't do
• 3:46
  anything else. It doesn't matter what GT model
• 3:49
  you're using if you really want to have that mode and you want
• 3:54
  to be able to. Use your.
• 3:55
  Proprietary data you need. To have fundamentals so reach,
• 4:01
  it walks us through how. She did that at a few of her
• 4:04
  jobs and I love talking to Richa, so I hope you.
• 4:10
  Enjoy it just as much as I did. And as always, if you thought
• 4:15
  this was useful to share it with one friend, that would mean the
• 4:17
  world to me. All right, let's So alright,
• 4:27
  reach out. Talk to me about your story
• 4:31
  because I know we met probably a few years ago and you were
• 4:35
  coming from Vanguard at the time and I love that you have been
• 4:40
  working in the financial department.
• 4:44
  So I want to dig deep on some of the regulations and all that
• 4:48
  good stuff. I think I might even use
• 4:50
  Vanguard to be honest. And that is another.
• 4:52
  To us and I just want to hear what's your how did you get into
• 4:56
  data? Yeah, absolutely.
• 4:58
  And side note, if you're considering using Vanguard funds
• 5:02
  for investment, do it. If you don't have a 529 for your
• 5:06
  kids, do it Languages. That you're going to give me
• 5:09
  some? Some influencer code like use
• 5:12
  reach a 29 at the end of you know vanguard.com/reach at 29 to
• 5:16
  get 20% off your Vanguard ETF. I am not giving any financial
• 5:21
  advice, I'm just keeping it very high level.
• 5:25
  But how did they get into data? So let me tell you a little bit
• 5:27
  about my story. Then start with data.
• 5:31
  My background is computer science and bachelors, masters
• 5:36
  and I. My one of my favourite subjects
• 5:38
  always was databases, data architecture and at at that time
• 5:44
  when I graduated, data was not big data, it was still at a
• 5:48
  smaller scale. But I still loved visualising,
• 5:51
  analysing data and little did I know the heart. 5-7 years down
• 5:57
  the line I was going to get into the world of data.
• 6:01
  But when I started at a college I worked as a software
• 6:06
  developer. I did production support, I did
• 6:08
  change management, I did development from ground up.
• 6:12
  I was also part of a company that was like in the that was a
• 6:17
  company that was providing service to for TV guides et
• 6:21
  cetera. So thinking about like the API's
• 6:24
  and how we were deploying things.
• 6:25
  So I learned a lot of best practises there.
• 6:28
  Then I went into a healthcare company where I took it upon
• 6:33
  myself to figure out which of the teams that are doing
• 6:37
  analytics. My job was to take an existing
• 6:41
  legacy application and move it over to Java and Spring Boot.
• 6:44
  So that was like again very heavy duty coding is it.
• 6:47
  They were analytics teams that were taking a look at electronic
• 6:50
  medical records. How do we extract data from
• 6:52
  them? How do we move it to Hadoop at
• 6:54
  that point in time? How did we run like Hive and
• 6:56
  Impala queries? And then I got entered into the
• 6:59
  world of cloud at a very early age when cloud was just hitting
• 7:05
  it up, was becoming like a buzzword.
• 7:07
  So I have very, I was lucky enough to have very strong deep
• 7:13
  in software engineering in understanding what does it mean
• 7:16
  to to write code with best practises, version control,
• 7:20
  change management and things like that.
• 7:23
  And then when I entered into Vanguard, I interviewed for the
• 7:26
  role of did engineering and I was the only engineer at that
• 7:30
  point of time. And the department that I joined
• 7:33
  was sort of like a startup in bank at Vanguard.
• 7:37
  So the mandate was for us to establish our core competency
• 7:41
  with data and then then think about hiring data scientists,
• 7:45
  think about doing prescriptive descriptive ML and then going
• 7:48
  into predictive. And that was like the overall
• 7:51
  journey I had for five years while I was at and that was the
• 7:55
  time, the tipping point where I got into data where I got into
• 7:59
  machine learning, that's not what I studied because when I
• 8:02
  was in school that was those were not the courses that were
• 8:04
  available. So a lot of it was like learning
• 8:07
  on the job, but I had very deep roots in technology.
• 8:10
  So at this point of time, if you think of leaders who lead very
• 8:15
  successful machine learning engineering teams that are doing
• 8:18
  Mlof's infrastructure to support all all of these, it's a very
• 8:22
  difficult to find folks who have knowledge of both.
• 8:27
  And yeah, through life through like knocking on the right
• 8:31
  doors, having the right mentorship etcetera, I found
• 8:34
  myself at a hybrid of these two roles.
• 8:36
  So that's why that's how I got into data.
• 8:39
  And then once I got into data, I never looked back and I've had
• 8:43
  like different roles which I can talk to you about.
• 8:45
  But yeah, I'm so happy that I love data from data.
• 8:48
  One day one I'm in there and I continue to make my career and
• 8:53
  and learn as I progress. You thought to yourself, I
• 8:56
  imagine, Wow, there's so many more bugs.
• 9:00
  That I can deal with and data. This is going to keep me busy
• 9:04
  for a long time. Long time.
• 9:06
  You said something there and you kind of glossed over it, but I
• 9:09
  wanna dig in, which is around the journey at Vanguard and how
• 9:16
  you were setting up the foundation and you had this nice
• 9:19
  little motif that prescriptive, descriptive and then predictive.
• 9:24
  Yeah, sure, happy too. Prescriptive is when you are
• 9:28
  working with the business and business knows the data in and
• 9:31
  out and it is sizeable digestible chunks and they can
• 9:35
  tell you what sort of analysis they want to see.
• 9:39
  They have for example the right KPIs or metrics in mind which
• 9:43
  they can say hey data reporting team, operations team, can you
• 9:47
  slice and cut the data for me? Uniform method we can use for
• 9:50
  decision making. Descriptive is that what you see
• 9:55
  in the data, the data science, the data analysts, data
• 10:00
  engineers able to see and just create some rough distribution,
• 10:04
  rough graphs, visualisations on their own and then present it to
• 10:08
  the business and have a dialogue as to which data points are
• 10:11
  important so that those can be then changed into automated
• 10:16
  dashboards etc. So one is like business driven,
• 10:19
  one is data-driven, but it's still what data is trying to
• 10:22
  tell you. What you see it as is protective
• 10:28
  is where you see historical data, you see the trends, you
• 10:33
  see the patterns, you apply your algorithms on top of it.
• 10:36
  You do your tuning, you figure out the right algorithm, you
• 10:40
  figure out what the business high level business problem
• 10:43
  you're trying to solve and then you predict based on top of that
• 10:47
  what could be your potential outcomes.
• 10:50
  So things like that, hey, when a customer call, what would be the
• 10:54
  next topic they would be interested in?
• 10:57
  That's not where I can look in snapshot of time right now and
• 11:00
  say, but I have to look back in time, understand the customer
• 11:04
  journey, understand various data points there demographics or
• 11:07
  transactional history and many other things and say OK, what
• 11:11
  would be the next best message they would want to hear?
• 11:14
  What is the channel they would respond better to?
• 11:17
  Where we where what would be mobile, some other digital
• 11:21
  channel for example or very real life you know situation is going
• 11:28
  to be what is the next best ad I should be showing you on the
• 11:32
  Instagram for example if I look about you or what you viewed
• 11:36
  right now I can tell you something but the Jews is when
• 11:41
  you analyse someones historical trends and data and then you
• 11:44
  predict what is it, what treatment you would respond to
• 11:48
  the best. And so you as a data engineer
• 11:51
  that we're growing in there and setting up these data
• 11:53
  foundations, how did? The system look.
• 11:58
  As you went through each one of these.
• 12:02
  Perspectives that you were looking at data through, Yeah,
• 12:05
  There's a lot that you could unpack here, so I'll try to
• 12:08
  remember. So think about it as like 3 high
• 12:14
  levels operational data, analytical data and data that's
• 12:20
  used for machine learning at a high level.
• 12:24
  So operational data is day-to-day data.
• 12:28
  How many clicks do you have? How many people are responded to
• 12:33
  a particular e-mail, how many people clicked on the like
• 12:36
  button, for example. That's the day-to-day sort of
• 12:40
  things that you can say when you have an offer, how many sign up
• 12:43
  for it, etc. day-to-day. That's more operational in
• 12:47
  nature and that's used to make a lot of, that's very, very
• 12:53
  powerful in making decisions. And the general mode is through
• 12:59
  reports and dashboards etc. Then is analytical which is a
• 13:06
  higher level of operational where you start aggregating data
• 13:10
  on monthly, quarterly, yearly level and start looking at
• 13:13
  trends. Start comparing your historical
• 13:15
  data to now how your your offers or your trends are going, your
• 13:22
  KPIs downward trending upward trend.
• 13:24
  How do you make data-driven decisions at the business when
• 13:29
  you start thinking about powering your machine learning
• 13:32
  models? But historical data?
• 13:37
  Huge sums of data that comes into play for many years.
• 13:42
  They will have always heard that data archival is important
• 13:45
  because it reduces the compute, it reduces the storage.
• 13:48
  But you have to be very mindful that the more the data you give
• 13:53
  to an algorithm. Of course in the Met you're not
• 13:55
  going to give 10 years worth of data, you're it might even
• 13:58
  choke. Like forget about.
• 13:59
  I'm not talking about like Elms and but like the traditional
• 14:02
  machine learning there you have to think about how do you store
• 14:06
  the data and warehousing, how do you surface the data with low
• 14:12
  latency, how do you make the data available for training
• 14:15
  etcetera. And there's always this like
• 14:17
  concept of ETL and ELT. So for operational data, for
• 14:22
  analytical data, if you load the data and then you transfer, you
• 14:28
  work on the transform data. So you work with your IT teams
• 14:31
  and say, hey, do this transformation, create these
• 14:33
  views, create these tables for me.
• 14:35
  You can work offline. For machine learning, it's more
• 14:38
  so as much raw data as possible. You could put in one place and
• 14:42
  create a semantic layer on top of it.
• 14:44
  And the transformation is something that the individual
• 14:47
  teams can do so that the meaning is not lost when you transform
• 14:51
  already the data. So all of it requires like
• 14:55
  different data architecture and you must have always heard of
• 14:58
  like the Raleigh or the semantic layer, the analytical rare.
• 15:03
  So very close attention needs to be paid to that.
• 15:05
  So when I when I said Vanguard, we always had to keep these use
• 15:10
  cases in mind and work with individual data producers as to
• 15:15
  where is the data, what format you're storing the data and how
• 15:18
  are you cataloguing the data. What's key in my is it in for
• 15:22
  example how bad could the history you're keeping.
• 15:24
  So those are the things that are important and the tools itself
• 15:28
  are very different. For analytical operational you
• 15:31
  could use let's see if you're on premise like your traditional
• 15:34
  SQL servers and things like that.
• 15:37
  When you take it to the cloud, of course there are tools that
• 15:40
  are there too. But like for your machine
• 15:43
  learning, there's a logic still flake debts, those are the tools
• 15:48
  that come into play that have that powerhouse of storing that
• 15:53
  much data, are giving you the capability to query it easily
• 15:58
  and then consume it within your model.
• 16:00
  And now there's a concept of like data mesh.
• 16:03
  Everything earlier was centralised when you used to
• 16:06
  think about like big data warehouses and again something
• 16:09
  that was very easy to plug and play for operational and
• 16:12
  analytical. But for machine learning, you
• 16:15
  don't only need the data that's in your department, you need to
• 16:18
  join it with, let's say other things about the customer, the
• 16:22
  marketing data, things like that.
• 16:24
  And then data mission comes into play that the data is
• 16:26
  decentralised and how do you think about like apps, micro
• 16:30
  service do monolith, right. That's the change that we went
• 16:34
  through. Data mesh is that for the data
• 16:36
  where data resides in individual teams and departments and then
• 16:41
  the consumers of the figure out how to like access that data,
• 16:44
  how to join it, how to use it for preferred decision making.
• 16:48
  Like I said, there's a lot to pack here, like on that data
• 16:50
  architecture, the tools, the centralization, the
• 16:53
  decentralisation. And then it gets even more
• 16:57
  complicated when you think about unstructured data, because a lot
• 17:00
  of our data is unstructured now we look into voice data, calls
• 17:04
  again, calls data for example, e-mail data, complain data,
• 17:07
  social media data. So how do we all store it
• 17:10
  together? How do we join that, create that
• 17:13
  thread between your relational databases and your non
• 17:17
  relational databases and then when you're extracting
• 17:20
  information of where do you store it.
• 17:22
  So there's a lot that goes into thinking about not just how you
• 17:26
  design your systems, but your data.
• 17:28
  So that's why, like data architecture is a very nuanced
• 17:32
  feel, and it's. Well, I gotta know.
• 17:34
  I mean you practically. Put the next question on the
• 17:39
  table and I'm intrigued to think about.
• 17:44
  What and how did you do that? Joining these unstructured data
• 17:49
  and the structured and knowing like certain databases have
• 17:55
  their strengths for one and other databases have strengths
• 17:58
  for others or it's just like BLOB storage and you throw
• 18:01
  everything into BLOB storage and try and work from there.
• 18:04
  How do you think about that? And inevitably, you're gonna be
• 18:10
  getting. A million databases I imagine.
• 18:13
  And so how do you like keep from?
• 18:16
  Just adding another database to service A need when it comes U
• 18:22
  yeah so back in the day when we were doing it we were evaluating
• 18:28
  different databases like you know Mongo for example.
• 18:35
  Yeah, because we had a lot of document that documents that we
• 18:39
  were trying to store, but then they were not just like
• 18:42
  challenge of integrating data across different databases, it
• 18:45
  was just a challenge of architecture, security and
• 18:49
  things like bringing those managed services or databases in
• 18:53
  house. So that was a reason why we we
• 18:56
  will put brakes on that. But yes, it is a question of
• 18:59
  like let's say even if we were to bring it in, how were we
• 19:01
  going to join it all? It's a difficult question
• 19:04
  because how do you keep all of them refreshed at the same time,
• 19:07
  How do you make sure that the the primary key or the the
• 19:10
  identifiers are the same And then it's a lot of data is
• 19:14
  loaded with PII. How do you, how do you
• 19:17
  unidentified that in unstructured data and store it?
• 19:21
  It is not an easy problem, there is no silver bullet to it and
• 19:25
  every company does it and a little bit of customise it
• 19:28
  because the rules change, the access controls change etc.
• 19:32
  So when we did it back in the day we used file storage like we
• 19:36
  uploaded a lot of them on the street packet.
• 19:39
  Then we had programmes that were constantly extracting the
• 19:43
  information, parsing and just sizing it down to digestible
• 19:48
  bite size. So that when our algorithms were
• 19:51
  reading the data, they were not reading the entire BLOB, but
• 19:53
  they were reading the most condensed, juiciest information
• 19:57
  that was present in there. And then when we were storing
• 20:01
  the information that we were deriving from it, then that's
• 20:05
  the place where we were making sure that the insights from it
• 20:09
  were being stored in. A place where if we were
• 20:12
  thinking about you as an end customer, Demetrius, and if I
• 20:15
  wanted to store your information about your transaction, what
• 20:19
  pages you visited, etc. The insights from your.
• 20:22
  Paul or your e-mail was also being stored in one place
• 20:25
  itself. If I think about EU, I think
• 20:28
  about you in a 360 frame. So it was more at the output of
• 20:33
  if we if the model had to like ping real time and say hey give
• 20:36
  me all the insights about you, that's what we optimise more
• 20:39
  than the storage layer. Our models were not getting
• 20:42
  refreshed in real time, there was a lot of bash.
• 20:46
  So it left us with the time to perfect the upstream and focus
• 20:53
  more on the downstream where it was more critical to make sure
• 20:56
  that all the data was in one place and easy to find.
• 21:00
  I hope that makes sense. 100% and of course.
• 21:04
  No, I wanna that was that's been your history now you're doing
• 21:08
  equally as cool stuff and you're at Chase, right so.
• 21:13
  Yeah, I would love to go through what's your day-to-day now?
• 21:16
  What are you doing? And we can go from there.
• 21:19
  Yeah. My my work had Chase.
• 21:21
  Of course, there's a lot that I can borrow from my previous
• 21:24
  life, but I must say, it's a lot of fun because I'm in cheese.
• 21:27
  Reward your credit cards, everybody.
• 21:29
  I tell you what they go, oh, you're the one who manages all
• 21:31
  of my data. Yeah.
• 21:33
  So my day-to-day is data reporting process automation
• 21:39
  where remember I talked to you about the operational analytical
• 21:43
  data. So we're facing those quick
• 21:45
  insights to the business for decision making.
• 21:48
  That's what I focus on. But through that I need to be
• 21:53
  very deeply ingrained in the business processes and knowledge
• 21:58
  of the business which are really enjoy.
• 22:01
  That's something that I did not do in my previous life, was very
• 22:03
  front and centre with technology and how the business is going to
• 22:07
  use. But now I have to really
• 22:09
  understand the card business and the reward space.
• 22:13
  Everyone. My name is Aparna, founder of
• 22:15
  Verizon, and the best way to stay up to date with ML OPS is
• 22:18
  by subscribing to this podcast. It's a lot of fun like
• 22:23
  brainstorming ideas because when I come to process automation,
• 22:29
  it's important for me to see that how a business conducts
• 22:33
  work end to end. What are the touch points, what
• 22:35
  are the manual steps, where are the errors?
• 22:38
  How do we bring in resiliency from a control standpoint?
• 22:41
  So my thought process has changed a little bit where I
• 22:45
  have to keep business, you know, front of my mind, but yet keep
• 22:49
  the wheels turning. How do we automate?
• 22:52
  Do we automate based on a workflow with a lot of local,
• 22:55
  local solutions available, do a thing about robotic process
• 22:59
  automation or is it the third category where we're trying to
• 23:03
  automate? How do we listen to calls?
• 23:05
  How do we drive inside? How do we figure out our pain
• 23:08
  points? How do we marry that with the
• 23:11
  customer journey on web or mobile and see where their
• 23:14
  dropouts are and they're calling to reduce the call volume?
• 23:17
  That's in the machine learning side.
• 23:19
  So my journey, I have not been achieved for a very long period
• 23:21
  of time, but my journey has been to figure out different buckets
• 23:25
  in which my work can reside. So there there's been a there's
• 23:29
  a lot of fun, there's a lot of room to make an impact and
• 23:32
  create that growth. But I am a core technical person
• 23:36
  in a business organisation, so I'm also trying to figure out my
• 23:41
  space and make sure that we continue to keep business front
• 23:45
  and centre, yet think about the technology solutions to propel
• 23:50
  IT at the same time. All I heard from everything you
• 23:53
  just said was that you can get me free first class upgrades on
• 23:58
  my transatlantic flights. Is that?
• 24:01
  True. Did I just hallucinate that one?
• 24:04
  You're absolutely correct. You know, I, I, I I like some.
• 24:10
  I couldn't get that for myself. But I can get it for you,
• 24:12
  Dimitri. Yeah, you you can go in the back
• 24:14
  end and tweak the numbers, right?
• 24:16
  That data, it's immutable or is immutable, let's be honest.
• 24:19
  So that is fascinating because I am a huge proponent and I know a
• 24:24
  lot of people in the community are also very gung ho about how
• 24:29
  as a machine learning engineer and even a data scientist,
• 24:33
  anyone who is trying to show value with these models that
• 24:40
  they're putting out. You really have to have a strong
• 24:43
  story on how it affects the business and it seems like
• 24:48
  you're going. At this through the lens of
• 24:53
  let's figure out the business needs before we figure out the
• 24:56
  cool tech that we can throw at it and what we can do with
• 24:59
  technology. You are very correct and that is
• 25:05
  a struggle that I lived and breathed.
• 25:08
  When we were at Vanguard, we were creating a lot of cool
• 25:11
  things. We know what we were trying to
• 25:13
  solve, but it was getting hard for us to translate it taped to
• 25:17
  the business of data literacy was a huge struggle.
• 25:22
  The other struggle was mania times folk sayings about let's
• 25:25
  bring the the biggest and the baddest and the strongest data
• 25:29
  scientists. And then data many times is an
• 25:31
  afterthought. And that also leads to a lot of
• 25:35
  latency and speed to market because if your foundational
• 25:39
  data is not correct, if we have a thought about the business
• 25:43
  needs and how you're going to be using the data, then all is
• 25:47
  lost. And this opportunity and cheese
• 25:50
  is perfect because it is giving me a chance to understand the
• 25:53
  business problems, understand the user journeys, marry that to
• 25:57
  the data that we have currently, figure out how do we want to
• 26:00
  store that data for best optimization.
• 26:03
  And then we're going to continue to incrementally deploy more
• 26:10
  nuanced algorithms, machine learning techniques to drive
• 26:14
  insights. Right now we're focused on the
• 26:16
  basic heuristics to learn from it, very simple data analysis, a
• 26:21
  little bit manual human in the loop.
• 26:23
  But as we go, we want to make sure that we have a very strong
• 26:27
  business story, very strong business impact.
• 26:31
  They're able to articulate and show progress and continue to
• 26:36
  put in efforts into our data strategy, tooling strategy
• 26:41
  resource and bend strength strategy to be able to go full
• 26:45
  steam. O as you're breaking down
• 26:48
  projects and you're trying to figure out, hey, can I add value
• 26:51
  here with? Whatever the data may be, maybe
• 26:54
  it's just analytics and it's a report, or maybe it's a full on
• 26:59
  machine learning model. What are some things that first
• 27:03
  come to your mind as far as? Metrics that you want to track
• 27:08
  and how you're able to make sure that.
• 27:12
  Those metrics are important and you can track them.
• 27:15
  Are you talking in terms of the data metrics that we want to
• 27:18
  track or are you talk about business?
• 27:20
  Metrics think it's more business metrics to be honest.
• 27:23
  It feels to me like at the end of the day, what you want to do
• 27:27
  right is show the ROI on your projects and you want to be able
• 27:31
  to say, look, we did all this work.
• 27:35
  It was a pain in the butt. We had to create new forms of
• 27:38
  capturing data. We had to bring on this new
• 27:41
  pipeline and this is all this stuff.
• 27:43
  But at the end of the day. I'm guessing your.
• 27:47
  Boss or your boss's boss? They don't really care about the
• 27:51
  cool data transform that you're doing in this new pipeline.
• 27:55
  They just want to hear how much money is it saving me or what
• 27:58
  are the things that they're looking to hear?
• 28:00
  Is it? How, how are we reducing churn?
• 28:04
  How, what are the big business metrics that you're trying to
• 28:07
  hit so that I guess how many people are getting free flight
• 28:12
  upgrades and that might be one business metric that you're
• 28:15
  looking at. So there are different parts of
• 28:22
  it. So without like getting too deep
• 28:25
  into our business, but the different journeys, right, what
• 28:29
  at a high level is you earning points at a high level as you're
• 28:33
  using and the whole management around it in terms of how many
• 28:41
  were redeemed, how many were combined, reducing the fraud
• 28:45
  around that um and how many people had easier interaction
• 28:53
  with our ultimate reward system that we we are trying to measure
• 28:58
  through through the call data, the reduction in calls or
• 29:02
  complaints et cetera. So those are like the higher
• 29:06
  level things that I can, you know, full transparency share
• 29:10
  here, but there's a different angle I can talk to you about is
• 29:14
  there's a component of process automation as well, right?
• 29:18
  Yeah. And it's in IT and IT and it
• 29:21
  effects our storyline at the end of the day because if you
• 29:25
  automate, we'd reduce the number of errors, we reduce the latency
• 29:30
  and how and when and how people are awarded their points or
• 29:35
  their points are clawed back or they get like the tax associated
• 29:40
  with it. Cetera.
• 29:41
  So at the very foundational level, if we make our processes
• 29:46
  a lot more automated, if they are governed and we have the
• 29:50
  right metrics to capture that in terms of before, how long did it
• 29:54
  take for us to run the task end to end when we've automated, how
• 29:58
  are we reducing the time to run it?
• 30:02
  Uh, how often did we have to accuse it or how error prone was
• 30:06
  it? Have you automated the quality
• 30:08
  cheques or in terms of were there any control or audit
• 30:12
  findings that we are closing? So those are the sort of things
• 30:16
  that are also front and centre and the reason is because the
• 30:21
  end customer might not be seen yet directly.
• 30:24
  It might not be a tangible impact, but it flows through the
• 30:29
  system and will have a tangible impact.
• 30:32
  So those are the sort of the things also at the automation
• 30:35
  level that my team is very involved with and the senior
• 30:37
  leaders are very interested in furthering out because it does
• 30:40
  affect our bottom dollar at the end of the day.
• 30:43
  It feels like the wise way of doing it and not getting caught
• 30:48
  up in other things and seeing like wow OK, these are really
• 30:52
  the things that matter and this is what we need to think about
• 30:57
  and the end user experience is the one that we want to optimise
• 31:00
  for. And then there's all these other
• 31:02
  parts. That the end user probably
• 31:04
  doesn't even think about, or doesn't need to think about,
• 31:07
  because again, you're optimising for that end user experience,
• 31:10
  and so do they need to be. Brought in on this part.
• 31:15
  Well, maybe, maybe not. And I imagine you you experiment
• 31:19
  with those different kinds of things.
• 31:22
  When you are setting up. This is because.
• 31:28
  I guess there's two pieces to this that.
• 31:32
  You have to go through this process and figure out.
• 31:37
  What you want to set U and how you want to set it U.
• 31:39
  And once you've identified something that you need to set
• 31:42
  U. What are some of the key
• 31:45
  tradeoffs that you're looking at right now and that you're
• 31:48
  thinking about? And again, this can be.
• 31:54
  Since you are, you have this vantage point of being so close
• 31:58
  to the business. Yet.
• 32:00
  Having the understanding of the engineering side, it feels like
• 32:05
  you can think about a lot of different tradeoffs, whether
• 32:09
  it's a technology trade off or it's just hey, if we have a
• 32:11
  different user experience that could be a trade off.
• 32:14
  So how do you look at those different trade-offs right now
• 32:18
  in your day-to-day? The way I think about this is a
• 32:21
  different lad. One is the technology that's
• 32:26
  available for us and a lot of it is legacy.
• 32:31
  A lot of it is also Morton. We are at this juncture where we
• 32:35
  have to interface between the two, so the trade off always is.
• 32:43
  Should we rush to push everything into cloud?
• 32:47
  Modernise it? Or should we figure out first
• 32:51
  how do we get the reports and the important decisions out of
• 32:55
  the business as soon as possible while we continue to work on the
• 32:59
  Morton tool stack? No surprise time is limited.
• 33:04
  Resources are limited. As a leader, I'm always grappled
• 33:10
  with the decision making or the catch 22 of where do we use our
• 33:15
  resources to the best of their potential?
• 33:18
  Should we pivot? Should we not get carried away
• 33:22
  by the the future state yet focus on the the state that
• 33:26
  we're in right now? And that's a decision that I
• 33:29
  cannot make in isolation. I have a lot of stakeholders
• 33:33
  that I work with, my product partners, my product teams that
• 33:36
  rely on the decisions that we give to them.
• 33:39
  So there needs to be a recognition of all the players,
• 33:42
  there needs to be a recognition of and to involve them for all
• 33:45
  of the decision making. Then the bigger the company, the
• 33:51
  more spread out the work can be. So our data modernization work
• 33:54
  has a lot of players, a lot of cooks in the kitchen.
• 33:58
  So sometimes it gets very, very hairy to figure out what
• 34:04
  progress is being made, what team is doing what, and what
• 34:07
  does the ownership and accountability lie?
• 34:10
  Um, so bringing it all together is like another very arduous
• 34:16
  task that takes away from time. So even though it is not
• 34:19
  something that is on my, is in my purview to solve, but it
• 34:23
  directly affects me to the trade offers I see it.
• 34:27
  Should I talk about it? Should I bring it up or should I
• 34:30
  let when blow and see where it's going?
• 34:34
  But as a senior leader, it is incumbent upon me to fight to
• 34:38
  identify room for growth and opportunities and bring all the
• 34:44
  right parties together for like decision making from the
• 34:48
  business side of view. Like I was mentioning, we have a
• 34:50
  lot of legacy tools and the knowledge is very centralised
• 34:54
  with some folks. So how do we think about
• 34:56
  breaking the the monolith and dividing it into domain driven
• 35:01
  architecture, microservices driven architecture, Think about
• 35:05
  like data with it as well. From a data architecture, data
• 35:09
  perspective that we talked about, what are the pros and
• 35:12
  cons of it like having that as an open dialogue when you do
• 35:16
  like data centre migration. That's another thing that we
• 35:19
  constantly think about and we evaluate the pros and cons and
• 35:23
  what timeline should we go about what road map, item should we
• 35:26
  have. So those are like some of the
• 35:29
  things that you have to bear in mind.
• 35:30
  So it's not just technology decisions, it's not just
• 35:34
  business decisions, they have to go hand in hand because
• 35:38
  technology decision drives business and vice versa.
• 35:42
  But which battle to pick, when and what to focus on is
• 35:46
  something. It's part art, part science.
• 35:49
  But the the message is at the at the scene, The more senior you
• 35:54
  are, the more interplay you have to understand and then make sure
• 35:58
  that you have the consensus from right parties.
• 36:00
  Understand the impact on every party and then do do your best.
• 36:06
  Yeah, it feels to me like. If you can fix it without having
• 36:11
  to go into technology, that's probably the easier route
• 36:16
  considering if you have. A lot of legacy and you don't
• 36:20
  know how long it's going to take to dig in and change the
• 36:24
  processes around the tech and everything that goes with it.
• 36:27
  And ideally. Not break anything along the way
• 36:31
  if you are trying to mess with the tech.
• 36:35
  So it's like it it's almost I could see it being like you
• 36:39
  would learn more towards can we find another solution that is
• 36:42
  not trying to do something with tech.
• 36:45
  And then after you've exhausted your creativity, there you go
• 36:49
  down the rabbit hole of OK, if we were to change the tech on
• 36:53
  this, what would it look like? Yeah.
• 36:54
  And that's true because in our world, our systems directly
• 36:59
  affect the end user. If you don't see the rewards you
• 37:02
  have, you go on Amazon, you're not able to make a purchase.
• 37:06
  We are going to affect the end customer like this.
• 37:10
  It's not like few days later you're going to hear about it
• 37:13
  and we can remediate it. It's right there right now.
• 37:16
  So it's like fight or fight. So we have to like fix it in in
• 37:22
  parallel. We have to keep thinking of how
• 37:24
  do we advance in our technology, how do we advance in our
• 37:27
  processes. So Yep, that's definitely a
• 37:31
  challenge. And like I said that time is
• 37:34
  limited, resources are limited and budget is limited.
• 37:37
  So we have to be pragmatic about how we go about it.
• 37:40
  Always, yeah. I wonder also though, at the
• 37:43
  same time so. How much do you think about?
• 37:49
  This idea of. After I just said, you know, you
• 37:53
  try not to touch the technology because if you can do something
• 37:57
  on a different level, then it's probably going to be faster to
• 38:01
  implement and and. Potentially you can have a
• 38:05
  quicker win, but at the same time I wonder about.
• 38:08
  How you analyse and maybe you have people that will go do full
• 38:13
  on reports and maybe this is what the product people, the
• 38:15
  stakeholders you were speaking about some of the other
• 38:17
  stakeholders? Are talking about with this
• 38:22
  legacy tooling that we have, it takes us way longer than it
• 38:25
  should to do something and if we had a new tool or if we had a
• 38:30
  new way of doing this on our tech side, then we could cut.
• 38:36
  The time of implementation or the time that our engineers are
• 38:40
  spending on this problem down significantly and that?
• 38:45
  Because of the cost of labour is going to increase our ROI.
• 38:50
  Yep. And those are conversations we
• 38:53
  do. And our product partners have
• 38:57
  their own corresponding duck teams who understand these
• 39:01
  systems. Our senior leaders, even above
• 39:04
  me, understand the pain points of this system.
• 39:07
  And we always come together to talk about what does our future
• 39:11
  look like? What does the modernization of
• 39:12
  our data, modernization of our tools and technologies look
• 39:15
  like. And if it is something that's
• 39:17
  going to be way too cumbersome to pull apart and create into
• 39:22
  different functions or domains, let's not mess with it.
• 39:26
  Let's try to create something lightweight on the side and then
• 39:30
  run them in parallel. If they're working, then unplug
• 39:34
  from the first legacy one and plug it to the next one.
• 39:37
  But yes, the this conversation or this thought is just not with
• 39:41
  me. It is with our product, tech
• 39:44
  partners, senior leaders, because we do see that it's not
• 39:48
  just the time it takes to get it done, but how does the data
• 39:51
  flow. Sometimes it's so hard for us to
• 39:53
  create a story, the data, because the data gets exchanged
• 39:57
  with a lot of different technologies or solutions or
• 40:00
  platforms so as to speak. So reconciling them gets hard as
• 40:05
  well under and with any legacy there is tech debt and then any
• 40:10
  organisation has to grapple work with that tech debt while
• 40:14
  continuing to innovate and that is something that we do as well.
• 40:19
  So the big question here is, I've heard rumours that some
• 40:25
  banks still operate with Cobalt. Is does Chase use Cobalt
• 40:30
  anywhere in their code? I'm sure some some mean frames
• 40:36
  are probably. Hard to get rid of.
• 40:39
  Talk about job security, right? Exactly.
• 40:43
  They're probably the cobalt engineers, probably getting paid
• 40:46
  a good chunk of cash to stay there and babysit that thing.
• 40:50
  Probably. I wouldn't be surprised.
• 40:52
  I wouldn't be honest, honestly. And I don't even know if it's
• 40:55
  just like bangs. And I think a lot of big
• 40:58
  institutions are going to have that financial industry for
• 41:01
  sure. But yeah, I mean, I'm, I'll be
• 41:03
  surprised if there are other industries who have completely
• 41:06
  washed their hands off of mainframe and cobalt and all
• 41:09
  that fun stuff, all that fun stuff.
• 41:11
  Yeah, I was expecting all with. You.
• 41:14
  That's a good question, Demetrius.
• 41:16
  Yeah, so well, this has been awesome, Richa.
• 41:19
  I appreciate you chatting with me and helping me to understand
• 41:24
  really like how you think through things, because that was
• 41:27
  the coolest takeaway that I've got from this conversation, so.
• 41:32
  Appreciate it. Thank you.
• 41:33
  Yeah, Thank you for giving me this chance to talk to you and
• 41:38
  to share my my day-to-day life and what I learned, what I don't
• 41:42
  learn, and. And it's going to be a work in
• 41:46
  progress at every different role, every different
• 41:49
  department, every different company, there's going to be
• 41:51
  something new. Sometimes you focus on stocks,
• 41:53
  sometimes you focus on business. But I call it honestly enjoy to
• 41:57
  become a go as a wholesome leader, that it's important to
• 42:01
  be able to connect the two very well.
• 42:04
  That 360 view is huge. Yep, and it is.
• 42:09
  It allows you to leverage so much more, so many more
• 42:14
  insights. Completely agree.
• 42:22
  Hey, this is Mike Gold, also Cofounder and CEO of Tecton.
• 42:26
  ML OPS community is the best way to stay in the loop on the
• 42:29
  latest ML OPS news and best practises.
• 42:32
  It's also a great way to connect with experts and get support
• 42:35
  from an amazingly helpful community.
• 42:37
  Subscribe and stay in the loop.