Latest / The Retail Podcast / Nagarro’s Rahul Mahajan: AI Demand Planning, Knowledge Graphs & Humanized Personalization

Transcript

Nagarro’s Rahul Mahajan: AI Demand Planning, Knowledge Graphs & Humanized Personalization · The Retail Podcast

• 0:00
  And that's the opportunity for businesses to think differently
• 0:03
  to also start exploring from a product provider, how do I
• 0:06
  become an end to end extended product to a service providers
• 0:10
  and stuff like that. Hello, and welcome to the Retail
• 0:23
  Podcast. Now we're in the series of
• 0:26
  meetings that should have happened, the NRF that didn't,
• 0:29
  just because of the sheer scale of NRF.
• 0:32
  I'm joined by Rahul Mahajan, who is Global CTO, Vice President at
• 0:38
  Is it Nagaro? Is that right?
• 0:42
  OK, fantastic. I'm winning.
• 0:44
  So fun and a thumbs up. So we're going to have a, a
• 0:47
  general discussion, but I I'm going to learn as much as you
• 0:49
  are because sometimes I'll get an invite saying, Hey, Nagaro
• 0:53
  doing this, this and this and AII think that's super
• 0:55
  interesting. And, and I'll say let's get
• 0:57
  let's get them on the show to have a, a chat.
• 1:00
  So roll. So if you don't mind, tell us
• 1:01
  why don't we start with you? What do you do at Nagaro?
• 1:04
  And were you at NRF by the way? Yes.
• 1:08
  So maybe I spend a minute about the about Nagaro, some of you
• 1:12
  may know, some of you may not know and then I'll absolutely
• 1:16
  come back to the question of that's OK, Alex.
• 1:18
  Perfect. So Nagaro is a global digital
• 1:21
  engineering organisation, 18,000 people and a billion euro
• 1:26
  revenue company listed in Frankfurt.
• 1:28
  We are in the business where client perceive us for complex
• 1:31
  engineering and especially clients love to work with us
• 1:34
  across industry. We are fairly diversified.
• 1:37
  So we work with surely retail, big part of what we do, but we
• 1:39
  also work with the travel, banking and so on.
• 1:42
  Clients love to work with Nagaro when it comes to complex
• 1:45
  engineering, when it comes to modern transformation, fast
• 1:48
  moving problems, agile, fast moving stuff.
• 1:52
  So we are perceived to be a company go with one, which is
• 1:55
  cool in terms of its engineering, its culture and and
• 1:58
  it's fast moving ways of working on complex problems.
• 2:01
  But that's a little bit about quick thing about Nagaro and,
• 2:05
  and, and, and back to the stuff in, in NRF, we, we had a very
• 2:10
  interesting event. We, we got to meet some amazing
• 2:13
  industry leaders across the spectrum, retail, digital
• 2:16
  transformation, consumer experiences.
• 2:18
  This is big part of what we do. Also to briefly introduce
• 2:22
  myself, I'm involved in the organisation as a technology
• 2:26
  leader in the global CPO capacity, in the global CPO
• 2:29
  capacity also. And hence I get to work with
• 2:32
  some amazing clients, tier one clients, all kinds of clients.
• 2:35
  But also I'm responsible from a business perspective for profit
• 2:38
  and loss with some amazing strategic clients.
• 2:41
  So I run the digital transformation portfolio across
• 2:43
  the organisation as an horizontal and I get to work
• 2:46
  with some amazing retailer, some amazing fashion companies,
• 2:49
  commerce companies, banking companies, automobile companies,
• 2:52
  CPG companies, and this always amazing stuff to learn from all
• 2:57
  of these industries given the rate that these industries are
• 3:00
  moving is different. So there's always stuff to
• 3:02
  learn. So that's a quick introduction
• 3:05
  about the company, about me individually, 15 odd patients.
• 3:08
  A lot of this is in the topics that we could be of our interest
• 3:12
  today in retail, in personalization data and in
• 3:15
  marketing and statistics around sales and stuff.
• 3:18
  And this is a reflection of how is a company from an innovation
• 3:21
  perspective, where are we moving, what kind of solutions
• 3:23
  are we creating? In fact, we just filed A
• 3:26
  provisional application in India.
• 3:28
  We'll talk about it around humanising personalization.
• 3:31
  This was also a big theme for us in NRF.
• 3:34
  We were there talking about how shifting personalizations are
• 3:38
  happening. Humanising the whole
• 3:39
  personalization aspect is huge part of what AI today can
• 3:42
  deliver. This is also what the part of
• 3:44
  what to do with some of our amazing year one planes.
• 3:47
  I think with that, I'll give it back to you Alex for.
• 3:51
  Well, what's the keeping it simple?
• 3:53
  What's the number one outcome you feel you can help clients?
• 3:57
  Where can you drive them? Like what?
• 4:00
  What are the if I'm the CEO of whatever retailer your, your,
• 4:05
  what outcome are you going to create for me?
• 4:08
  Sure, it's, it's a, it's a heavy question, but let me take some
• 4:11
  one or two examples and hopefully that should answer
• 4:13
  your question. So we are working with one of
• 4:16
  these amazing CPG companies in the larger beauty fashion space,
• 4:20
  an amazing company. They are working with us on the,
• 4:22
  on the, on the demand planning side, on the supply chain side.
• 4:26
  And and the, the leadership came to us, the ecosystems has
• 4:29
  evolved. We are companies are not just
• 4:32
  selling on retail outlets or not just selling on commerce.
• 4:35
  It's a fairly distributed eco system that companies to operate
• 4:39
  in. So for the CPG company, they
• 4:41
  have their own retail stores, buy brands, some of the brands
• 4:44
  of their own stores, some of the brands selling in 3rd party
• 4:47
  outlets like 7 Elevens and so on.
• 4:50
  They also sell through commerce. They first party commerce, but
• 4:53
  they also sell through marketplaces, which are big
• 4:56
  Amazons and then ebays and and then some focused industry
• 5:01
  segment specific commerce players in different countries.
• 5:03
  But they also sell through third party retail, e-commerce and,
• 5:06
  and, and, and it's an amazing complex stuff.
• 5:09
  When you talk to these clients now managing demand and and
• 5:13
  making sure on supply chain side you have right amount of orders
• 5:16
  in this complex ecosystem is not trivial.
• 5:19
  Classically, companies would have gone to some big daddy
• 5:23
  softwares. I don't want to name anyone
• 5:25
  here, but but there are established existing
• 5:28
  contemporary players. What Arrow came involved as kind
• 5:32
  of now answering your question from a value perspective, how
• 5:35
  you can use today AI to kind of look into demand?
• 5:39
  Demand is an attribution of what promotions you are running.
• 5:42
  Demand is as an attribution to what historical sales and the
• 5:46
  classical stuff, purchase patterns and stuff like that.
• 5:48
  Demand also has drivers coming in all the way from competition.
• 5:53
  But now in case of marketplaces like Amazon's, suddenly there's
• 5:57
  a price drop by a competition or there's a season coming and you
• 6:00
  see an A spike in your sales and, and, and their seasons
• 6:04
  changing. There are influencers talking
• 6:06
  about your products, about your skin care products, about your
• 6:10
  fashion products on and and those influences messages change
• 6:15
  your demand. In this kind of complex
• 6:17
  scenario, surely you can go and run auto regressive forecasting
• 6:22
  model and then create a fancy dashboard and live with it.
• 6:26
  The problem is that the accuracy for this tier one company was
• 6:31
  fairly low and, and, and, and hence there was a need of by
• 6:35
  SKU, by category. How can we look at demand in
• 6:39
  very unique ways? And that's where the, the
• 6:42
  promise with AI was that for every SKU, can we have a unique
• 6:45
  model? Can we have these factors
• 6:48
  dynamically picked in depending on what significance these
• 6:51
  factors have on specific categories and so on.
• 6:54
  So that's a big shift, a use case which nobody would have
• 6:57
  talked about like 3 years back because the channel complexity
• 7:00
  was not this complex. And then and, and trade planning
• 7:02
  was primarily limited to trade management, largely the trade
• 7:07
  optimization TPO was not tightly integrated into it.
• 7:11
  Even the dealer signals were kind of missing.
• 7:14
  But today, that's the opportunity.
• 7:16
  And it's not about just deploying some AI models.
• 7:18
  The another big opportunity, Alex, for us is how do we
• 7:22
  humanise back to the NRF theme. Also, how do we humanise the
• 7:25
  consumption? Can we make demand planners talk
• 7:27
  to systems in natural languages? Can I go and talk to my system
• 7:31
  using generative AI? Hey, just tell me what all SKU's
• 7:34
  are ageing in my in my larger region and, and canny I come
• 7:38
  back with to me have been a beautiful rendered widget.
• 7:41
  Here are top five products which are actually ageing.
• 7:45
  And here are some great promotion strategies that you
• 7:47
  can start considering all the way from Bobo to to price
• 7:50
  discounts and stuff like that. And that's the opportunity.
• 7:53
  A classical contemporary system just cannot do this in, in near
• 7:57
  to real time. But that's on the B2B and
• 7:59
  enterprise side. Let me shift my gears to
• 8:02
  consumer facing experiences. The promise again is and then
• 8:06
  one of the stuff that we have recently filed a patent for also
• 8:09
  is on using generative AI to generate advisories. 2 big two
• 8:14
  or three big trends that I'm observing in my leadership
• 8:17
  conversations with some of these big clients.
• 8:20
  One, it's not just about products anymore.
• 8:22
  Definition of products is changing.
• 8:24
  It's not just first party products.
• 8:26
  Clients want to consider their partner products and not just
• 8:28
  products, clients also want to consider the partner services.
• 8:32
  So goal is to get embedded into the lifestyle beyond
• 8:36
  transactions. Can you get embedded into the
• 8:37
  lifestyle at an emotional level, at a humanising level?
• 8:40
  So for example, if I'm buying an automobile and I want to
• 8:44
  customise it given the terrain and given the vacation, can AI
• 8:48
  help me connect to my accessory partners, to my service
• 8:51
  partners, to my insurance partners and services and create
• 8:55
  a a very simple consumption view out of it?
• 8:59
  And that's the promise with the shifts around ecosystems of the
• 9:04
  shift around services. How do we humanise this whole
• 9:08
  thing? And you cannot do this, Alex,
• 9:10
  without, for example, without having some of the new
• 9:12
  capabilities like Agentech. In past, what we would have done
• 9:16
  is typically go to service based architecture written some
• 9:20
  workflows. Today, if your ecosystems are
• 9:22
  getting plugged in, insurance provider is coming in, a service
• 9:25
  provider, a door service, a service provider at your
• 9:28
  doorstep is coming in. How do you write these
• 9:30
  workflows? You cannot write these
• 9:32
  workflows. You cannot wire these workflows.
• 9:34
  What you can rather do is today is go on agentic systems, use
• 9:38
  agentic systems, deploy these workflows in a way that they can
• 9:41
  on in runtime learn and adapt and stitch these additional new
• 9:46
  services in a marketplace type of a pattern seamlessly.
• 9:49
  So suddenly if you have a doorstep service provider coming
• 9:53
  to your door to provide some services, automatically those
• 9:56
  services get integrated into a workflow and that workflow then
• 10:00
  gets converted to a humanised consumption format.
• 10:03
  And, and you cannot take a, you cannot afford a downtime as a
• 10:07
  client, as a retailer that let me take a downtime, let me
• 10:10
  integrate those services, let me write those workflows, go back
• 10:14
  and do the engineering for three months and come back.
• 10:16
  These providers are coming in on the fly again, technology do it.
• 10:20
  That's the opportunity for all of us.
• 10:23
  And that's exactly the kind of engineering shifts, Alex, we are
• 10:27
  seeing with some of our amazing tier one clients.
• 10:30
  These are things they're pushing, pushing for example, a
• 10:33
  pharmacy company in retail context, they just don't want to
• 10:36
  sell medicine anymore. They want to sell lifestyle
• 10:38
  services, which mean diet services, Wellness services.
• 10:42
  And imagine these kinds of contexts where suddenly from a
• 10:45
  transactional product context, you shift to a lifestyle
• 10:49
  emotional service provider in the larger ecosystem context.
• 10:53
  For this kind of a context, you need a very different
• 10:55
  architecture, very different kind of an AI and, and, and
• 10:58
  hence of those things. Now these are the values that we
• 11:00
  are promising, Alex, to your point to the leadership of some
• 11:03
  of these organisation which are willing to cope with us and
• 11:06
  become futureity before they get disrupted by an online
• 11:09
  marketplace or an online SAS provider.
• 11:13
  So I mean, if we were going to break that down into its
• 11:17
  components, obviously you, you'll build a unique AI model
• 11:21
  for your, your client, because what comes across there is that
• 11:25
  your ability to execute your engineering prowess is obviously
• 11:30
  what you're strong in. Where does the, the, so the
• 11:33
  patents that you hold or the patents that you're, you've
• 11:36
  applied for, how does that come into this?
• 11:38
  What are you actually patenting? Is it the AI model itself?
• 11:43
  And then where does that AI model sit?
• 11:45
  Is it tied to, I don't know, Microsoft or or ChatGPT or
• 11:50
  Anthropic? What, what can you just expand
• 11:53
  on that bit? Sure.
• 11:55
  I'll be brief here because this gets super technical, but just
• 11:59
  to talk about it, there's amazing opportunity, Alex, from
• 12:02
  us for all of us, all of the engineering community and
• 12:05
  solutions and consulting providers to think differently.
• 12:08
  We went back to think of data structures from scratch because
• 12:11
  AI can't consume structured data, video data, multimodal
• 12:14
  data and give you advisories. So for example, if an influencer
• 12:19
  is talking certain, giving some feedback on product, a good
• 12:23
  feedback, let's say, or a certain, a certain way you have
• 12:27
  to use the product or apply the product or, or, or around
• 12:30
  product related scenarios. How can you bring that stuff
• 12:33
  real time into your subsequent advisories, into loyalty and so
• 12:37
  on, which means you need a different kind of data
• 12:39
  structure. We are heavily using knowledge
• 12:42
  draughts. Knowledge draughts has amazing
• 12:44
  promise. I'll keep moving because
• 12:46
  otherwise I'll keep speaking. So knowledge draughts is 1 big
• 12:50
  amazing kind of a promising thing We are exploring.
• 12:52
  Second and a huge thing, we are vectorizing data at an amazing
• 12:57
  speed. Now data has been sitting
• 12:59
  classically inside structured databases or or recently.
• 13:03
  No SQL databases, big databases. But that's not something what
• 13:07
  languages can consume directly. Language needs a semantic
• 13:11
  substitution, so we are vectorizing the databases.
• 13:14
  We are writing new semantic layers.
• 13:16
  That's second big shift in the way we are now thinking of
• 13:18
  solution, because language has capability to talk to data,
• 13:22
  language has capability to even talk to AI.
• 13:24
  Language has capability to talk to us as consumers.
• 13:27
  So it has all that. You can use language to go
• 13:33
  either ways, which means you need a strong semantic
• 13:36
  catalogues to complement your architecture.
• 13:38
  So that's point #2 in terms of where new solutions will have
• 13:42
  novelty point #3 the governance. Now you need this big explain
• 13:47
  ability around some big decisions happening in your
• 13:49
  demand planning, all in terms of advisories.
• 13:52
  We need to go back, double click and see what model ran in, what
• 13:56
  version of data ran in, what kind of observability was it,
• 13:59
  who approved it, what data version, what lineage from a
• 14:03
  master data management perspective and so on.
• 14:06
  Who had the access right on this kind of a data.
• 14:08
  And then you have all data privacy, which is GDPR and CCPA
• 14:12
  and and India data protection and so on.
• 14:15
  So that's the third big opportunity around AI
• 14:17
  governance. By the way, it's not just about
• 14:19
  security and data privacy. How does AI even talk in a human
• 14:23
  friendly format? There's a huge opportunity to
• 14:26
  make AI not talk like a computer agent, computational agent, but
• 14:30
  rather talk like how a marketing person will talk.
• 14:33
  For example, in beauty industry, a dry hair can be said as rough
• 14:38
  hair, dull hair and so on. So brand and marketing talk in a
• 14:43
  certain way, but computational agents, they typically would
• 14:46
  just talk in in standard languages.
• 14:48
  Now, how do you, Tony, how do you change the tone so that it
• 14:52
  speaks like how a brand would like to talk?
• 14:54
  So that's another part of the governance and that's governance
• 14:57
  to me, the third big area of where novelties can be created.
• 15:01
  So imagine the amount of novelty opportunity that we have on.
• 15:05
  And and then last but not the least, you talked about Entropic
• 15:08
  LLMS from open AI clients today wants to build systems because
• 15:13
  innovation is happening. Alex, we all know in A at
• 15:15
  amazing speed in January, you could be seeing Gemini doing
• 15:19
  better. In February it could be Entropic
• 15:21
  doing better, in March it could be again GPD doing better.
• 15:25
  And and who knows, in April some, some other company in some
• 15:28
  other country may be doing better clients, at least when I
• 15:31
  talk to them, they want to externalise their system so that
• 15:34
  there's a certain amount of interoperability.
• 15:37
  They can predict use X tomorrow, they can trace the same thing
• 15:40
  with Y provider and Z provider and so on.
• 15:43
  So externalisation means you can't just use the core APIs and
• 15:48
  do stuff with them. You have to externalise the
• 15:50
  vectors, you have to externalise the governance, you have to get,
• 15:54
  you have to do grounding in a way which is again common to all
• 15:57
  of these raw providers and so on.
• 15:59
  So in terms of creating novelties this huge and then you
• 16:03
  don't get patent until unless you don't bring in the
• 16:05
  application aspects on, on, on all of it, it cannot be a
• 16:10
  problem, right. So that takes me back to what
• 16:13
  kind of new application interfaces can build with this
• 16:16
  kind of new, which I already touched in some ways.
• 16:19
  And that's the opportunity for businesses to think, think
• 16:23
  differently, to also start exploring from a product
• 16:26
  provider, how do I become an into an extended product to a
• 16:30
  service providers and stuff like that.
• 16:32
  There are companies, big CPG companies who are saying we just
• 16:35
  don't want to be selling products.
• 16:37
  Rather we want to get embedded a service provider into the
• 16:39
  consumer lifestyles at an emotional level.
• 16:42
  And and you would have heard this from some amazing fashion
• 16:45
  companies, beauty companies, sports companies and so on.
• 16:48
  Alex But but easier said than done, The engineering for this
• 16:52
  needs a very unique kind of an approach and that's the
• 16:56
  opportunity for us to from a novelty point of view.
• 16:59
  That's wonderful. Thank you.
• 17:00
  Where can people find out more information, Raul?
• 17:02
  So absolutely I'm I'm always accessible on on my LinkedIn, on
• 17:07
  my, my e-mail idrahul.mahajan@nagaro.com can
• 17:11
  reach out to Nagaro websites, reach out to us and then some of
• 17:15
  us are actively available in different events and so on.
• 17:18
  Lot of what I said by the way is, is is there on our, in, is
• 17:22
  there on, on our and our marketing materials and
• 17:24
  brochures. So we would be happy to talk to
• 17:26
  any of our interesting clients. That's brilliant.
• 17:29
  Thank you so much. I really appreciate.
• 17:31
  It thank you Alex for the opportunity.
• 17:33
  Thanks for inviting me. Thank you for your time.