Why Enterprise Integration Still Fails and How AI Can Fix It

FULL TRANSCRIPT

This transcript is based on the episode’s English auto-captions and has been formatted for readability. Please allow for occasional transcription errors in names, acronyms and specialised terms.

[00:00:00]

The issue of integration or inability to integrate will persist unless you start thinking of another layer which sits on top of the applications and says okay you don't need to speak any language I will figure out how to talk to you and I'll figure out how to talk to him and I'll make them both talk to each other and once that mediation layer comes into existence it'll be like what people used to call ether Right? It is there everywhere. So that mediation layer once it spreads into the system then uh you will find that you can concentrate a application A can just concentrate on what it wants B can concentrate on what it wants and the communication will be done by this AI based layer mediation layer a strict workflow that means a workflow that is predefined but AI augmented to ensure that certain things that need to be transformed and certain things that need to be uh brought in and understood etc. that is done using AI and the rest of it is done using traditional workflow. You will s solve more than 50% of today's problems.

[00:01:16]

M >> I want the world to wake up to a situation where I can put together a business process that covers a shipping company, an insurance company, a a government agency, a bank and we can assemble the whole thing together, find our knowledge work and then create your workflow okay underneath which will talk to AI for various things but still be executing your command. Those solutions are what you need to do straight away and that will give you far higher success percentage to keep the AI momentum going. Hello and welcome to the enterprise tech talk. I am your host Saumitra Kalikar.

[00:02:23]

Now enterprises have spent decades pursuing one ambition that is seamless automation. from ERP to APIs and from RPS to agentic AI. The technology have promised um elimination of workflow manual workflows and creation of um connected and intelligent enterprises. Yet despite all these technology innovations, the the challenge of enterprise integration and automation continues to resurface. In fact, in most of the organizations even today, uh manual handoffs, uh disconnected systems and fragmented business processes continue to be part of day-to-day operations. So, have we have have we um addressed the automation problem or are we just getting better at working around it? To unpack this topic, I'm joined today by um Madhav Sivdas. Uh Madhav is the entrepreneur with uh almost 30 years of experience in building enterprise software. Uh he also holds many uh patents um including the uh the one which subsequently resulted in uh creating robotic process automation technology. He is currently CEO and CTO of Telgro where he continues to build next generation um enterprise AI. Mad welcome to the podcast. Great to have you here.

[00:03:51]

>> Thank you very much Saumitra. Very nice to be here. >> Uh Mu uh you have worked across um enterprise software uh integration platforms RPA and now of course enterprise AI. So I want to draw on your experience. So if we if we look back if across last three decades in your view how enterprise automation has evolved what what fun how fundamentally it has changed and equally how it has not changed. >> Okay um it's going to be a long answer um if you if you wish to have a long answer. So uh as uh you had previously mentioned uh I've been around looking at these things for nearly 30 years. Um and uh specifically I've been involved in um distributed systems. Uh so from from the early '90s through um various kinds of uh early generation distributed systems using uh Corba and RPC all the way through to the existence of Java and net technologies and then later on now what has happened is um there is a in in 1968 there's an old um article by a person called Melvin Conway and he wrote that enterprises build systems according to the way the organizations are structured and that will later on went on to become what's called Conway's law.

[00:05:26]

>> So which means in large enterprises or any enterprise the software systems directly represent or reflect the way the enterprises are organized. Now that means there will be lots of applications. So I'm starting the premise from that right that there will be lots of applications and each functional unit of an organization subfunctional unit and so on will partition their uh their their software systems in certain organizational um mirroring. Now these systems need at some point to communicate with each other but they may not have been conceived at some point to speak with each other. So the problem has been around for all of these three decades. There were things like um NQC or it still does exist but messaging was one mechanism where two or more disparate systems could try to communicate by sending messages over a over an asynchronous pipe. Um then you had of course distributed systems using things like Corba etc which were trying to talk to multiple applications residing remotely and talking to each other using certain network protocols.

[00:06:51]

And then we went into the world of uh you know uh web commerce etc. >> E e- business so to say. So that that started perhaps 1996 9798 99 all the way up to the com boom and then those technologies came. So point therefore be there have been waves of things happening waves of technologies and still that old Conway's law that enterprises will build software according to their organizational structures that holds and they these systems may not always communicate with each other that also holds. So as a result we've go gone through all these different phases of creating various kinds of integration technologies but never solved it completely. >> That is the state of affair today.

[00:07:49]

>> Okay. Now uh the other uh aspect overlaying this is that all of the past integration technologies have been on the uh basic understanding that when two systems need to talk to each other they need to develop a standard of communication and it is so A talks to B because there is a standard C that A will talk to C and then C will C will somehow communicate communicated back to B. So this underlying framework and that is the foundation of messaging systems, web methods, all of the old IBM technologies, all the Oracle technologies and so on. So then what happens is when these systems need to talk, they need to agree upon that standard that communication protocol that and that is what is called EAI, enterprise application integration. Now all these systems may not wish to speak in a given EAI language because they come from different uh vintages.

[00:08:56]

So it is with that uh in that uh scenario that around uh 2004 uh we hit upon or almost accidentally hit upon this thing called UI integration. >> Okay. Um and you user interface integration becoming a means of enabling completely unknown applications to be able to speak to each other by talking to their screens. Okay. Now the hardliners may say that's not enterprise integration because you know it doesn't follow the the the hardline um criteria of what what enterprise integration is all about. But then it enables A and B to talk to each other without needing a C to u uh that A has to comply with and B has to comply with. But on the other hand, now the C is something that will figure out what A is like and what B is like and then do the communication. Now that that nuance is important because there is something called the onus to comply. It's something it's a term that kind of >> figured out in my head that in an integration scenario the onus to comply is on the participating applications in the in a conventional sense when A wants to talk to B A needs to know how to how to communicate using this C protocol and then B also needs to be able to speak to that. So these guys need to change their systems to uh to comply with that and the onus is on A and B in order to talk to each other.

[00:10:41]

So when this world of UI came, UI integration came along that onus shifted to the integration technology unlike previously where the integration technology said I'll give you a standard you talk to me in that standard you talk to me in that standard and I'll join you up. Here there is another kind of integration technology that says don't worry I don't need you to tell me anything of I I don't need you to comply with anything that I have. I will find out what you do and what this guy does and I'll make the two talk and that was the birth of UI integration which then became our client. Yeah. So uh to answer your question uh is these are this is the state of affairs in in the uh as things have moved over the last 30 years people have the these technology there will be an ex there will always be disparate applications there'll always be heterogeneous applications and there will never be a monolithic underlying standard that everybody magically is able to okay so the issue of integration or inability to integrate will persist unless you start thinking of another layer which sits on top of the applications and says okay you don't need to speak any language I will figure out how to talk to you and I'll figure out how to talk to him and I'll make them both talk to each other so um I'll pause here and uh you know enable you to continue.

[00:12:22]

>> Yeah. No, that's that's really good good uh background overview uh mother. >> So in that journey, do you now see agentic AI >> as a next step of of integration automation where >> the new protocols emerging like MCP and >> um and A2A etc. >> Where where would you put those in in this whole journey? >> Okay. um MCP etc. Again those are things that a given system may put out to enable tool calling by agents by AI enabled pieces of software um that it will work in scenarios where that MCP layer is available but again in the world of enterprises today that is not the case. So you may have uh a 15-year-old loan origination system in a bank uh that needs to make a communication with another dozen other things. There are no MCP servers for all of them. And uh in any way if they need to communicate with each other if any of these participants needs to modify itself you are back to the world of projects and you have to you know fund it. you have to put people onto it etc.

[00:13:47]

So, agentic AI will create another set of things to integrate with. But the key thing according to me would be for the things that today are or for all the things that are unintegratable, we can create a mediation layer. >> Okay? And that mediation layer can be Mhm. >> Thus that mediation layer will be able to say okay none of you have to make any changes. We will make all the adjustments needed. So if A has a certain data format and A emits everything in a certain format and B wants it in another data format, we now have using AI all the wherewithal to convert it from that A's format to the B's format.

[00:14:48]

Okay. prior to this prior >> sorry just to understand there um so how is that specifically different to a typical middleware uh era where yes it supported the different formats or different u models data model conversions >> okay so in a different in the middleware world when A needs to communicate with B he has to be able to figure out what B's protocol is and someone has to do the mapping >> y So it is either the project owner of A or the project owner of B that of system A or system B that needed to call in and create a project that would then uh make that mapping happen. Okay. So one of the two participants of this communication leg would have had to modify their systems in order to be able to make this communication happen.

[00:15:49]

And that's a project cost. Yeah. >> And this continues to exist today. The 30 odd years that we've been in the industry, we can see there is no end to it. It is no end to it because up until now this ability for something to figure out how these two things are able to talk or what they want and then be able to take something from one and give it to the transform it and give it to the other the way the other one wants it. That has not been possible because of the uh the the the impedance in the knowledge of what A likes and what B wants. play. But now with AI, the ability of some of these specific models, you don't need to have frontier models to do all this. You don't need to have a uh an AI frontier model that knows a cheesecake recipe to also provide you data mapping. You don't need to do that. You can have a very simple model, a simple AI model that can be very adept at enterprise data transformations for example, at enterprise communications for example and those are much smaller, much cheaper to implement and much quicker to uh you know provide you the success of this kind of an integration. And once that mediation layer comes into existence, it'll be like what people used to call ether, right? It is there everywhere. So that mediation layer once it spreads into the system, then uh you will find that you can concentrate a application a can just concentrate on what it wants, B can concentrate on what it wants and the communication will be done by this AI based layer, mediation layer. So that's that's the kind of a holy grail I think that we should be trying to move towards.

[00:17:45]

>> Yeah. Is this what I think you are positioning as intelligent transaction network? Um is this the same is this the same concept or is this is that something different? >> Okay. Um the in what I do from a from a work perspective is uh andro is is promoting a a technology called intelligent transaction network. Yes. um that is a scenario from a businessto business integration perspective. So when you've got multiple entities, businesses and when they need to collaborate today, there are very little or very few um integration mechanisms unless you go for some of those very very heavyweight things like electronic data interchange, businessto business gateways and some very very specific uh businessto business connections. But for the large majority there is nothing. So the intelligent trans transaction network aims to provide the same kind of capability in a B2B world. But what I mentioned to you earlier is a is a layer that uh an enterprise can have within >> Yeah. >> Okay. and and that can be something that they should and that is something from a from an enterprise architecture perspective people should be looking at according to me uh far more urgently than you know jumping on to AI to do all kinds of uh all these things that they think it can do.

[00:19:20]

>> Yeah. Yeah. Okay. Um so I wanted to move to a diff new different topic um mano just to understand um how organizations should be focusing on in um when they initiate automation programs um and the and the a scenario I can potentially paint uh in front of you is if you take let's say an insurance industry right and and a claims processing claims settlement as a as a end toend business process. Often what happens is um the process is spread across multiple fun teams or departments within within insurance uh company. Um yes maybe uh an um a claims team might actually automate document processing for example or um an underwriting team might actually use AI assisted risk assessment function something like that.

[00:20:21]

Yes. So, so in part those automations happen but from customer perspective he still waits for days for a claim to be set up right. Um so I wanted to get your thoughts on how organizations should be thinking when it comes to the automation programs in uh you know uh broadly because um it seems that many times we focus on departmental or specific siloed automations without looking at end to end picture as to how the business process should be re-engineered and so net effect from customer perspective still potentially may not be significantly improved Exactly. >> And how do you see uh what kind of advice you will provide to customers on that front? Um how they should be looking at building the proper automation programs? >> Uh the first uh word or phrase comes that comes to my mind is outcome. Okay. So once you start focusing uh your your um your results on an outcome that means I need the claim to be done within 15 minutes of it arriving the border of my perimeter of my organization.

[00:21:33]

Uh then you will not be looking at oh this department or that department let me digitally transform them or they will do their stuff they will do their stuff etc. If when you are in those kinds of worlds because that's what happens when people are in large enterprises and organizations they just get covered by the walls of their own departments that they don't see beyond and they don't see the actual outcome that the business wants. They see the outcome of when it enters their system and when it leaves their system. That's all they know. But the business looks at it at a much no much higher altitude. looks at it from much higher altitude. Therefore, that outcome is the thing that should drive the transformation and when that happens they the some entity at that level should be able to say this is how the whole thing works. That means it first comes here. You've got uh let's say document assessment.

[00:22:34]

You've got uh claim settlement. You've got all those various steps that that happen. What are the things that need to be put in place to automate the entire thing so that I get that single outcome of a 10-minut turnaround time or a 5 minute turnaround time etc. and a 90% or 95% accuracy. Right? Um and when you look at it from that perspective, you will then start looking at all the solutions needed in each of the blocks that comprises the entire that comprise the entire solution. >> Right? So uh that is what needs to be transformed because if it if it look if everyone looks at it from their own perspective they'll just do something that that is relevant to them but the total outcome or the to overall time will still remain the same. >> Yeah. >> Okay. Same thing goes for underwriting. Same thing goes for uh you know things as complex as trade finance. Okay. Uh trade finance also involves receiving documents from different kinds of entities. You got buyers, sellers, issuing banks, settling banks. You got uh all all kinds of entities involved in a single transaction. And if if the bank or the the organization only looks at their thing, the outcome will still remain many days uh of latency.

[00:24:02]

One entity may say, "I've done my part." So u yeah so outcome is what you need to look for and you need to stitch together all the technologies needed to make the entire outcome uh within that particular figure or that statistic that you wish to achieve or that metric that you wish to achieve. Yeah. And and just building on that, I also wanted to get your thoughts on how enterprises should be looking at first thing building their maturity about around automation and but um how should they be measuring the maturity around automation as at enterprise level because uh as as as we discussed most of the programs focus on siloed metrics like for a particular department how much productivity I achieved But if you are to now start thinking at at an enterprise level I the organization is really pushing for automations into automations what what kind of key indicators they should be focusing or a driver they should be focusing on so that uh they become more enterprise level success measures. >> People have attempted to do this so many times. Okay. uh they have create they have create for every new technology that comes in they bring centers of excellence and they sit and create all these different metrics I've not seen any of them working okay but when it now comes to uh the the new wave of AIdriven automations uh uh I think what they should be looking at I'll go back again to that phrase that they should be looking at the end result that they wish and all the they should bring together or assemble all the participating entities and start looking at how those communication channels can be bridged >> in the in the shortest possible way. And the key thing to be done is all of this has to happen with little or no changes to any of the participating organizations or participating systems because the moment you make changes to these systems then you're we've gone back to the old era of of complying with something that is a product of this wave.

[00:26:20]

When this wave goes gives way to another wave then you will again go back to changing your systems. M >> so you have to switch to a world where every time something happens the these composing or component applications don't need to make changes >> because they probably they're most likely doing a very good job of the functionality that they were built to do. That is a fundamental thing, right? Uh a financial system, a loan system or core banking system, an underwriting system, they do their things pretty well, rock solid, because they've been around for 20, 30 years, at least 15 to 20 years. You make them do you you open them up and change them or rip them out and create a new one just in order to comply with something that has come in new.

[00:27:18]

Maybe uh it was a uh thick client uh system but and now that's too old and archaic and I don't want to use Citrix so let me make it web- based. So when you made that you throw away 25 years of u you know a lot of engineering and a lot of knowledge. >> Mhm. So now so when you keep doing that so when you comply with things just to fit the technology of that wave then you're not doing any any good is according to me okay of course you you need to web enable definitely you need to you can't be living in the client applications I fully agree but the the core point being that in order to now automate if you need to make changes >> that is something you should avoid and you should bring in something else that will do that bridging of that communication for you so that your internal systems or your core systems remain the same.

[00:28:18]

>> Right. Okay. Let me let me test that because that's an really interesting point or perspective that you you have painted. Uh let me test it from uh by come bringing the or draw the parallel with how we use RPA right. um uh in the last decade. Now, RPA was definitely a success in the last decade and what potentially in my view why it was successful was it was uh it really helped in automating the mundane repeatable tasks within existing processes, right? So, it really gave very quick benefits in that in that sense. So, definitely played a really uh key role in realizing quick benefits. But an argument also always goes against RPA. Uh that while it help us to realize quick benefits. It just automated the the existing processes as they were being executed. It didn't allow us to re-engineer the processes. It didn't allow us to find ways to optimize those processes. Right? if we just use RPH to automate what was already being done manually. And I and I wanted to get your thoughts on that. uh as to do you see that that was potentially a learning or even if if I use a strong word a mistake in in the RPA world that we should be learning from when we are now actually getting into agentic era world and we want to use now this new technology for process automation.

[00:30:00]

Should we be the simple question is simple is should we be again using agentic AI to just automate what is already existing processes or we should actually take a step back and really re-engineer the processes before we invest into agentic >> AI. Okay. So yeah this has been a a subject of this this topic has been around u let's say from some 200 13 14 onwards okay 12 is around the time I say that the word RPA started becoming uh you know repeated many times by people uh 2012 okay so yes and I I'll give you an elaborate answer for this uh when we created UI automation and we never called it RPA by the way. Uh it wasn't initially meant to do what it is doing today.

[00:30:59]

Okay. Uh our background has been enterprise integration. So we are from EI world and we got into this because uh uh there were people who we were selling EI to who strongly rejected our our uh sales pitches of uh you know the the you know web sphere web logic type of uh uh integration technologies. They said we are a large regional conglomerate. We've got hundreds of applications. We're not going to buy your $20 million thing and do things because these things these applications are running pretty well. Yes, we do want them to talk to each other and uh but for that I'm not going to make big changes. That's what was told to me and people along in my team when we were working in other companies trying to sell conventional integration. So we came from the so we created UI integration in response to that.

[00:32:00]

>> Mhm. >> We didn't create UI integration to uh you know rapidly uh convert all these human activities etc. All that happened much later. So the birth of UI automation initially according to me was not for uh the kind of things that it is ended up doing today. So that is point >> Mhm. Then it is around 200910 that uh when we went to from Singapore to India to see certain scenarios in BPO organizations that's when we realized that oh this is something that uh our UI integrator tool can be used in which is uh it can be put in front of an agent's desk and what they've been doing with five or six applications this thing can do on its own. the agent can go back home or they can a 20 uh person process can now be managed by five people. That was how it this this thing took birth.

[00:33:01]

Uh it was being done of course by another competing company in the contact center world. >> Okay. In in your in US uh Israel etc. But in the uh non-cont center world, we were the ones who brought in the concept that maybe this UI integration can do this. Now, at some point, the situation went out of control and people that we spoke about this to, they decided that we're going to take every process that is there and we're going to put this thing and try to speed it up. Mhm. >> That's when you have this scenario of, you know, you make um an inefficient process automated. You only, you know, magnify or create further uh uh issues when you when you take an already inefficient process and you try to speed it up.

[00:33:58]

uh so those things like I said came much later when people you know you introduce something to the world but then the world decides to take it in a certain a different direction. >> Okay. So uh that is what I think or I I feel from my perspective with what happened. uh and it is then that people came up with the idea that oh we must uh do process re-engineering we must understand what is the lean waste what is the you know you they bring in six sigma lean and all of that and then you they used to do time and motion studies first have a again good old word a good old phrase center of excellence then a center of excellence has a team they start reviewing all the different u processes of the organization and then it's a stack of them and some boss decides I'm going to take the next first 15 of them and try to engineer them or re-engineer them etc. So those were things that people some people were doing but then when uh you started seeing 50 60% 80% benefits there is a natural vacuum or a suction that happens. Let's deploy this everywhere and then when that happens and things fail then there is uh you know the traditional the the hype and then the trough of disillusionment.

[00:35:16]

>> Yeah. So the dissolution started it went up because there was a heightened expectancy. Uh and then uh not everyone was capable of understanding which processes need to be done need to be automated. How should those automations take place? You have you have a tool but you know if you don't use the tool properly you may create very dangerous uh outcomes. So and that is what created all the failure scenarios uh which then became very famous the consultants who created uh various studies of how RPA has failed you know 40% failure here 30% failure there etc and then that became uh you know a sort of uh another cloud you know whenever you use you hear the term RPA yeah there are certain good things but then there's that cloud of failure years. >> Mhm. Now that can happen with AI as well. So when it come when it come when you come and bring bring the world of AI, agentic AI, you're likely to face the same kind of uh situations where the initial success of a few things will cause some eager people to you know spray it onto a whole bunch of things and it'll all start failing. Uh there is a term it may not be politically correct uh but it's still something worth uh every every every IT leader should know and that term is or that that that saying is a fool with a tool is still a fool.

[00:36:55]

>> Okay. So you give an intelligent or a very smart tool to someone who is not capable of understanding its usage, you will end up creating solutions that may not be adequate and it will not be uh you know serving the purpose and in fact it may it may be contrary to what you were wanting it to do. >> So just having a tool is not going to be enough. You need to have people who understand how to use the tool where to apply it. So you need all of that and only then you can apply whether it was RPA whether it is agent KI today. >> Yeah. Yeah. And um so because many organizations today are at at this juncture where they are initiated most of the organizations are definitely experimenting with agent care. Some organizations are now trying to productionize it and I think to your point many organizations could be at that juncture where if they see success they will start accelerating AI investment more faster and the risk could be the same as that happened with um in the era of RPA where >> yes >> organizations will they will spraw a systems and solutions across different departments within organization without having a proper understanding of what should be the approach or where process which process should be engineered. Uh but also what are the important architecture foration those should be laid out. Sorry.

[00:38:31]

>> Absolutely. >> Right. So >> absolutely >> what advice would you provide to those organizations where they might be uh sensing some success but what kind of precaution or advice you will provide to them to how they should be approaching the proper scaling of agentic solutions. >> Okay so you know the old style the old the general consulting style would have been center of excellence. Okay, bring in buy bring in a bunch of five or six people in your organization who will do this. But I've seen centers of excellence fail. Okay, and they become bottlenecks. So the I will I will refrain from going by the you know using that very easy solution uh center of excellence but I would say that the uh technology organization of of every enterprise must uh look at these um uh for for one thing they should not do is they shouldn't be looking at these frontier models to do all these things as if somehow by magic it is going to solve problems for you. Okay, a frontier model has a whole bunch of training data. Some of it is good, some of it may not be good. Today you will use it only for providing a whole bunch of context. So this is how you your typical your token usage happens. You provide it a lot of context and you say okay what do you make of out of this and then you build an agentic layer out of it saying you you will prompt the creation of agents by saying get this review it go and check something and put it into a loop see if this is there see that is there right now every single thing in the world does not lend itself to being agentically solved in fact Today the most important things that need to be done really don't need this kind of agentic solution.

[00:40:40]

You still need AI but you don't need this AI that self-thinks all these things. Even if you were to create a workflow AI that means a strict workflow that means a workflow that is predefined but AI augmented to ensure that certain things that need to be transformed and certain things that need to be uh brought in and understood etc that is done using AI and the rest of it is done using traditional workflow you will solve more than 50% of today's problems. Today's CTO problems, CTO, CIO problems can be solved today. >> Okay. You use AI for those things that you could not have used uh conventional technologies for before and still rely on an underbelly which is still a workflow executor.

[00:41:39]

If you then replace the workflow executor with an agentic AI where it can do whatever it feels like that I think will result in a lot more failures and it'll again bring you down into this trough of dissolution. >> Yeah. >> So rather than going that way that bring agentic AI everywhere and for everything we create agents we try to take a more mature approach. Let's look at augmenting our workflows with a with AI maybe agentic if required and make the actual workflow complete successfully. I think that will be my that will be my approach and that will give us according to me lot more uh a much higher success percentage which will then help continue to push the AI agenda >> otherwise what will happen same thing uh there will be a research study that says AI has failed agentic AI has failed 30% of projects have failed 9 million9 billion have been wasted tokens who's exhausted all of that story. >> Yeah, that was I think uh an MIT research which indicated around 90% of AI pilots do not get to production, right? >> Um and I think I would also probably um add to what you just said. I mean yes um traditional workflows are may al would continue to deliver value and also when it comes to AI agentic AI is just just one aspect to that your traditional AI the ML based AI that has given organizations benefits for last many years and that would still continue to deliver value for many many use cases >> of course so where there is analytics there's analysis of large volumes of data you can use those techniques You can customize you don't need to use again frontier models for that you can customize uh things for yourselves within your large organization uh so that you query that AI based system to derive answers for your uh those kind of judgment related or assistance related queries. >> So a lot of the customer u customer service u all of these could use AI significantly. Okay. So typical thing today or in the past we've uh tried to talk to robotic agents or typed things into chat bots you know for your in in in telco companies banks airlines I'm not able to book this I'm not able to rebook this ticket what's you know what can be done and you get responses that are based on very stock standard situations right even today >> despite having uh so what what have they done they've just set up a language generator meaning it can respond to you in English or whatever language but it doesn't have the the skills to be able to understand what you're saying and say oh you come up with a very unique problem you're not able to change your uh booking for of this airline and this thing because of a unique situation with your fair class on this and that and that and My online system can't solve it for you because it's not geared for that. My agent is not able to solve it for you because they're not geared for that. This is how it can be solved.

[00:45:14]

>> Yeah. >> You still don't have that. And if you can do that, you will create a customer um uh satisfaction and customer delight beyond what you can imagine. But they still have not done that. So that's another use of AI. >> Yeah, absolutely. the the difference I'm saying in um in the when RPA took off versus the current where we are currently in terms of agentic world uh it's not only the the enthusiasm from techn technical people that is of course there and no surprise there it's a new technology but there is a a top-down pressure from boards now this is a new thing I don't I don't remember that kind of top down pressure was there in the RP era or any other era Right. Um because for whatever reason agent has become such a visible thing. Um there is a pressure coming right from top down as to what we are doing in the agentic era world. What are our you how we are investing or what use we are delivering.

[00:46:15]

So to your point I think uh the middle management the the the the architects the uh and those those guys need to provide much clear and transparent and honest view as to really where the the new agentic use cases are important versus where we can deliver value with existing and existing workflows. I yes and uh to to can further your uh point uh unfortunately in the world today a good chunk of work that people are humans are doing can actually be done now automatically through the use of these these these technologies. uh maybe politically incorrect, maybe socially uh you know creating uh situations but those jobs those kind what they call knowledge work >> those jobs came into existence because of inadequacies of existing or at that time prevailing software systems.

[00:47:24]

>> Okay. So you had two or three systems that were that that people had they wouldn't talk to each other. So um a human was brought in and that human started doing this particular activity and then they became specialists in that right now with AI being able to solve do all of that this particular human's role becomes uh questionable. Uh similarly uh there could be various other uh decision or analysis types activity which we call knowledge work uh which can now be completely done by AI based systems appropriately trained again not using vanilla models but appropriately trained models >> in addition to uh you know providing the right kind of search data context etc. you can and some workflow. Okay, all of that together will be able to create solutions that completely remove the need for huge chunks of humans.

[00:48:30]

>> Yeah. >> In an organization and that is is going to happen. It's a matter of time. Now, one may wish to resist that. Okay. uh there may be social implications but I feel that uh eventually the the profitability etc of a business an organization of enterprises will give way and these these uh optimizations will happen. >> Yeah. >> Okay. So from a board perspective, what people would think of is since I'm able to sit on on my chat GPT and I can do all these great great things, why why aren't you doing this for the company? That's what the board member is asking. That's why you have all these pressures. The same pressure, you're right, didn't didn't exist in RPA because there wasn't a board member who could use an RPA on his mobile and then say, hey, this is looking good. Why don't you do it? Right. So, but now with with AI, you can do that. Yeah.

[00:49:30]

>> Right. Uh but the consequence is many many chunks of you know positions will now become redundant. Okay. So because the whole thing can be done now to be safe you will keep one person or two people or you know 10% of the workforce but the remaining 90% may not be relevant uh once this AI thing comes comes through and it's AI already has all the capabilities today to make that happen. It's not like it needs to do anything more or improve in order to make right now. You can you can do it. >> You're right. Yeah. Agree. >> Okay. Uh look uh there is one um specific area I wanted to explore with you mother when it comes to integration and automation and mainly because um you predominantly nowadays work into the B2B integration scenarios. Um um for in the in the B2B world uh organizations have historically invested into um more into digital portals uh to start with when it comes when it came to automating some the task right uh they build it supplier portals partner portals what not right customer portals >> while it helped to certain extent definitely improving the experience but uh in the hindsight it shifted the work from manual work into the digital space.

[00:51:02]

Right? The core focus about eliminating the work itself and automating an what I would say straight through execution of transactions across customers and um u uh companies or between the partners and company etc. that that didn't u go away right uh that that was not realized and I wanted to get your thoughts firstly on why the straight through execution uh continues to be challenge in the B2B world I I come from telco background I know uh how in Australia for example how NBN and Telos now they have established this whole B2B but it took a long time to build build that but with most of the integr industries this is continues to be a tough challenge as to how partners and whole ecosystem can work together. Right? I wanted to get your thoughts initially on why that continues to be such a big challenge. Uh most of the organizations settle on partner portals but not get going all the way to straight through executions and what are what can be done within industries to improve upon that.

[00:52:12]

So this is a a topic more dear and near to me at this point in my my uh organization. >> Okay. So as you rightly said um the existence of portals and what we call self-service portals uh either it could be employee self-service or customer self-service or partner self-service. So basically uh one entity has created u an application uh that is called a portal and is available on the web and all their counterparties whether they are employees, customers, partners, they will log into that application. Okay, great. So for every communication that you need to do with that particular entity, you log into their app and do things. So their side is great is working. Okay. At least this part of it. >> Yeah. >> Okay. But what about the entities that need to actually sit in front of a PC and do this thing?

[00:53:13]

Okay. So you're right. Even in Telco and many other systems, this problem is there. The interim solution has been RPA. >> Okay. Where some of these players will say, "Oh, it's so difficult for me to do it. So let me put an RPA to communicate with that app. We in 2007 implemented one such RPA for one such Australian telco to be able to communicate with another Australian telco. Okay. So you can imagine who those telos are, right? So, one of the Australian Telos had a self-service app that provided bulk leasing and bulk line provisioning capabilities and the smaller Telos would have to log in and do a lot of things in order to provision those lines and set up the roots etc. And that was creating issues.

[00:54:23]

uh uh in in that scenario it was most of most of it was rekeying issues meaning the the keying of data even if they go one digit here and there it can create big mistakes so we created a UI automation in those days it wasn't called RPA okay so the history therefore has been that when you have these kinds of latencies that one entity provides an interface and thinks everything is over everything is done the all the other partners or counterparties sit and struggle to enter things. Each counterparty which has got money will invest in some kind of RPA to automate this route. But that's a sporadic thing. Some people would, some people won't now that creates still the latency that exists in this entire system is still present. What companies and organizations haven't understood is that the world is not themselves only.

[00:55:26]

Most organizations think this is my world and everything revolves around my organization. But in reality, a business transaction goes through many organizations. Lots of business transactions go through many organizations. and sporadically fixing some point-to-point solution will not help using RPA. So, you need some capability that can do everything. Now, some entities, Telos, banks etc. try to achieve automation by creating standards. So that is the traditional way of automation or B2B uh smoothening things out is you create a standard once you create a standard then it's up to all the participants to comply with the standard and then game over there everything works right but there are sit situations and that may work for a few things there have been standards for EDI for example there's been standards like uh swift and the ISO standards for for money stands for certain things work but then I would see this as an iceberg the things that are working up just this the the tip of the iceberg sticking out of the water there is still the largest chunk the 95% of things that happen between companies cannot be standardized the other thing that goes against standards is the cost of complying with a standard So people will bear the cost when it comes to the largest or the most valuable or the most meaningful things in their company in their enterprise.

[00:57:13]

Everything else they'll push it under under the carpet and they will send it to overseas capability centers outsourcing shared service whatever they want to call it today. Okay. And it'll be done manually. So the scenario is that in reality you want to wake up to a world where businessto business is also automated and composable. This is what is uh things that we are working on which is dear to me. Mhm. >> I want the world to wake up to a situation where I can put together a business process that covers a shipping company, an insurance company, a a government agency, a bank and we can assemble the whole thing together and that will be an end to-end process that is completely automated with all the work being done by the individual entities IT systems. We don't have the capability to to do all the business logic. Okay. So the the point we're trying to make therefore is all of this uh standardization is not going to work everywhere.

[00:58:30]

Point-to-point RPA kind of solution is not going to work everywhere. Simply creating um a self-service portal and telling everybody to hit it is not going to work also because all of them will end up in lots of inefficiencies and the end toend business process will To solve that you need a system that can wherever required use RPA, use EI, use EDI and achieve the outcome of connecting from business A to business B somehow and do all the transformations that may be needed in order to get the data packet from A to B. Simply having a a self-service portal will not solve the job. It is definitely a step forward is better than not having it. >> Okay. But having created one, these people should not go back home and saying, "Okay, our job is done."

[00:59:46]

A CTO or an organization that understands that will then parti will will lend themselves to participating in a network and that network will then be able to do this kind of businessto business or enterprise to enterprise automation. It will involve use of a AI, use of agentic AI, use of workflows and the use of integration technologies. So all four things. >> Mhm. >> Okay. And main thing is it should not cause anything on the participant side to change as a result of this participation because again if you do that then you're going back to the world of I I need to comply with you to in order to talk to you. Yeah. >> The moment I need to comply with you to talk to you, then I need to run a project inside to make my systems comply with you.

[01:00:48]

>> Yeah. >> Okay. In a larger scheme of things, that will not work out. So the world has to come to a a system that can allow people to plug and play. >> Yeah. Sure. >> Okay. So that is where I think the world should be moving towards and AI provides all of us uh a very good set of tooling to start know walking that direction. >> Yeah. No, that's good. Thanks. Look, as we start wrapping wrapping up, Madona, um I wanted to get your final thoughts on u what what advice you offer to um uh CIOS or CTOs who would be at this juncture thinking of or started initial investment into AI uh um building their AI strategies uh for the long term. How what coming to what we discussed earlier? key things, key learnings they should they they should be um building upon from from previous integration eras. Um and what key one or two important inputs or advices you will offer to them. Okay. Um key thing would be um look at all the knowledge work that is going on.

[01:02:10]

Okay. And you will find that a whole big chunk of it uh can now be completely taken out. Everything that perhaps you used to do in terms of let's say in the old days okayif uh 60s7s and early 80s uh everything that had to do with storage of data with one sweep the world of the technology of databases killed it. No more ledgers. No more writing into long paper, right? No more filing of things. With one sweep, database turned that out, took that out. In the same way, with one sweep, the current state of maturity of AI can take out a huge chunk of knowledge work or so-called knowledge So you should look at how that can be achieved within your organization.

[01:03:13]

It will be the first thing. Now doing that will not the the the natural tendency will be to uh to use a for one of a better word to abdicate like I don't know how to do you do it. Okay. And AI you do it for me. So there is a natural tendency that you will just say I'll bring a strong AI frontier model and I'll say this is the problem and the AI will just give me the solution that will again lead to failure. So the advice would be find out knowledge work and then create your workflow okay underneath which will talk to AI for various things at but still be executing your command. Those solutions are what you need to do straight away and that will give you far higher success percentage to keep the AI momentum going. If you throw everything over the wall to a frontier model to do and then it gives you something back and it fails then some of the some of it will work some of it won't and then you will get a mixed reaction.

[01:04:27]

Okay. So these are the two main things I would want people to do. I am refraining from saying set up ex centers of excellence. Okay. Because uh that's an easy thing to do. Okay. Have you know few selected people and the the way that has played out in two waves. So RPA was one wave. S SOA just before RPA there was another wave called serviceoriented architectures. Okay. So these two have we have seen how COE's centers of excellence have played out in uh S SOA and we've seen how centers of excellence have played out in RPA. Some of them have worked. I'm sure those who have participated in this they will say oh it worked. Okay yes it worked. Some places it worked. But uh you know it has also caused lots of frustration because then that COE becomes the uh a choke point >> choke point governance order yeah >> yes too much you still you need governance you can do all of that so you can have your governance your AI related stuff all of that's underlying those are foundational things but uh you know I wouldn't uh you know look at it that way anymore. more that is the normal tendency from a consulting perspective to go that way and I've done it. So perhaps I may have been one of the earliest creators of the power points for establishing a COE for RP for UI automation. Again, it wasn't called RPA in those days.

[01:06:08]

>> Okay. And we had to really, you know, scratch our brains. What do you think we should have? Well, let's go back to SOA. SOA said, you know, have four people from here, two people from here, and uh create governance. So governance is a nice term. Put governance there. and you know so we start right we go right so it's easy to bring things uh from the past and just repeat it uh but I think uh you you some people will still end up doing it right so uh but I would say you continue to do whatever is uh you know the logical thing but the main thing to do is knowledge work >> okay understand all the knowledge work that is happening in the organization and create a workflow plus AI combined that will be able to replace that knowledge work. >> Yeah, >> that's what I think uh is is the way to go according to me.

[01:07:09]

>> Yeah, Mad on that note thanks thanks for your time. Thanks for joining. I think there were so many um u nuggets of wisdom I would say and advice to our to very to all the business and technology leaders. Uh there definitely not many positive takeaways for all of us. Um so thanks again. It was really pleasure hosting you. If you found this discussion valuable please follow and subscribe to Enterprise Tech Talk and thanks for listening. I look forward to seeing you in the next