Transcript: SE Radio 707: Subhajit Paul on ERP Automation and AI
Source: SE Radio | Duration: 59 min
Summary
Here is a comprehensive summary of the key points from the "SE Radio 707: Subhajit Paul on ERP Automation and AI" podcast episode:
Opening Context:
- The guest is Subhajit Paul, who has over 20 years of experience leading ERP implementations at large global enterprises, particularly in electronics manufacturing and supply chain.
- The main topic is how ERP systems work in the real world, including how they are implemented, where they tend to break, and how machine learning and generative AI are starting to be used within ERP.
Key Discussion Points and Insights:
- ERP (Enterprise Resource Planning) systems integrate and automate core business functions like finance, procurement, production, inventory, and sales across an entire organization. This eliminates siloed processes and improves visibility.
- The core ERP processes covered are order-to-cash, plan-to-produce, and procure-to-pay. These link customer orders to production, procurement, and invoicing.
- ERP has evolved from on-premise to SaaS (software-as-a-service) delivery, and now is incorporating machine learning for predictive analytics and generative AI for more conversational, agentic capabilities.
- Failures in ERP implementations can be chaotic, halting core business functions like production, procurement, and shipping. Proper planning, testing, training, and coordination are critical for successful rollouts.
- Machine learning in ERP can enhance existing workflows, such as optimizing warehouse bin placement to improve picking efficiency. Generative AI is starting to add more flexible, conversational interfaces and autonomous decision-making.
- Challenges exist in determining the right scope and granularity for AI agents, as well as when to use vendor-provided models versus building custom ones.
Notable Technologies, Tools, or Concepts Mentioned:
- Machine learning models like k-means clustering, linear regression, XGBoost
- ERP vendors and products: SAP, Oracle, Microsoft Dynamics
- Emerging technologies: MCP (Model Context Protocol), A2A (Agent-to-Agent) communication
- Vendor-provided AI capabilities: SAP Conversational AI (Joule), Oracle Fusion AI Agents, Microsoft Dynamics Copilot
Practical Implications and Recommendations:
- Organizations need a clear data and AI strategy when implementing ERP, evaluating both vendor-provided and custom AI/ML capabilities.
- The process for ERP implementations with AI integration should include steps for AI evaluation, data integration, model selection/training, and model integration.
- Determining the right scope and granularity for AI agents within ERP processes is important, as is understanding when vendor models are sufficient versus needing customization.
- Integrating AI/ML into ERP can enhance existing workflows, but organizations must also be prepared to extend or modify vendor-provided AI capabilities to meet their specific business requirements.
Overall, this episode provides a comprehensive overview of how ERP systems work, the challenges and evolution of ERP implementations, and the growing role of AI and machine learning within ERP platforms. It offers practical insights for organizations looking to leverage these technologies to improve their core business processes.
Full Transcript
[00:00:00] This is Software Engineering Radio, the podcast for professional developers on the web at se-radio.net.
[00:00:07] SE Radio is brought to you by the IEEE Computer Society and by IEEE Software Magazine.
[00:00:12] Online at computer.org slash software.
[00:00:18] Hello everyone, welcome to Software Engineering Radio.
[00:00:22] Today we are talking about ERP or Enterprise Resource Planning Systems and how they actually work in the real world.
[00:00:29] We'll get into how ERP systems are implemented and where they tend to break.
[00:00:35] We'll also talk about where machine learning and Gen AI are starting to show up inside ERP
[00:00:40] and what people mean when they talk about agentic systems,
[00:00:44] basically systems that can decide what to do next instead of just following a fixed workflow.
[00:00:50] And our guest is Subhijit Paul.
[00:00:52] Subhijit has over two decades of experience leading ERP implementations at large global enterprises,
[00:00:58] particularly in electronics manufacturing and supply chain.
[00:01:03] Welcome to the show, Subhijit. Great to have you here.
[00:01:06] Is there anything you'd like to add to your bio before we get started?
[00:01:10] Hi, everybody. Thank you, Kanchan, for having me.
[00:01:12] It's a pleasure to join in this podcast.
[00:01:15] You have covered my introduction very well.
[00:01:18] So let's get started.
[00:01:19] Before we start, I would like to point out our listeners
[00:01:23] that SE Radio has several episodes that go into the background of AI and ML.
[00:01:28] and LLM-based apps. I will point to a few of them. A recent one is episode 698, Srujana
[00:01:35] Merugu on how to build an LLM app. Also episode 641, Catherine Nelson on machine learning and
[00:01:42] data science. And the last one I'll point to is episode 305, Charlie Berger on predictive
[00:01:48] applications. With that, let's start with an introductory question for someone who's never
[00:01:55] worked with ERP. Subhijit, how would you explain what it is and why it's central to running a
[00:02:02] business? Enterprise Resource Planning System, that's the full form of ERP. So we can see like
[00:02:08] in the full form also, the enterprise is there. So this is a total software solution for enterprise
[00:02:15] to run the business. ERP will integrate from finance to procurement, to production, to inventory,
[00:02:22] to sales to logistics to shipping and each and every business system. So without ERP all these
[00:02:30] processes will be running in silos. So that will be creating inefficiency that will be creating
[00:02:35] lot of lagging and there will not be visibility of the business processes at any level. That's the
[00:02:41] main reason if you see all the fortune 500 companies are running in one kind of ERPs or
[00:02:48] some of the companies will be running in multiple ERPs. I can tell the ERP vendors who are really
[00:02:54] leaders in ERP industry like SAP, Oracle, Microsoft Dynamics and now from the example I can tell some
[00:03:03] kind of you know idea like who have never worked with ERP or who doesn't know what is ERP. So let's
[00:03:09] say without a total software solution like ERP sales department got the sales order so it will
[00:03:15] be in an excel and then sales department will be sending those sales order requirement to production
[00:03:21] department in a mail or excel and then from that production department they will be requesting to
[00:03:26] the warehouse for having the components for manufacturing the finish goods so now we can
[00:03:31] see here there are a lot of manual processes there are a lot of manual interaction with
[00:03:35] different departments so there is a lot of data lagging and inefficiencies it can happen but with
[00:03:41] ERP, it will never happen because it will be a total same software platform.
[00:03:46] And you can say ERP is the nervous system of an organization to run the business.
[00:03:51] Thanks, Vijay.
[00:03:52] With that background, can you walk us through a flow, how ERP shows up in day-to-day operations?
[00:04:01] You mentioned it joins multiple departments together.
[00:04:04] So maybe could you talk about how a customer order turns into procurement, production, shipping, and finally cash?
[00:04:13] Yeah. So if we are talking about the basic ERP flows, I would like to mention three basic ERP flows.
[00:04:19] Order to cash, plan to produce, and procure to pay.
[00:04:23] Now with order to cash, ERP manages those customer orders like a sales order through delivery to customer invoicing.
[00:04:30] and then for plan to produce process it will be a demand from sales order so there will be
[00:04:37] production order to manufacture the finished goods and then with procure to pay process
[00:04:42] ERP will create purchase order to external vendors for the components which are required
[00:04:49] to manufacture the product. So I can give one example from electronics manufacturing industry
[00:04:55] where the products is like laptop, mobile, cloud computing infrastructure like racks, servers,
[00:05:03] then industrial drones and you name any kind of electronics component, it will be a product for
[00:05:08] electronics manufacturing industry. So now let's take one example for one familiar product like
[00:05:14] laptop. So we all are ordering laptop. So whenever we are ordering laptop, we are basically selecting
[00:05:20] a different specification right like we are selecting what will be the microprocessor speed
[00:05:25] what will be the chassis what will be the ram so from erp point of view these are all components
[00:05:31] and these components are in a bill of material so that's the way we are tracking all the components
[00:05:39] for manufacturing the laptop now let's say from manufacturing the laptop if sales department got
[00:05:47] a sales order for manufacturing let's say 100 numbers of laptops so for 100 numbers of laptops
[00:05:52] there will be components all will be needed for procuring if those components will not be
[00:06:00] in the company stock so with procurement like a external purchase order to vendors all the
[00:06:07] components will be coming to company stock then only all these manufacturing orders will be fulfilled with the components to manufacture the hundred numbers of laptops So after having the 100 numbers of laptops that will be delivered to customer Now
[00:06:21] if I am just mapping all these processes in these three basic ERP processes. So order to cash is the
[00:06:29] sales order and delivery of the 100 numbers of laptops plan to produce will be the manufacturing
[00:06:35] order, which will be internal to the company, where this 100 numbers of laptops will be
[00:06:41] manufactured. And then procure to pay cycle will be the external purchase order that went to the
[00:06:46] vendors for supplying all the components. So in this way, you can have all these business processes
[00:06:53] mapped in the three basic ERP process. Thank you. That's very helpful. Now, before we dive in further,
[00:06:59] Can you just take a step back and describe how ERP has evolved over time?
[00:07:05] So in the beginning, there were traditional on-premise systems.
[00:07:08] Then there was the SaaS, evolution to SaaS, and how that set the stage for ML, Gen AI, and now agent-based systems.
[00:07:17] Yeah.
[00:07:17] So ERP is an old software concept.
[00:07:22] So it's nothing new, right?
[00:07:23] So I would say ERP has begun maybe in the 1990s.
[00:07:27] And then I started working on ERP in on-premise software.
[00:07:32] So it was around 2005 and then it really evolves from there.
[00:07:37] So around 2010, it went to SaaS ERP delivery model where all the infrastructures like all the servers, application software, data are managed by the ERP vendor.
[00:07:49] So company doesn't have to take any kind of initiative for installation, maintenance of the ERP software.
[00:07:56] and that SaaS ERP will be available over the internet to the user.
[00:08:00] Now from SaaS ERP, like I would say two, three years back,
[00:08:06] we started with AI capability.
[00:08:08] So that time we are talking about ML model,
[00:08:10] that is the machine learning algorithms
[00:08:12] where we can have all the patterns with the big ERP data.
[00:08:18] From there, ML model will be having predictive analytics
[00:08:22] for any kind of business function like anomaly detection,
[00:08:26] like warehouse efficiency improvement and now we are talking about JNI agents. So if we are having
[00:08:34] a chatbot we are always familiar with the chatbots. Now if we are thinking about the chatbot that will
[00:08:40] be directing us how to operate the business process. That's where the JNI agents have come.
[00:08:47] Like you can just write in a chat GPT like thing and then that JNI agent will direct you for the
[00:08:54] correct insight from the ERP. So this is the way it really evaluated. Do you have a real example of
[00:09:01] what happens when an ERP system goes down or went down? What broke first in the business and did it
[00:09:08] escalate? Just walk us through what the chaos might have ensured. Yeah, it will be a complete chaos,
[00:09:14] obviously, because as I said earlier, ERP system is the nervous system of a business. So if ERP
[00:09:21] system goes down then production order will be halted. Procurement will not be happening because
[00:09:27] the buyer will not be sending the purchase order to vendors. Shipment will be delayed so order will
[00:09:33] not be fulfilled so there will be a lot of revenue loss. So to have this kind of downtime because ERP
[00:09:39] is a software right so for any kind of software this update maintenance is required. So to avoid
[00:09:45] this kind of disruption, there are planned downtime that is required always for having the
[00:09:52] update in the ERP for latest kind of software update. And that time, there will be a different
[00:09:58] strategy that organization needs to take. Like there will be a lot of offline entries, like the
[00:10:04] manual entries that will be done. And after that, it will be entered in the ERP. So in that way,
[00:10:10] those kind of downtime can be managed. Got it. Now that we've covered the basics of ERP,
[00:10:17] let's talk about implementation. What does an ERP implementation actually look like from start
[00:10:23] to go live? Yeah. So ERP implementation is a structured process. I would say it will be
[00:10:31] six to seven stages. So it will be starting with the project preparation where the business problems
[00:10:37] will be finalized and the budget will be finalized. Project manager and project team will be assigned
[00:10:44] and then there will be business blueprint phase where that business problem will be discussed with
[00:10:50] business team to understand what will be the changes in the ERP system and then as is and to
[00:10:56] be documents will be prepared and then in the next phase it will be system design and realization phase
[00:11:02] where project team will be changing in the ERP system.
[00:11:07] So it can be an enhancement for the existing functionality
[00:11:10] or it can be a new transaction that is needed for users
[00:11:16] or it can be the integration with the third party systems
[00:11:20] like it can be MES system that needs to be integrated with the ERP
[00:11:25] that is the manufacturing execution system
[00:11:27] which is very common in electronics manufacturing industry
[00:11:30] because all the production operations in electronics manufacturing needs to be tracked
[00:11:35] at very granular level. So, MES system will be integrated for that we need to develop different
[00:11:42] framework and then we have to also do different framework for business to business communication
[00:11:48] with different customers and vendors like we have to send maybe inventory data to the customers and
[00:11:54] they want to have the visibility of company stock how the raw materials are coming according to that
[00:12:02] they will be placing the orders so having said this will be the phase where ERP team will be
[00:12:09] really doing the changes in the system and then there will be the next stage where all these
[00:12:17] changes will be tested it can be testing within ERP system it can be testing with the MES system
[00:12:23] for MES system there will be different team it can be the testing with the vendor or customer team because vendor and customer also needs to test their own functionality with this B2B communication so after all this testing successful there will be final preparation
[00:12:41] and go live phase where there will be user training because user needs to know what will be the new functionalities
[00:12:48] already some of the users like super users they will be testing those functionalities and after
[00:12:54] they are testing certification only all these new changes will be going to the production
[00:12:58] so for all other users there should be training to operate the new functionalities and then for
[00:13:06] new site implementation there will be data migration like there may be some of the new
[00:13:12] inventory that needs to be updated in the new site and there will be different new beans that
[00:13:18] needs to be created so those are some kind of example for the data migration so after all this
[00:13:24] preparation training data migration strategy program then there will be go live so in this
[00:13:30] phase it can be technical go live and after that it can be business glow live sometime what will
[00:13:36] happen the bigger changes which can affect the whole ERP system globally those changes will be
[00:13:42] sent at the system downtime so the business go live will not be happening immediately that business
[00:13:49] go live will be happening after the technical go live so after go live obviously we need to support
[00:13:55] the system if there will be any kind of issues usually it will be saying as hyper care support
[00:14:01] like we have to resolve that issues very fast.
[00:14:04] So there will be very strict SLA for that.
[00:14:07] So after this hyper-player support,
[00:14:10] it will be handed over to support team
[00:14:12] and that will be the end of that project.
[00:14:15] So this is a pretty good flow
[00:14:17] of the ERP implementation methodology.
[00:14:19] So you did call out integrations.
[00:14:22] You give an example of that.
[00:14:23] You talked about data migration
[00:14:25] and then you also said in some cases
[00:14:27] there are enhancements needed.
[00:14:29] Do you have an example of an enhancement?
[00:14:31] Yeah, sure.
[00:14:32] So let's say we need to have one interface that needs to build so that we can send the inventory data to a customer.
[00:14:42] And we do not have this B2B communication with that customer.
[00:14:46] So we have to do the changes in ERP system and we have to do that testing with the file, whatever we are sending, whether that file is getting received by the customer or not.
[00:15:00] So this is a kind of enhancement that we have to do in our system as well as we have to do integration testing between customer system.
[00:15:09] So this is an example, like if you are building some kind of new interface.
[00:15:14] Could you maybe share an example of an implementation that went well and talk about what made it succeed?
[00:15:21] So I would like to share one of the biggest implementation that I was part of and that went very well.
[00:15:30] So it was SAP S4HANA implementation three to four years back.
[00:15:35] So as I said, SAP is one of the biggest ERP vendor and their latest technology is 4HANA.
[00:15:42] so hana is in-memory platform so the transactions of the erp will be very fast so that's the main
[00:15:51] benefit by going to sap s4 hana so earlier we are having on-premise system as i said earlier so it's
[00:15:58] a kind of evaluation so we went to cloud and to sap s4 hana cloud so that time three to four years
[00:16:06] ago there is no program available from sap like today it is available sap rise where you can go
[00:16:12] to private cloud that time it was not there so we have to basically go through our own methodology
[00:16:18] own framework own testing strategy own role strategy role means like the authorization
[00:16:24] what authorization will be having for the user so like warehouse manager will be having different
[00:16:29] authorization warehouse operator who is only doing putter or picking there will be different
[00:16:34] authorization and then all the transaction are kind of a changing with the new SAP S4HANA system
[00:16:41] so we have to go for rigorous testing and rigorous training because there are a lot of customization
[00:16:48] in our system so to adopt all this new process we really need to test each and every process so that
[00:16:57] nothing will fail in the production and we went for the big bang go live where all the 100 plus
[00:17:04] sites are affected worldwide like there are sites in US, Mexico, Brazil, in Asia and in Europe. So
[00:17:14] you can understand there are multilingual users in different time zone in different business process
[00:17:20] So it was really not easy to go with direct implementation.
[00:17:25] So we went with a little bit phased out implementation.
[00:17:29] But the final implementation of SAP S4HANA, it was kind of a big bang.
[00:17:34] So with this implementation, there is a good amount of savings like $1.8 million yearly cost savings.
[00:17:43] and there is a lot of implementation experiences that has been shared in different SAP conferences.
[00:17:51] So I would say it was a big strategic implementation that I would like to share.
[00:17:56] What was it specifically during the implementation that caused the success?
[00:18:00] I would say it's a proper planning, team collaboration and team coordination because there are multiple teams.
[00:18:07] There is a business team, there is a technical team, there is partner teams, there is a different vendors team.
[00:18:14] So it's a proper planning that is key of success.
[00:18:18] And there is a lot of webinars that we have done.
[00:18:22] We have gone for the different languages also for doing the user training.
[00:18:28] And obviously there is a technical upgradation.
[00:18:32] So that has been heavily supported by ERP vendor SAP.
[00:18:36] So all these things all click together and that the main success we got with very little downtime and disrupting the business So proper planning was a key to the success Now could you share an example of an implementation that went sideways and why and what were the early warning signs
[00:18:55] Yeah, I can give a recent example from our first SAP BTP project in inventory management,
[00:19:02] where pre-receiving process was defined and implemented in more than 40 sites globally.
[00:19:08] but initially it was a failure so what was the reason so what is SAP BTP let me tell first
[00:19:15] so SAP BTP is the latest innovative platform from SAP where you can develop your own solution
[00:19:24] that will not be affecting ERP and all the solution will be connecting to ERP with different
[00:19:31] APIs so that is the advantage so you are not doing a lot of customization to the ERP system
[00:19:37] so now in this case we have developed this pre-receiving platform where user
[00:19:43] will not be able to enter the material whatever is coming from the vendor
[00:19:48] because those materials will not be having the correct number or maybe the
[00:19:54] purchase order is not having the capacity for doing the good receipt or
[00:19:58] maybe some wrong descriptions or some wrong values are there so whatever is
[00:20:03] the reason user is not able to do the good receipt in the ERP so that's a
[00:20:06] business problem so after developing this platform what will happen buyer can go into this platform
[00:20:13] and see what was wrong with that particular good receipt and then buyer can correct that so with
[00:20:19] these functionalities we went live with all the processes and testing everything but after going
[00:20:27] live user came back and told that they are not able to do the physical inventory process so what
[00:20:35] is that physical inventory process like you will be going to the warehouse physically and you will
[00:20:39] be counting that and that is needed for their daily checking of the inventory so then again we have to
[00:20:47] stop so from project management point of view it was a strategic project so it was a very big failure
[00:20:52] there was escalation so we have to again start from the beginning so there will be more budget
[00:20:58] obviously needed because the project team will work with the new functionalities and then after
[00:21:02] doing all these functionalities implemented like the physical inventory and then the other
[00:21:07] functionalities were there. They will be required different printout and different reports. Then it
[00:21:13] went live and now it is in 40 sites. It is running successfully. So it's a lesson learning and doing
[00:21:20] the correct business process testing and then it's a success. Those are good examples. Just thinking
[00:21:26] through them, the two examples that you walked through in the failed case, nothing really broke
[00:21:31] technically, just that a real operational process wasn't modeled or tested and surfaced only after
[00:21:38] go live. In the successful rollout, there was proper planning, testing, training, and good
[00:21:44] coordination. With that contrast in mind, let's move to AI. What's really new with AI? Is it adding
[00:21:53] brand new functionality or helping with some of these issues by surfacing missing processes earlier
[00:21:59] or by reducing cost and effort?
[00:22:02] Yeah, it's a very good question.
[00:22:04] So ERP is a static transaction, right?
[00:22:08] So all our static rules, static workflows,
[00:22:10] everything is kind of static.
[00:22:12] Now with AI means it's artificial intelligence, right?
[00:22:15] So we are getting predictive analytics from ML model.
[00:22:19] So it is basically increasing the efficiency,
[00:22:24] how the ERP process will work.
[00:22:27] Can you maybe just spend a minute
[00:22:28] explaining practically what's the difference between ML or predictive models and generative
[00:22:34] or LLMs? Yeah, sure. So ML is machine learning algorithm. So that means we are getting predictive
[00:22:41] analytics kind of functionality from there. So it is not giving any kind of generative functionality
[00:22:49] like in with whatever we are getting from JNAI or LLM. So where this kind of JNAI and LLM
[00:22:55] functionalities are creating new content, like creating new text, creating new documents,
[00:23:01] creating new images. So these are the basic differences between these two.
[00:23:05] Let's start with ML, which is, as you explained, the predictive side of AI.
[00:23:10] What are some of the high-impact ML-driven capabilities that you are seeing in ERP today?
[00:23:16] Yeah, so I can give one example from my earlier implementation with third-party EIML into SAP,
[00:23:25] like that's the ERP that we have implemented we have integrated that third-party ML so this is for
[00:23:31] having the bin efficiency where we are talking about more than 20,000 bins in electronics warehouse
[00:23:39] so picking cost is more effort for the due to large amount of bill of materials and it's a big
[00:23:49] warehouse so warehouse operator needs to travel a lot. So just stepping back you're talking about
[00:23:54] the plan to produce flow and how that is enhanced with ML. Yeah. Okay, continue please. So in plan to
[00:24:02] produce what is happening that we have already discussed, there will be a production order
[00:24:05] and for that we will be required the component. Now we are talking about the components,
[00:24:10] how to get it from warehouse and I also talked about if any component will be needed. So that
[00:24:15] needs to be coming from the warehouse. Now here to get it from the warehouse, the warehouse storage
[00:24:21] beans which will be really near to the production floor and if the warehouse operator is going to
[00:24:27] those beans and taking all the components then this picking will be very very faster and the
[00:24:33] production order will be fulfilled very fast but now let's think over like warehouse operator needs
[00:24:40] to travel a lot and they needs to travel to the end of the warehouse and then they will be coming
[00:24:45] from there so the travel time will be lot so to mitigate that risk and to make that picking
[00:24:51] efficient for the faster manufacturing order this ML model is required here to have that kind [00:25:00] The predictive analytics, which will predict what will be the production schedule for the future and then how the material is getting stored according to the layout of the warehouse with the bins which are near to the production flow.
[00:25:16] How does this data actually get supplied to the model? Isn't the model already trained?
[00:25:21] in this case the model was not trained with erp data because we are talking about third party ml
[00:25:28] model which are very general model which are not erp specific pretend model i would say i can give
[00:25:34] some of the examples like k-means clustering linear regression xg boost so those kind of models
[00:25:40] are basically tested so for this we have to really train those ml models with erp data we have to
[00:25:48] send all this ERP data like inventory data then production schedules data then
[00:25:55] that demand production demand data because all these parameters are needed
[00:25:59] and also the physical bin size and the package size because otherwise it will
[00:26:05] not be possible to calculate how much space is needed in the particular
[00:26:10] storage bin so with those data has been sent to different data lakes and from
[00:26:16] their ML model got this data and getting trained and then come back with the bin solution.
[00:26:23] And we got it from that in the ERP.
[00:26:25] So you mentioned training the models.
[00:26:28] Don't vendors supply ML models out of the box?
[00:26:31] So where are other use cases where you would extend or augment these models versus training
[00:26:36] them?
[00:26:37] Yeah, it's a very good question because this was a question earlier in my mind also.
[00:26:42] So with embedded AI functionality or the AI capability which are given by the ERP vendors, ERP vendor is already supplying the pre-trained ML model.
[00:26:53] Now, in my earlier example where we need to really train the ML model for custom AI or third-party ML model like K-means clustering or linear regression or XGBoost.
[00:27:07] So those ML models which are not having any context of ERP data that needs to be really, really trained.
[00:27:13] Now for embedded AI capabilities or where ERP vendor will be supplying the model.
[00:27:20] For an example, I can give very latest innovation from SAP, which is called SAP RPT1.
[00:27:26] It's a pre-trained transformer model.
[00:27:29] So this model is basically trained with SAP ERP data with programming context also.
[00:27:36] So you do not have to train a lot and it can be also validated with different testing.
[00:27:44] So that means whether it's a vendor supplied model or whether it's a third party ML model or the custom AI, you have to really test to see whether you will be getting the desired output or not.
[00:27:57] So the inventory management example about placement of inventory in the right bins at the right place.
[00:28:04] Are there other examples?
[00:28:05] I can give one example that I have seen the capability like it's anomaly detection and here
[00:28:13] I will be talking about the models which is already supplied by the vendor so in this case
[00:28:20] it will be SAP, SAP Joule I will be referring so Joule is the conversational AI that means it's a
[00:28:27] chatbot and it's a ai chatbot like a chat gpt but jool will be sitting within sap erp that means it
[00:28:36] will not be available outside like we are doing in chat gpt and jool will be having capability with
[00:28:43] all the erp context so there will be sap knowledge graph with that jool will be having all these
[00:28:50] capabilities for SAP context. So in chat GPT if we are going and typing we will not get that kind of
[00:28:58] ERP context or the inside the organization ERP data but here if you are comparing that you will
[00:29:05] understand SAP Joule will be having that kind of capability. So now for one example anomaly
[00:29:12] detection. So that is actually coming with SAP Joule. With anomaly detection if any equipment
[00:29:19] that will be failed that will be predicted with that particular ml model like k-means clustering
[00:29:27] which is basically running in in the background and getting the data from iot devices from the
[00:29:35] manufacturing line so with the historical data that kind of ml model will be basically setting up
[00:29:42] one baseline if there will be any kind of difference data that they will be getting that
[00:29:48] that time that particular ML model will be basically notifying well so this particular
[00:29:55] parts that can be failed maybe in 10 days so there will be actionable insights and SAP Joule will be
[00:30:04] notifying the users if there is no stock you have to procure so do you want to create one purchase
[00:30:12] ordered. So this kind of capability we can see in future in the ERP system. You mentioned
[00:30:18] knowledge graph. Can you provide some insight into that? So SAP knowledge graph will be having
[00:30:25] all this SAP insights, means SAP context, like what is there in the ERP business processes,
[00:30:33] inside the ERP business data, like what event will be happening after what. So this is a kind of a
[00:30:40] knowledge graph that will be used by Joule. And you mentioned in this case, you're able to
[00:30:45] directly use the vendor supplied ML model. There is no training or fine tuning required.
[00:30:51] In this case, it is not because this is one capability that is getting provided by SAP Joule.
[00:30:57] Like if SAP Joule is there in the ERP, means in SAP, then all the underlying ML models that will
[00:31:05] be active and if you are activating that functionality basically you have to
[00:31:09] activate all these functionalities so if you are activating that functionality for anomaly detection then it will be actually already pre So you do not have to train to get the desired output
[00:31:21] But isn't there a difference in the data that SAP would have trained on and what is
[00:31:26] valid anomaly at the customer side? Yeah, definitely. Because, you know,
[00:31:32] I have told about setting up the baseline, right? So the setting of the baseline is actually a
[00:31:38] framing of the enterprise-related data. So already that particular ML models is having
[00:31:46] the knowledge for SAP data, but the enterprise-related data will be there to train the
[00:31:53] model and that will be basically happening in the background so that they will be setting one
[00:31:57] baseline with the historical data. Where do you think most enterprises are today in terms of being
[00:32:03] ready for ML with ERP? Yeah, so enterprises are basically passed with the experimental phase.
[00:32:11] Now everybody is looking for implementing AI functionality and looking for more use cases,
[00:32:17] more real-life example. So with those use cases and real-life examples, those will be the
[00:32:23] parameter to decide whether leadership will be investing in AI or not. Let's shift to Gen AI now.
[00:32:30] So as compared to ML, how is Gen AI changing the way in which people interact with ERP?
[00:32:37] The examples you gave for ML were more around enhancing already existing processes.
[00:32:44] Is that right?
[00:32:45] And then with Gen AI, is that the same or that's changing basically the model of interacting with ERP?
[00:32:52] JNAI will definitely change the interaction with ERP because the thing is with ML we are getting predictive analytics right so it is only giving the insight now with JNAI it will be adding the conversation layer with natural language processing and like my earlier example SAP Joule so with SAP Joule it's not only predicting it is actually guiding
[00:33:22] the user like a colleague so there will be in oracle also there will be oracle ai agents which
[00:33:27] can summarize the documents orchestrate the workflows and then even create different business
[00:33:35] documents like purchase order maintenance order or production order so these are the functionalities
[00:33:43] that will be getting from jane ai where it will be working as a colleague it will be helping the
[00:33:51] user with actionable insights. As you explained, the conversational interface is really on top of
[00:33:58] the existing ERP workflows and context. Underneath, these workflows could still be largely deterministic.
[00:34:05] For example, how procured to pay or plan to produce executes would not be fundamentally changed.
[00:34:13] But what has changed both in your examples is that machine learning is making those deterministic
[00:34:18] flows better, like using ML to optimize the warehouse placement bin so that picking is
[00:34:24] faster, or anomaly detection to surface equipment or supply risks earlier before the workflow even
[00:34:30] hits an exception. However, as I understand with Gen AI, besides the conversational interface,
[00:34:37] parts of the workflow can start to feel more agentic where the system can choose which step
[00:34:42] or tool to invoke next based on the ERP context in this case.
[00:34:47] If that's true, how do you decide when a workflow should stay strictly deterministic
[00:34:52] versus when it starts to make sense for it to become more agentic?
[00:34:58] So deterministic orchestration is nothing new.
[00:35:00] It was there, traditional ERP also.
[00:35:04] So with time, it evolved.
[00:35:06] So like I can give an example, like PO release strategy means purchase orders,
[00:35:12] are created let's say for a thousand dollar now buyer is not having the approval limit of thousand
[00:35:20] dollar maybe buyer is having approval limit of five hundred dollars so if any purchase order is
[00:35:25] created with thousand dollar it will be going to purchase manager now if it will be created
[00:35:30] with maybe five thousand dollar then it will go to senior purchase manager so with this hierarchy
[00:35:37] and approval limit it's called release strategy in sap so this is a kind of example for the workflow
[00:35:45] where that deterministic orchestration is happening now let's talk about the ai agents
[00:35:52] so with ai agents there will be the task that really require the adaptability that will be
[00:35:59] requiring the problem solving it's not just a programming or the static workflows that will be
[00:36:06] just working with like like the pure release strategy so for an example Joule can identify
[00:36:13] high-risk suppliers I'm giving example from procurement side only like procure to pay
[00:36:19] so Joule can have that kind of ability by analyzing the supplier performance then compliance
[00:36:27] scores and potential supply chain disruptions because Joule can analyze the supplier delivery
[00:36:33] data from there Joule can see whether supplier has supplied all the orders in
[00:36:40] that correct timeline or not that is very important because if supplier is
[00:36:44] not supplying the orders regularly then that supplier will not be feasible to
[00:36:51] supply big orders in the future so with that it can suggest alternative suppliers
[00:36:56] it can update directly the ERP documents for that supplier so that will be kind
[00:37:03] of a capability with Jule or AI agents. What is the right granularity for these agents?
[00:37:09] Let's talk about the same example. So I talked about creating the purchase order earlier, right?
[00:37:15] Like release documentation or release strategy. So now in this scenario, if you are talking about
[00:37:21] procure to pay cycle creating the purchase order that will be a granularity for AI agent but it not the full procure to pay cycle like the supplier collaboration or the supplier determination or the good receipt
[00:37:36] Because for good receipt, there will be different agent.
[00:37:39] So that will be the granularity level because it will be bounded by the business object and business object.
[00:37:46] It will be purchase order, sales order or production order.
[00:37:50] So in procure to pay scenario, creating the purchase order will be one agent.
[00:37:55] So as you mentioned, the vendors are delivering a lot of these agents out of the box. You gave a few examples on what these agents can do. What are you seeing in practice? Are you able to adopt these agenting capabilities as delivered by the vendor or is there need for extension?
[00:38:14] Yeah. So to answer this question, I need to really go to the different technical approaches or technical architectures like embedded AI, MCP, A2A.
[00:38:24] And I will be basically explaining all those.
[00:38:27] Before you get into the technical implementations, can you give an example of where you feel it'll be?
[00:38:32] Take an example of the agent that you had and explain why do you think it might be important to enhance it or extend it?
[00:38:40] so for an example I talked about the agent where we are doing anomaly detection so with that anomaly detection what is happening we are only getting the actionable insights from Joule that I need to create the purchase order now let's think about one scenario where there will be an agent or there will be much more capabilities required like whether it will be a purchase order or whether it will be a purchase order or whether it will be a
[00:39:09] plan to plant transfer order so if that actionable insights are only giving purchase order
[00:39:18] functionality then we need to actually write some kind of custom codes or do some kind of adjustment
[00:39:24] to have that kind of functionality where we will be getting that particular component
[00:39:30] from different plant so if that is not learned by the actionable insights of Joule then it will
[00:39:40] not be there so I'm just you know trying to give one example where it can happen that we'll be
[00:39:46] having the agents which is already or ML models which are already supplied by supplier but
[00:39:54] sometime it will not be able to fulfill the whole business requirements. Thanks about that
[00:40:00] Now we can get into the technical aspects that you started to talk about and you mentioned MCP or model control protocol and A2A or agent to agent.
[00:40:10] Before you get into the details, maybe give a quick intro on what is MCP and A2A.
[00:40:16] I'd also like to point our listeners to episode 689, Amai Desai on model context protocol.
[00:40:22] So the approaches and technical architectures of this MCP and agent to agent, I would say these are kind of in a paradigm shift and it's really latest innovation in ERP space because all the big ERP vendors like SAP, Oracle and Microsoft Dynamics, they have included this kind of protocols directly into ERP.
[00:40:49] so i can give some example then then i can start like for embedded ai we talked about embedded ai
[00:40:56] right so with embedded ai we have talked about sap jewel and then for other vendors just to
[00:41:03] have some kind of idea like for oracle it will be oracle fusion ai agents then for microsoft
[00:41:10] dynamics it will be microsoft dynamics copilot we all are aware of copilot but with microsoft
[00:41:16] dynamics co-pilot it's a kind of embedded ai agent that you can use also now coming back to
[00:41:23] mcp and a2a so mcp stands for model context protocol so this is defined as an open source
[00:41:32] standard how ai application will communicate with external systems through a unified interface like
[00:41:41] how it is happening without MCP right so if any AI agent which is a foreign
[00:41:47] AI agent or customer agent it can communicate with API or it can
[00:41:52] communicate with different method where we'll get all the communication from the
[00:41:58] third-party ML model and will store in ERP right that we talked about now with
[00:42:04] MCP this will be really easier because MCP will be the medium for communicating
[00:42:10] with any kind of custom AI with ERP so having said that MCP will be act as a server and custom AI
[00:42:22] will be doing query to MCP and then MCP will be having all this ERP context from ERP module and
[00:42:31] get back to the AI agent. That's the way it will happen. So let's say for an example, if any kind
[00:42:40] of SAP agent wants to reconcile accounts or supplier invoices. So in this kind of scenario,
[00:42:47] what will happen? It will query to MCP server and then MCP will be querying to the invoice
[00:42:54] management module. And from there, it will be getting all this context back to the AI agent.
[00:42:59] so in this case the vendor is actually exposing their business logic through the mcp protocol yes
[00:43:06] that's correct so it will be kind of a unified platform earlier we have to create a separate
[00:43:12] separate api for let's say separate separate agent or for separate separate integration of
[00:43:18] different tools here we do not have to do that because if mcp will be there mcp will take care so
[00:43:24] So the traditional API method, it will be better than the traditional API method, I would say.
[00:43:31] Are you seeing a lot of functionality and business logic exposed as MCP servers?
[00:43:36] Well so this is a really emerging technology and we have seen the news or capabilities that is basically included in different ERP systems like recently in SAP TechEd SAP announced about this capability So we are yet to see how it will be happening
[00:43:54] in future. We would like to see more use cases and real-life examples to understand how it will
[00:44:01] be happening. Talk about A2A now, maybe with a concrete example. Yeah. So again, A2A is also
[00:44:08] emerging technology that is agent to agent communication so it can be custom ai agent
[00:44:16] or it can be the agent which is provided by the erp vendor so now with mcp both agents are
[00:44:24] integrated with erp right now if this two kind of agents what is provided by erp vendor or the
[00:44:31] custom agent if they will be talking with each other it will be the agent to agent communication
[00:44:36] So for an example, if a supply chain planning agent interacts with an inventory agent and a procurement agent for proactively managing the stock levels so that there will be new purchase order that needs to be created to place to the vendor so that the stock will not be having any kind of shortages.
[00:45:01] So in this kind of scenario, what will happen?
[00:45:04] This supply chain planning agent will be talking to inventory agent and procurement agent with A2A protocol.
[00:45:12] Do you envision using agents across multiple vendors?
[00:45:16] Yes. So that's the capability basically.
[00:45:18] So like with MCP, it is possible.
[00:45:21] Like if there will be different, different vendors are creating, because it doesn't say that it has to be any ERP vendor specific agent.
[00:45:29] It can be anything like it can be custom agent also.
[00:45:33] So that can be easily integrated with MCP2 ERP.
[00:45:35] You touched upon this a little bit, but maybe could you summarize on what determines whether the vendors Gen AI works out of the box versus needing customization?
[00:45:45] So with out of the box, like we are getting the use cases.
[00:45:50] So all these use cases are very specific to the ERP software, right?
[00:45:58] So it's not specific to a company.
[00:46:00] It's not specific to a industry.
[00:46:02] So we need to have that kind of use cases, which is specific to the similar company or
[00:46:08] similar industry.
[00:46:10] So because all the use cases are very generic, right?
[00:46:13] So now one size will not fit for all because every enterprise will be having that different
[00:46:21] business processes, different business requirement.
[00:46:23] So, for any kind of standard ERP process, when we are doing a lot of customization to fulfill the business requirement and to have that kind of capability, whatever is needed for a particular company.
[00:46:38] So let's say for electronics manufacturing, whatever capabilities are needed that may not require for one auto manufacturer.
[00:46:45] So these are the different capabilities will be required for the agents which are basically provided by the ERP vendors.
[00:46:55] So this is where we need customization if the business requirement is not getting fulfilled.
[00:47:02] With that, let's get into our final section or discussion about how ERP implementation is changing now that AI is part of the implementation.
[00:47:11] To start with, Sukhujit, what is the process to evaluate upcoming AI enhancements, either what vendors are supplying or the new models or new technologies?
[00:47:20] So with ERP implementation methodology, we talked about six to seven phases.
[00:47:25] Now with AI integration, it will be required similar kind of stages, but with AI steps, obviously, because we need to have more AI specific activities now. So the first stage, I would say it will be AI evaluation and strategy. So where with the business problem discussion, we need to evaluate the strategy for AI.
[00:47:50] So we have to talk about the data strategy of the organization.
[00:47:55] I'm not talking about any kind of one project.
[00:47:58] I'm talking about the organization-wise policy and I'm talking about the organization-wise data strategy.
[00:48:03] Because with AI, there will be a lot more dependency with the data security.
[00:48:09] So after this, there will be business process analysis and data integration.
[00:48:15] And where there will be discussion from the earlier data strategy.
[00:48:20] how it will be connected with AI. So with embedded AI functionality, it will be a little bit easier
[00:48:27] because the data directly will be connected with the embedded AI. So you do not have to do a lot
[00:48:34] more investment there. But with custom AI functionality, we have to do a lot more investment
[00:48:42] or strategy for integrating the data because data is the key, as I said earlier. And then we need to
[00:48:48] think about the AI model selection training so with embedded AI functionality if it will fulfill
[00:48:56] your requirement then we do not have to think over we can easily utilize the already pretend
[00:49:02] models from ERP vendors but if it is not fulfilling the business requirement so in that case we have
[00:49:12] to really think about how to do the customization on top of the vendor provided models or if we have
[00:49:20] to take some different custom models and with MCP we can really integrate with ERP. So some of the
[00:49:28] models that I can tell like it was decision tree, random forest, k-means clustering, linear regression
[00:49:35] those are the models actually we looked earlier and these are very popular models and then after
[00:49:42] model selection there will be the integration of ai model if it is embedded again we do not have to
[00:49:50] think about that but we need to do that testing so that is really important embedded ai model
[00:49:57] may not be giving you the desired output [00:50:00] Because with AI implementation, you will not get the desired output on the first day.
[00:50:04] It will take time because the model will learn.
[00:50:08] And with pre-tend model, it will be a little bit easier.
[00:50:10] But with the custom AI model, it will be a little bit more time for getting the desired output.
[00:50:17] Then in the next phase, we have to do that AI adoption preparation.
[00:50:22] This is really important because organization change management is very critical in this scenario.
[00:50:29] We need to go for the expandability training. So we need to train the user so that user will not be having that kind of feeling that AI will be replacing their jobs. AI will be helping them in the operations. And then after all this testing, training and all this policy in place, organization needs to sign different documents also because AI security policy and AI policy will be at organization level as I said earlier.
[00:50:58] then after that AI deployment will be happening so with AI deployment as I said we'll not get the
[00:51:05] result on the first day so we need to really monitor the changes whatever is happening with
[00:51:11] the AI model deployment and we need to see whether we are getting the desired output or not
[00:51:17] if we are not getting then we have to evaluate where is the fault like whether we need to
[00:51:24] use more custom things to have that on top of the embedded AI or our custom AI, whatever we are
[00:51:32] using, whether we have to change that model or we have to fine tune that model. So basically,
[00:51:37] these are the phases that I can think over with AI implementation. Even stepping back a little bit,
[00:51:43] what's your methodology to even research on what are the latest and greatest technologies
[00:51:48] or models? You mentioned several models. How do you keep up? So whatever models I have mentioned,
[00:51:55] those are actually available in GitHub. So as a ERP consultant, there will not be that kind of
[00:52:03] capability to select the models. Obviously, we need to work with the AI architects and the data
[00:52:09] scientist but the training the business process this is the area where ERP
[00:52:17] consultants will work very closely with AI architects and data scientists because AI architects or data scientists will not be having the pleasant idea about the business processes so those models will be trained with the ERP business data right so we have to really prepare the data as
[00:52:36] a ERP consultant and the correct data so that is where ERP consultant roles
[00:52:42] will be very critical so after selecting and testing only validation will be with
[00:52:49] the business data, right? So in that case, if one model will not perform with the desired output,
[00:52:57] then we have to see different other model or we have to fine tune the model. Everything is
[00:53:02] depending on the testing result. Thanks for covering all the different new phases,
[00:53:08] but a lot of your focus with it was in explaining that testing and monitoring is very critical,
[00:53:14] especially to make sure that you're actually getting the benefits of AI. You didn't really
[00:53:18] talk about extending with MCP and A2A, which we covered earlier in the Gen AI context. Why is that?
[00:53:25] Basically, we do not have any use case to tell, frankly speaking, because this is really a new
[00:53:30] capability that has been shown by SAP. I'm talking about SAP because I have integrated all these
[00:53:36] things with SAP. So not only SAP, if you are talking about some other big vendors like Oracle
[00:53:42] and Microsoft Dynamics.
[00:53:45] So a lot of this MCP and A2A are integrated in this year.
[00:53:49] So we are here to see the use cases.
[00:53:52] We are here to see the pilot implementation
[00:53:54] or at least small kind of implementation with ERP partner.
[00:54:00] It doesn't have to be have that big customer or something.
[00:54:04] Somebody can do prototype and show,
[00:54:06] well, this is happening like this.
[00:54:07] So we are here to see all these things
[00:54:10] to have that kind of validation.
[00:54:12] What are the roles or skills that now become essential for an implementer to have?
[00:54:19] So I would say for AI integration and whatever we did, like we are having our ERP team and then we are having our AI center of excellence where there are AI architects and data scientists.
[00:54:32] So now whenever we are talking about two different capabilities like ERP and AI together and all the ERP vendors are giving all the AI capabilities as an embedded kind of a model.
[00:54:46] So we need to more focus on the data science as a ERP consultant and we need to know the basic AI models how it will work like for an example SAP RPT1 So if you are having any kind of CSV file you will be
[00:55:02] uploading to SAP RPT1 and then you will be trying to ask some question there. So this is a latest
[00:55:09] innovation from SAP and that has been released. So I am trying to say that there will be trend
[00:55:16] where it will be easier for a ERP consultant having the basic knowledge of AI with the next
[00:55:25] level evaluation from ERP vendor so that ERP consultant can evaluate the models easily.
[00:55:32] So what's your advice for teams starting to integrate AI with ERP? What is the most key
[00:55:37] technical investment or organizational investment they should make early in the journey?
[00:55:42] Yeah, from my experience, obviously it will be data. I told earlier also data is the key and then that AI mindset. So there will be AI awareness training all over the organization and there will be AI security policy. And these are all organizational mandate. This is not particularly for one ERP project. So there will be a lot more systems. So whenever we are talking about AI, it will expose a lot of business data.
[00:56:08] To mitigate the risk, we need to have AI security policy all over the organization.
[00:56:14] And then there will be a lot of training on AI awareness, not only ERP.
[00:56:19] And after that, whenever ERP AI will be implemented, like we'll be integrating AI within ERP with different business function,
[00:56:28] that particular business users, they have to be trained wisely so that they can understand the value of artificial intelligence.
[00:56:37] Is there any topic that we haven't covered, Subhidu, that you'd like to drill into?
[00:56:41] Yeah, I'd like to mention that in recent days, I can see a lot more traction or interest from organization perspective.
[00:56:50] Like in my organization, there are a lot of workshops are happening for having AI in place into ERP, all the AI capabilities with different ERP partners, different ERP vendors.
[00:57:03] So one point I would like to mention, in each and every workshop, we are looking for the use cases.
[00:57:09] We are looking for the pilot implementation at least by the ERP partner So those are basically missing Like there are a lot more AI capabilities which are explained on the powerpoint but that may not be useful for a company which will be investing a lot of money on ai innovation so if there will
[00:57:28] be a lot more use cases a lot more real life example there may be small kind of pilot implementation
[00:57:34] so erp vendor will be basically collaborating with customer and doing this kind of pilot
[00:57:42] implementation or small, small use cases. And then that will be also a good point for selling
[00:57:48] the product for ERP vendors. Thanks for that. Validation that there's a lot more happening
[00:57:53] with AI now. How can listeners contact you or learn about your work?
[00:57:58] Please check out my IEEE papers, AI redefined implementation methodology for ERP,
[00:58:05] and then digital twin integration with AI enabled ERP because digital twin is an
[00:58:11] important part for any kind of manufacturing scenario so where digital twin have that predictive
[00:58:18] thing for having new product or new product line and in this context we can get all this
[00:58:25] digital twin capability that actionable insights directly into ai and all that capabilities or
[00:58:33] insights can be used in the ERP. So apart from that, I will be having IEEE webinars,
[00:58:41] and I am also an active member of American SAP user group, where I will be having presentations
[00:58:48] in SAG chapters, meetings, or in SAP software. You will put some of those links in the show notes.
[00:58:56] Thank you so much for coming on today, Svajit. Thank you very much. Thanks for having me.
[00:59:00] Thanks for listening to SE Radio, an educational program brought to you by IEEE Software Magazine.
[00:59:07] For more about the podcast, including other episodes, visit our website at se-radio.net.
[00:59:12] To provide feedback, you can comment on each episode on the website or reach us on LinkedIn, Facebook, Twitter, or through our Slack channel at seradio.slack.com.
[00:59:23] You can also email us at team at se-radio.net.
[00:59:27] this and all other episodes of se radio is licensed under creative commons license 2.5
[00:59:33] thanks for listening