Saurabh Gupta — President, HFS Research[00:21]
Welcome, welcome everyone, to another episode of HFS Unfiltered. And today we’re going to be talking about how to turn AI ambitions into reality. Look, AI has dominated every boardroom conversation, every CFO conversation that I’ve had over the last two years now. And one of the jokes that I’ve had is that AI is dying a death by 1,000 POCs. I invited Khalid, my good friend Khalid, who’s the global offerings lead for finance operations at IBM Consulting, to tell us about some of his clients and his experiences where POCs have actually gone into production in the finance operations space. And to talk about what the outcomes are, so we can see whether some of our AI ambitions can actually be turned into reality. So, thanks for joining, Khalid.
Khalid Siddiqui — Global Offerings Lead, Finance Operations, IBM Consulting[01:22]
No, it’s a pleasure. It’s always great speaking with you. Thirty minutes provides me at least six months of intelligence as I speak with you. So it’s also humbling to be part of this. Good to be here.
Saurabh Gupta — President, HFS Research[01:41]
Fantastic. So, Khalid, tell me, as I was just mentioning, we are seeing a lot of interest in leveraging AI, whether it’s predictive or machine learning, whether it’s GenAI, and now agentic AI, in helping finance operations, especially a lot in order-to-cash. Tell us about your experiences of where you actually moved the needle from POC to production.
Khalid Siddiqui — Global Offerings Lead, Finance Operations, IBM Consulting[01:41]
No, it’s a great question, Saurabh. What we’ve been seeing, at least interacting with the C-suite and the stakeholders in our line of business, is that a lot of them are extremely apprehensive as to whether the impact and effect is actually real. Because of that apprehension and those questions, the adoption of AI is still not what we want it to be. And hence, you’re right by saying that AI is dying through 1,000 POCs. What we’ve been doing differently from an IBM perspective is that any POCs that we’re undertaking, we’re undertaking with actual client data and actual client challenges. So what I mean by that is one of our very good clients basically stated that if you’re able to execute something within 6 to 10 weeks and show benefits, it’ll be a great story. So what we are also inferring from understanding the different lines is that a POC should be short, it should be quick, but more importantly, show tangible benefits in an area. The client that I’m going to quote is a building-materials client where this team of about 200-odd resources manages about 1.2 million queries annually. Now, if you understand the magnitude of 1.2 million queries, this is coming through different modes of communication — through WhatsApps, emails, Teams, Slacks, and any other chat or SMS mode. And all of these queries relate to a product being delivered from point A to point B. Now, in a building-materials industry, the product, which is cement, as it gets delivered, if it doesn’t go to the right place, it’s actually solidified, so you’re actually losing the value of that product as the delivery takes place. And during this process, there’s a multitude of queries: has the order been released? Is there a proof of delivery? Is there a statement that I can see? Is there an invoice I can see, so that the effective services can be provided? And here, what we see is that not only are the queries important, but the queries are real time. So we took a portion of these queries and said, is it possible for us, for a particular segment, to streamline the way all the queries are coming in and our ability to then process them. So just to give you some numbers, an average query takes anywhere from 29 to 30 minutes to process, and they need to then get connected back to the system to ensure that an auto block is released, a credit limit is increased, an invoice payment passed through. So we were able to take a segment for this customer and not only bring down the processing time from 30 minutes to 2 minutes, but also create entries within the ERP system that also processed the order. It sounds extremely fancy, but if you look at it, there were multiple agents that were brought to life by doing this, and that’s how the POC got enabled. One agent had the ability to bring all the queries from multiple channels into one single channel. Another agent was able to categorize and bifurcate the queries — are the queries commercial, are the queries contractual, are the queries billing related, are the queries customer-service related. One agent, for example, if it was an order-release query, was triggered to understand the financial risk of a customer and then understand the credit worthiness by running a risk-worthiness assessment check. And one query was an integration query that basically triggered the respective entry into the ERP system. So we created a bunch of, let’s say, 5 capabilities within 5 agents that were orchestrated so that when a query comes in, it was able to execute it and hence bring down the person time. So when we started this engagement, it took us about 12 weeks to complete this exercise, but then we expanded to not only what IBM managed, but also what the customer managed. So it was both for the customer organization and the IBM organization that resulted in almost a 60% efficiency in query management. Now, what has also happened with this is that not only were we able to get a product faster, but because of the resolution time we have on proof of deliveries, invoice statements, and analytics on credit worthiness, we also improved cash flow. So when we actually ended the year, we delivered about 994 million, which is about two days, for that region within the same financial year itself. So what worked in our favor was the ability for us to implement the POC, expand it to a broader organization, and then deliver benefits within the same financial year. Now we have a plan to extend it to almost the field and sales teams — that’s almost 1,000-odd resources — and then they will extract benefits based on the interconnecting capabilities. So what has happened is that it was also a learning for IBM that it is not about query management and creating agents to solve for queries, but an ability to link different agents so that information can be extracted in whatever form or shape and then provided. And that example that I just talked about is easily replicable in supplier management, easily replicable in supply chain, and also in controllership activity. So this was more a receivable management, or trade receivables, type service, but it’s going beyond what we’re doing. So this is, let’s say, a snippet of what we’ve been able to do with this particular client.
Saurabh Gupta — President, HFS Research[09:10]
That is fantastic, Khalid, because that’s agentic AI in action, but that’s also what we at HFS — you know, we coined this term “services as software” last year. That’s almost, if not completely, services as software; it’s services as almost software. And the question that I have for you then is, you have hundreds of other clients in finance operations and in consulting. Why are not all of them jumping onto this? This is almost like a new S-curve of value creation. This is the kind of stuff that we started to expect when we unearthed offshoring. We all remember in the early 2000s, that’s the kind of impact that this has. So why aren’t others jumping onto this all over the place?
Khalid Siddiqui — Global Offerings Lead, Finance Operations, IBM Consulting[10:18]
Within IBM we have also created something called an AI fabric, which is an orchestrating agent layer that orchestrates different agents to execute controllership, receivable, supplier management, HR, talent, and customer-service type activities. Now, for that ecosystem to be brought to light, what is very, very important is data — the availability of data, the availability of integrating systems, and our understanding of what skill of the agent will be applicable for what business process and function. A lot of our clients want to go into the journey, but they also are trying to solve for, okay, am I able to bring data into a place, or whatever data I’m having in different data repositories, is it clean? Step number 1. Step number 2, how do I simplify the landscape of my system of records and engagement? There could be a client with multiple legacy systems. There could be a client with a single ERP. So the complexity of their system of engagement and their system of record also poses a challenge — how many integrated connectors are we going to make with our agents to make it run? And then thirdly is also change management. So whether we like it or not, people like to see the same work being done through the hands into the eyes, and there’s an amount of change that is also required to make it happen. So I’ll give another great example, where for a telecom company, we’ve been able to improve the way billing queries are being managed to ensure we’re able to collect and improve their international collections quite effectively. But what we brought into play was analytics — agents that were automating and matching bill-inquiry information. And then our ability to then create the next best action for different collectors for their customer so they can execute. Again, it is not about a collector saying that the next best action is going to be X, Y, Z, but linking queries from a billing perspective, matching it to their account statements, and executing it. Traditionally they already had an extremely heavy analytics-based solution, but with this it exponentially improved the way they’re doing it, and we almost generated close to about 200 million in cash flow through DSO improvements even for them, but again through an agentic layer. Now what we’re seeing here is, let’s say, a trend around receivable management being a platform for GenAI to be created, but we’re also seeing this in FP&A and reporting as well. For another consumer-based client in the UK, we’ve been able to help streamline the multiple management reports that they are generating into a few. But more importantly, also generating insights. So this is where, you can call it, large language models have come to play. Automatic inferences for working-capital ratios, liquidity ratios, trends in margin, cash flow, cash-forecasting trends, and then inferring narratives have immensely helped controllers in this organization. To then wait for the report to be generated with AI based on the month close, and then edit it with a human in the loop and then send it to the CFO. So in this particular example, a controller spends an average 11 to 15 hours per market, and there are about 52 markets to enable a month close. So that’s a lot of controllers and a lot of work. We were able to crunch that 11 hours to almost 2 to 3 hours. It can still further go down, but there are still nuances with different regulations that are required. But again, in this example, it’s an ability to bring agents and also large language models, so that it simplifies the work for the controller, and it is not extremely complicated from a control viewpoint, because you have a simple user interface, you’re able to execute through it and then get an output. So, yes, clients are — we are slowly going on this journey, but there are still those three challenges that I talked about that they’re still trying to solve for, and we at IBM Consulting are helping them in that journey as well, at least for our clients. Because unless we help and then guide our clients as to how to make their life easier and change the way of working, we will not be able to expand into the different use cases that we’ve been targeting with our clients and expanding in different areas.
Saurabh Gupta — President, HFS Research[15:41]
No, these are some really fantastic examples, call it. So thanks for sharing that. And I think you’re absolutely right. I don’t think the problem is the technology. The technology is right here, right now, and it works. The 3 or 4 examples that you shared are enough evidence that this works. I think the problem is that we in our enterprises have collected a lot of debt. We have data debt, we have, as you mentioned, process debt, we have cultural debt, we have technical debt. And there is a limit to how much you can buy on your credit cards. At some point in time, you have to pay your credit-card bills, and I think that’s what we are struggling with. And I think if we are able to pay our debts, we will see a hockey stick in agentic AI and services as software. So, thanks a lot, Khalid, for sharing your views, for sharing some of these examples. I think these are super useful. The more that companies like you share these examples, I think the more enterprises sort of see the value in this, and that’s how we’ll move beyond the death by 1,000 POCs, Khalid. That’s my hope — that a few people who listen to this see the light at the end of the tunnel and feel that this is possible. So thanks a lot, Khalid, for taking out the time and sharing this with us.
Khalid Siddiqui — Global Offerings Lead, Finance Operations, IBM Consulting[17:08]
No, most welcome, sir, most welcome, and thanks for this opportunity. Like I said, 30 minutes with you gives me 6 months of intelligence, which is quite a bit in my lifetime, so that’s very good. We should do more of it. Thank you very much, Saurabh.
