gsriramana

Domain: Finance and Banking Processes We all know that AI agents and solutions are designed to respond quickly in Finance and Banking operations. AI solutions helps in making high impact decision and quickens the actions and speeds up the approval process, etc. Processes that took long hours or few days (like credit approvals, fraud decisions, payment releases, exception handling) are now completed in few seconds or few minutes. We all know that AI solutions helps in real time scoring, instant approvals and automated actions. This speed and responsiveness has improved efficiency and customer experience. At the same time, the increase in speed also has increased risk rather than reduce it. Acting quickly results in increase in errors, wrong assumptions that results in irreversible decisions. Let’s look at few examples from the Finance and Banking domain. Example 1: Fraud Detection In card and payment systems, AI has helped in detecting and blocking fraudulent transactions in few milliseconds. How AI acting quickly can lead to serious consequences if the : • Confidence is moderate • Context information is incomplete • Transaction appears suspicious due to location or amount but legitimate • Time sensitive purchases blocked instantly without confirmation AI agents & solutions can block the transaction instantly without asking confirmation and could stop critical payments leading to losing customer trust and increase in support interactions. How AI system should recognize that it needs to slow down: AI solutions should have the ability to recognize the need to slow down when it sees the signals as below – • Fraud probability is medium • High customer impact • First time variation In these cases, instead of blocking the payment immediately, the AI should pause and do the below – • Initiate step-up authentication • Request customer confirmation Example 2: Invoice Approval In accounts payable process, AI helps in quick invoice validation & approvals by automating the process and helping to reduce cycle time and manual effort. At the same time, approving too quickly can lead to lot of problems. How AI acting quickly can lead to serious consequences if the : • Checks are bypassed • Invoice technically matches purchase order but deviation in tax rate • Slight deviation from contract terms • Approvals are done based on patterns How AI system should recognize that it needs to slow down: AI agent and solutions should have the ability to recognize the need to slow down when it sees the signals as below– • Repeated deviation • Ambiguity in contracts • Materiality thresholds are crosses In these cases, additional validation or policy confirmation is required. Overall, an AI agent or solution should not look at speed alone but also adjust its behaviour based on confidence of the decision, potential impact and how quickly it can reverse its decision. Overall, the triggers for slowing down should look at – • Low or medium confidence prediction • High financial impact • High reputational impact • Conflicting signals • Regulatory issues • Compliance issues • First time scenarios • Complex scenarios If AI agent or solution sees the above conditions, it should pause, look for additional context, involve humans, apply a staged decision and escalate. Conclusion: AI agents and solution can act quickly but it also need to recognize when delay is a risk mitigation strategy. Effective solutions are the ones that combines speed with restraint and this balance will provide the difference between efficiency and intelligence.

Domain: Finance and Banking Processes We all know that AI agents and solutions speeds up the decision making in Finance and Banking operations. AI solutions helps in providing instant recommendations, quickens the actions and speeds up the approval process, etc. Processes that took long hours or few days (like credit approvals, fraud decisions, payment releases, exception handling) can now be completed in few seconds or few minutes. We all know that this speed has improved efficiency and customer experience. At the same time, this also has introduced the risk that decisions are made – · Before critical or important context is fully understood · Before it’s well-informed · Before it’s properly governed · With poor assumptions · Weak judgement Let’s look at few examples from the Finance and Banking domain. Example 1: Credit Approval In Credit approval processes, AI has helped in increasing the speed and improved the outcomes. How AI reduces decision time: Creating credit scoring model to reduce the time taken to approve loans · Quickly evaluating credit history and transaction behaviour · Approve or reject loan application in real time How AI introduces new risks or blind spots: At the same time, speeding up the credit decisions with the help of AI model may lead to poor assumptions, incomplete context, weak judgment, bringing in new risks and blind spots as below – · Not accounting for temporary income disruptions · Not accounting for life events · Increase in short-term loan volumes · Increase in long-term default risk · Fast rejections affecting customers · Increase in customer appeals What checks to put in place: To mitigate the above mentioned risks or blind spots, checks such as – · Confidence thresholds · Human in the loop for borderline cases · Periodic outcome based validation These mechanisms will help that speed does not create risks. Example 2: Invoice Approval AI has helped a lot in speeding up invoice validation & approvals by automating the process. How AI reduces decision time: · Automatically matching invoice to purchase orders · Automatically matching invoice to contracts · Reduces cycle time · Allows users to focus on exceptions How AI introduces new risks or blind spots: At the same time, faster approvals can increase the errors bringing in new risks and blind spots as below – · AI system may learn to approve invoices with minor tax discrepancies due to historical tolerances · Increase in audit findings · Small findings leading into material risk · Post payment corrections What checks to put in place: To mitigate the above mentioned risks or blind spots, checks or controls such as – · Exception trend monitoring · Dynamic risk scoring · Human in the loop for final approvals · Periodic rule based re-evaluation These mechanisms will help that faster processing does not create risks or weaker controls. Conclusion: AI agents and solution can significantly improve financial decision making when it eliminates unnecessary delays without eliminating meaningful judgment. The risk arises when organizations try to equate faster decisions with better ones and not reassessing the assumptions that are part of the models and data. AI systems must have human in the loop and deliberate checks. In finance and banking processes, the goal is not simply to decide faster, but to decide faster with discipline.

Domain: Finance and Banking Processes We all know that AI agents and solutions are getting implemented in Finance and Banking operations to remove existing constraints, reduce manual efforts, improve decision making, mitigate risks & improve the control mechanism. But in many cases, by removing few constraints, AI has created new kind of constraints like data dependency, governance delays, trust issues or it has shifted the constraints to another process or function. Let’s look at few examples from the Finance and Banking domain. Example 1: Invoice Validation In accounts payable processes, Invoice validation is a very common process where AI agents are deployed. This is done to address issues like – · Manual data checks · Exception handling By deploying AI agents and utilising capabilities like OCR, Machine learning – · Efforts are reduced · Speed of processing the invoices increase · Reduces the backlog · Straight throughput (STP) increases However, once validation is taken care by AI, a new constraint often emerges downstream. The bottleneck shifts to data dependencies, data governance and exception handling and new constraints arises along with signals as below - · High volume of exceptions · Increase in Human in the loop · Inconsistent vendor master data · Ambiguity in tax rules · Interpretation issues If the upstream master data quality is not good, AI agents likely to generate more exceptions and shift the issue to Exception handling teams. Example 2: Loan Underwriting In Banking processes, loan underwriting is a common activity where AI agents are deployed. This is done to address issues like – · High turnaround time · Inaccurate and inconsistent efforts By deploying AI agents, the models help in – · Automating the credit scores process · Automating the risk assessment · Improvement in speed, accuracy and consistency However, once automation is taken care by AI and AI decisions lack transparency - new constraints arises like policy interpretation, policy explanation and trust. Also the constraints shifts to other teams or individuals like relationship managers, risk officers and compliance teams. Due to this, there are issues along with new signals as below - · High rejections · Delay in approvals · Conflict with people · High volumes of manual review · Increase in escalations · Explanations to regulators for high rejections · Slow down in final decision making Conclusion: By going through the above examples, we understand that AI rarely eliminates constraints and also shifts the constraints from one team to the other. AI was good in increasing the speed, accuracy and consistency in the processing but it created other weaknesses and constraints like – · Data quality · Governance · Trust, Ambiguity & Interpretation · Readiness issues and showed signals like · Increase in exception volumes · Increased escalation rates · Increase in cycle time in the downstream process · Increase in dependency on exception handling teams

Domain: Banking I agree that in many organization, the processes runs on hidden assumptions. At times, we forget to ask questions and we solely believe our customers & vendors are always right and our workload, risk mitigation & behaviours are on the right track. This belief has become a part of the culture in many organizations and many do not question it and don’t realise it until something fails. Let me share few specific examples where there are few blind spots and how AI agents help to eliminate them and mitigate risks. Example 1: Duplicate billings in Invoice Validation process High-level steps: · Vendors send the invoice on monthly basis. · Accounting team checks for duplicates. · Approvals are done quickly due to tight timelines (say peak quarter or seasonal). · Duplicate invoices slip through occasionally. · Finance team discovers or misses duplicates during the payment. · Audit team discovers duplicates during the audit. Assumption: · Duplicate invoices are random errors and there are no patterns to it. Blind Spot: · Team missed to look at the seasonal behaviour. · Team missed to analyse year-over-year patterns. · Team assumed the duplicates were occasional mistakes. How AI agent helped: · AI agent performed analysis for each & every season. · AI agent found that the duplicate bills spike during peak & holiday season (Oct-Dec) for high volume vendors (ABC & XYZ). · AI agent can perform additional validations to detect duplicates. · AI agent + Human in loop (HIL) to improve controls mechanism. Example 2: Inaccurate Tax calculation in Invoice Validation process High-level steps: · Vendors submit the invoices. · Accounting team performs validation checks. · Tax checks are ignored without extensive reviews. · Exceptions are raised only when discrepancies are identified. · Finance team approves the payment. · Payment gets processed. Assumption: · Vendor systems are already automated and they are likely to submit accurate tax calculation on their invoices. Blind Spot: · Team missed to question tax accuracy for few vendors. · Team overlooked recurring small mismatches. · Team assumed automation done by vendor will lead to accurate invoices. How AI agent helped: · AI agent performs analysis on the invoices for the past 12 months. · AI agent detects patterns in vendors invoice tax calculation. · 1-2 % tax deviations were detected by agent for vendor X and A. · AI agent identified correlation in line item count & errors in shipping charges. · Implement new validation rule if line items are greater than 3. · AI agent + Human in loop (HIL) to verify GST calculation manually. Conclusion: Blind spots are exposed by AI agents because they analyze data without relying on human assumptions. They detect hidden patterns and helps the team to redesign processes based on evidence rand not by intuition. AI helps to automate work and supports in strengthening controls.

Agree that AI systems can generate artwork, write scripts and solve problems for us, but I don’t treat this as creativity. I believe that AI primarily learns from the data that we feed and it tries to remix. Let me give an example of how AI helped in solving few of the process problems in my organization and at the same time it tried to be more creative/re-mix things. Process: One of the process in my organization involves multiple teams working on financial reporting, ledger updates and investment accounting. These activities are performed on a daily and monthly basis. Below is the high-level process: · Pricing team sends pricing & valuation files to the accounting team. · Accounting team validates the details to ensure it is correct and booked on a timely manner. Considering the volume, complexity & the risk of errors, we thought of implementing an AI agent to support the team members to review, validate & automate the process. We considered below activities as part of our scope to implement Accounting Validation AI agent: 1) Extracting essential pricing and valuation information 2) Presenting extracted information to an accounting reviewer 3) Supporting decision-making (proceed with validation or put on hold) 4) Running accounting validation checks 5) Handling business-level outcomes (Validation OK or Exception) and 6) Sending notifications to stakeholders While implementing the AI agent, we noticed that the agent accurately & consistently extracted pricing/valuation information & presented it to the accounting reviewer. At the same time, the agent while supporting the decision making & handling outcomes (OK/Exception or mismatch) for Pricing calculations & invoices, it tried to be creative & remix things. We all know that AI depends on human created data & architecture and cannot experience emotion or cultural context and they do not create concepts outside their trained data. Examples of AI trying to be creative/re-mix things in the process: 1) The AI compared the historical accuracy, timelines, volatility & past delays for Pricing related information and forecasted that “Prices from vendor A this month are likely to have price miscalculations because the Pricing calculation template changed last month” There was no rule written by us that treat this case as miscalculation but it remixed data patterns and produced a new warning which was more of a creative act in analysing information. 2) Instead of flagging an error like “Mismatch found” for a variation of Price in the invoice (8% vs agreed 5%) for vendor B, it identified that same discrepancy happened in June due to a system migration and recommended to validate the master file of vendor B unnecessarily. This was not programmed but it generated the response based on re-mixing insights.