Q 807.  In Business Excellence, processes evolve constantly through continuous improvement initiatives, customer feedback, or regulatory changes. But many AI solutions risk becoming outdated if they aren’t designed to adapt.

Think of a process in your domain that changes frequently.
How would you design an AI agent to stay aligned with those changes — without needing to be rebuilt from scratch each time?

The best answer will be selected on the basis of:

• Relevance of the chosen process

• Practicality of the adaptation approach

• Insight into balancing stability with flexibility
   

Note for website visitors -

• This platform hosts two weekly questions, one on Monday and the other on Thursday. 
• All previous questions can be found here: https://www.benchmarksixsigma.com/forum/lean-six-sigma-business-excellence-questions/. 
• To participate in the current question, please visit the forum homepage at https://www.benchmarksixsigma.com/forum/.
• The question will be open until Monday or Thursday at 5 PM Indian Standard Time, depending on the launch day. 
• Responses will not be visible until they are reviewed, and only non-plagiarised answers with less than 5-10% plagiarism will be considered for winner selection. 
• If you are unsure about plagiarism, please check your answer using a plagiarism checker tool such as https://smallseotools.com/plagiarism-checker/ before submitting. 
• All correct answers shall be published, and the top-rated answer will be displayed first. The author will receive an honourable mention in our Business Excellence dictionary at https://www.benchmarksixsigma.com/forum/business-excellence-dictionary-glossary/ along with the related term.

• Some people seem to be using AI platforms to find forum answers. This is a risky approach as AI responses are error-prone because our questions are application-oriented (they are never straightforward). Have a look at this funny example - https://www.benchmarksixsigma.com/forum/topic/39458-using-ai-to-respond-to-forum-questions/

• We also use an AI content detector at https://quillbot.com/ai-content-detector. Only answers with less than 45-50% AI-generated content will be considered for winner selection.

Applying AI systems in practical business environments plays a vital role in evolving process, but AI systems deployed with static model might fall behind when processes constantly evolves with new regulations, customer requirements, improvements. So, AI system should be designed as modular, configurable, adaptive and incremental learning so that the model can be aligned with changing process without building AI agent from scratch. These approaches help AI to adapt changing process which is explained below:

• Continuous Learning Loop - Dynamic learning pipelines uses custom logic layer with automated retraining model caused by data drift or process changes
• Modular Architecture – Build the AI system with distinct modular components and new rules can be embedded to specific modules in case of any process change instead of changing entire AI system
• Self-Correction Agents – Agentic AI uses self correction and reasoning agents that queries updated KB instead of full retraining of model for any process change
• User Feedback Mechanism – In case of complex cases, AI uses human in the loop mechanism and learns from user corrections/feedback and use these inputs to improve AI agents

Adaptive AI agents can be designed for process changes in finance and accounting domain and below few process change examples have been provided:

• New regulatory updates – Adaptive AI updates new tax rate introduced for specific services and process automatically in tax calculation mode using updated tax rate without changing rest of the system
• Change in Travel & Expense policy – Adaptive AI applies the new allowance mentioned in policy to all future expense claims avoiding redevelopment since the rules are modular and easy to edit
• Change in workflow for invoice approval – Approval for high valued invoice amount > $10K require 2 layer approval from managers. Adaptive AI updates new approval process and auto routes high valued invoices to 2 layer approvals without retraining system
• Enhanced fraud detection model – Adaptive AI retrain fraud detection model with examples of new pattern of fraudulent activity and then flag suspicious transactions matching with new fraud detection criteria keeping the rest of system unchanged
• Enhance customer ticketing system – Adaptive AI retrains ticket categorization when new complaint categories are identified for new services or any new type of escalations are raised and then routes the ticket to respective functions

The chosen process is Investor support services in Primary markets.

During IPOs the transaction volumes are at times unpredictable due to various factors  especially market conditions and global financial movements. Furthermore, the sector being highly regulated and transactions to be executed in timely manner calls for fine balance between robust processing and adaptability to changing Regulations .

An AI agent handling Investor queries must flexible in design to adapt to these changes to remain effective without requiring a complete rebuild.

Designing an Adaptive AI Agent

The primary requirement of the AI agent would be to always be aligned with the evolving process. For this propose a modular, learning-based architecture with continuous feedback loops would be most suitable.

- Modular Architecture

Breaking down the entire process into units and developing solutions around the activities would ensure that process changes are incorporated with minimal efforts without impacting the overall solution. 

- Continuous monitoring and feedback

Since the AI solution is handling critical function, the responses and actions needs to be monitored and error identified and corrective actions taken immediately to ensure minimal or no complaints or escalations are triggered.  Also as a best practices it would prudent to mention that the transactions is/was processed by a Bot and in case if the recipient notices any error to kindly highlight the same immediately. Therefore, seeking timely feedback ensures that the solution meets its purpose of reducing human efforts rather that spending more time in making the solution work effectively.

- Change management

With changes in Regulatory environment and Industry standards, these have to be incorporated into the process and AI solution to ensure relevance and usefulness. This can be achieved through engagement with Regulatory bodies and Industry experts to foresee changes and anticipate the timelines of these changes having an impact on the process. 

- Stability of the solution

Though the solution would be able to handle volumes however, like any technology intervention needs to have a fall-back mechanism to ensure the process is not stalled due to any reasons. Since being a modular Architecture the component failing can be isolated and work around can be deployed to ensure continuity. 

A Human-in-the-loop mechanism would prove to be an additional monitoring mechanism along with ensuring that the AI model is trained and performing the tasks as planned. This can also be utilised as a fall-back mechanism in certain scenarios.

 

Overall the solution could prove to be an effective model as it leverages the existing applications and technologies due to its modular architecture and provides flexibility to incorporate changes in structured manner with least impact to production output and thereby being cost effective.

In the banking domain, KYC (Know Your Customer) checks are one of the processes that change very frequently. Reserve banks keep updating their guidelines, thresholds, and documentation requirements, and if an AI solution is built in a rigid way, it can quickly become outdated.

To handle this, I would recommend designing the AI agent that focuses on three things as mentioned below:

1. Modular design for regulatory requirements – AI model would handle pattern recognition (like spotting unusual customer behavior), while all regulatory rules, thresholds, and Reserve Bank and other regulatory requirements sit in a separate configuration layer. With this, compliance teams will be able to update rules or any new data files without touching the AI code. I would also ensure that this bot pulls the latest regulatory requirements (like sanctions, circulars etc.) through API or secure feeds.
2. Integration of Customer feedback– Would connect the AI agent with customer service platforms (chat, email, survey tools) and enable it to identify the cases where customer has been consistently flagging pain point or any other issues.
3. Integration of Employees feedback integration – Would also create a customized forum (e.g. tagging cases in applications) for Relationship managers, operations staff, and compliance officers to provide their feedback, as in many cases they are able to spot issues before they become escalations. Embedding a simple feedback mechanism
4. Real time CI dashboard – Would create a real time governance dashboard for risk, compliance, and business excellence teams so that they can monitor what the AI flags, adjust rules, and test small changes before scaling.
5. Learning cycle – Every KYC alert should be tracked: Was it valid? Was it ignored? Did it prevent a risk? That evaluation becomes the input for retraining the model and reducing noise over time.

With consideration of the above points, I would create an AI agent that works like a flexible engine that plugs into live regulatory data and evolves with it. This would ensure that our AI bot is not only well equipped with regulatory updates but also learns from customer, banker, and other employees’ feedback.

Let’s understand how one would design an AI agent to stay aligned with regulatory or business requirement changes, be adaptable — without needing to be rebuilt from scratch each time.

Example - Frequent changes in GST call for changes in overall direct and indirect tax structure. The GST slabs and calculations need to be reconfigured and aligned to match to stay complaint. Many times, in built GST calculators say in Tally or ERP are static and would need to be scrapped or would require change in configuration. This would call for manual efforts, off the system calculation, adding to delays and errors, till the time the tool is rebuilt. This may have an impact on cost and possible tax liabilities.

AI agent design would need adaptive compliance engine to address the listed core challenges. Instead of hard coding rules directly into the system, AI agent with “Dynamic Rules Layer” that learns, validates, and applies updates automatically.

 

Key Design Principles

Regulation aware Natural Language Processing (NLP) engine - Use NLP to continuously scan official government notifications, circulars and tax websites.

Automatically extract changes in tax slabs, exemptions, and compliance deadlines.

Example – AI identifies “18% slab changed to 12% for foreign exchange services effective from April1, 2026”.

Modular Rule Database - Rules to be stored in a configurable knowledge graph or rule library, instead of hardcoding. Each GST rule is represented as a modular entity (rate, category, effective date, applicability). The AI agent will update the database rather than rewriting the software. 

Automated Impact Simulation - AI runs “what-if” scenarios on existing transactions to simulate the effect of new rules. Example – AI identifies and aligns to “If GST rate changes to 12% for IT services, projected tax liability reduces by 8%.

Human in the Loop Validation - Compliance Manager (Tax team) get AI generated draft updates and approval requests before deployment. Thus, ensuring trust, accountability, and avoids blind automation risks. 

Plug and Play API layer - The AI agent exposes updated rules via APIs. ERP/Tally systems pull updated rules dynamically instead of requiring rebuilds. API integration will give the required flexibility for scalability.

Illustrative prompt for AI agent can look like – “Act as a compliance update assistant. Your role is to 1) monitor government tax portals for GST notifications, 2) extract and summarize changes in slabs, exemptions, or filing rules, 3) update the modular tax rule database with effective dates, 4)simulate impact on past and future transactions 5) provide clear recommendations in a supportive, non-technical manner to tax  teams, 6) expose updated rules via APIs for ERP/Tally integration. Always flag ambiguities to Compliance Manager for approval.”

 

The approach would thus position AI gent as the living “Tax Brain” and benefit business by optimizing costs, stay compliant and adaptable to meet any changes in regulation irrespective of the tax category.

With the agentic AI in all domains in our day-to-day activities, AI is also being leveraged by SAP to enhance the business processes related to BTP, Success-factors, S/4 HANA, Clean Core technologies, etc. In SAP there are many fields amongst which payroll is the most sensitive and dynamic area which requires special attention and continuous upgrade due to yearly regulations, localizations and changing policies based on Compensation benchmarks.  AI is already being incorporated in Payroll processes like advanced reporting, data processing and integration with payroll control center.

Below are few pointers to train the AI model to avoid the rework related to building an AI agent with every change coming into the payroll process.

Integration with SAP - AI should be integrated with SAP Support enabling automatic updating of Knowledge Base with new payroll updates /releases/notes/patches/ compliance changes in SAP and enable users to keep themselves up to date with latest regulatory changes.

Users Feedback --Continuous interaction with the users of the Agent to check the relevance of the answers, identifying missing information pieces, and updating the knowledge base based on the feedback.

Interaction with Business users -Continuous interaction with the Business users to explore the changes in payroll related business processes, issues and refinement areas and update the AI model to facilitate SAP implementations.

Prediction-Train the models to predict the near future trends in SAP payroll related to employee overtime, compensation changes, unauthorized absences, based on current and previous data and patterns and inform users to take proactive measures to ensure compliance and avoid potential risks which will optimize the payroll processes and help users to take data driven decisions.

Objective: I want to build and maintain an online Mutual Fund tracking system for my client. Since 2023 there has been many changes and to keep us with the changes. I have found the following areas which are critical and every changing. Thanks to AI many things can be managed. Please find some of the area:

 

1. Real-Time Fund Performance Tracking

 

Use of AI-driven to track NAV & volatility monitoring:

ML models analyse historical NAV trends, macroeconomic indicators, and sentiment from news/social media to predict short-term risks.

 

Build Dynamic dashboards & alerts:

AI can trigger alerts for sudden NAV drops, liquidity stress, or regulatory breaches.

 

2. Real time Personalized Investment Recommendations

 

Hyper-personalization beyond KYC:

 

AI Can uses behavioural data, transaction history, and life goals to recommend funds (e.g., SIP vs lump sum, ESG vs sectoral).

• Predictive portfolio optimization:
  • Using AI one can simulates market scenarios to suggest rebalancing strategies.
• Compliance-aware suggestions:
  • One can use AI agents to ensure recommendations adhere to SEBI suitability norms and avoid mis-selling by in-corpora ting the change very fast. 

3. Automated Reporting & Communication

• Natural Language Generation (NLG):
  • AI converts complex fund data into simple, jargon-free investor reports. Which can be seen in real time.
• Chatbots & voice assistants:
  • AI-powered bots handle investor queries real time in multiple languages, improving accessibility.

 

4. Customised Fund Screener & Comparison Tools

• Customised AI-powered screeners:
  • Filter thousands of funds by custom metrics (e.g., ESG score, volatility, expense ratio, SEBI risk-o-meter). one can look at their portfolio break up into small, medium and large including funds invested in international stocks. 
• Sentiment & performance scoring:
  • Through NLP models can analyse analyst reports, news, and social chatter to rank funds dynamically.

 

5. Cost Optimization for my Fund Houses

• I can do Real time Automated KYC & onboarding:
  • AI reduces onboarding time by 40% and lowers operational costs. This can be done 24/7 and any changes can be implemented real time.
• Process automation:
  • AI agents handle repetitive tasks like NAV validation, reconciliation, and investor communication.

In a contact center, things rarely go exactly as planned. You might have a forecast that looks solid, but during the day it can change fast — a few people don’t show up, call times go higher than expected, or the client suddenly says ,for example - “focus on collections more than sales.”

If I were building an AI agent to handle this, I wouldn’t want to rebuild it every time something shifts. I’d do three simple things:

First, keep the backbone steady. The AI’s main job — spotting when SLAs are at risk, comparing forecast vs. actual — doesn’t need to change. What does change (like priorities or staffing numbers) should just be stored as data that’s easy to update.

Second, plug it into live inputs. The AI should read from WFM or call systems so it knows today’s reality, not last week’s plan. And it has to be smart about limits — you can’t just move anyone into collections if they’re not trained for it. We can move only cross-skilled agents. 

Third, keep humans in the loop. The AI can suggest overtime, break shifts, or call-backs, but supervisors see the impact before hitting “go.” Over time, the system learns what works without being retrained from scratch.

That balance — stable logic but flexible levers — is what keeps the AI useful even when the floor keeps changing.

Congratulations to Monica Salunkhe, whose detailed design of a GST “Tax Brain” agent emerged as the winning entry. By combining NLP-driven regulation scanning, a modular rules database, automated impact simulation, and human-in-loop validation, her response demonstrated the strongest balance of adaptability and practicality. Close runner-ups include Gaurav Saxena, for his clear and flexible KYC compliance engine with regulatory feeds and stakeholder feedback loops,

B. Ravi Shankar and Sarveshvar, who offered practical modular and real-time adaptive designs for finance and contact center processes, and Tushar Ghosh, for his comprehensive mutual fund tracking system that integrates compliance, personalization, and reporting.
 

We also acknowledge the contributions of Gopu Nair and Deepti Kumar, who provided relevant and publishable designs for IPO investor services and SAP payroll adaptation.