Q 766. Think of a specific challenge or opportunity in your domain where you believe an AI solution could make a meaningful impact.

Describe the problem clearly, and then explain which of the four AI solution-building approaches (conventional AI, fine-tuned LLM, training from scratch, or flow + prompt-based design) you would choose to solve it — and why that method fits best.

 

🏆 The best answer will be selected on the basis of:

• Clarity and relevance of the chosen problem

• Sound reasoning behind the selected AI approach

• Practicality and originality of the proposed solution

 

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.

Problem: Auditing faces significant challenges in gathering and analyzing relevant evidence. Traditional audit processes are time-consuming and labor-intensive, requiring auditors to sift through extensive financial data, contracts, invoices, emails, and other records. Manual identification of patterns, inconsistencies, or anomalies is tedious and prone to human error. Additionally, reliance on sampling methods can lead to missed issues, especially in large datasets. Timely, reliable, and actionable insights are crucial, and without proper tools, audits may fail to uncover potential risks or fraud.

 

Recommended AI Solution: Fine-Tuned LLM

 

• Natural Language Understanding: Fine-tuned LLMs excel at understanding and extracting key insights from unstructured text like contracts, financial statements, emails, and internal reports. They can automate the extraction of relevant clauses, highlight inconsistencies, and identify potential risks across large text corpora.
• Time and Cost Efficiency: Fine-tuning leverages the vast general knowledge of pre-trained models, adapting them to the specific needs of auditing. This process is more efficient and cost-effective than building a model from scratch.
• Handling Complexity: Audit data is complex and varied. Fine-tuned LLMs can analyze different document types, perform sentiment analysis on communications, detect inconsistencies, and identify irregularities in financial records—tasks challenging for conventional AI models based on hardcoded rules.
• Scalability: Fine-tuned LLMs can scale effectively as audits grow larger and more complex. They can be retrained or adapted to new data or audit requirements, providing flexibility to keep up with changing business environments.
• Interactivity: Auditors can interact with the fine-tuned model via natural language queries, asking for explanations of flagged transactions, summaries of contracts, or generating reports from raw data. This conversational ability streamlines the audit process and makes it easier to obtain relevant evidence on demand.

Conclusion: Fine-tuning an LLM is the most appropriate approach for solving the challenge of gathering

I genuinely do believe all front and back-office enquiry handling in the BPO industry can and will be handled by AI.

For the purpose of this question, I would like to focus on Customer enquiry handling. The problem statement will have to be long wait times; the enquiry process itself is time consuming and labour intensive. With agents dealing with high volume of repetitive tasks/enquiries the margin for error is always fluctuating depending on time of day, energy levels and variables that are not consistent which is inherent of a human being. Further to this, these elements then lead to poor satisfaction levels due to the above-mentioned challenges when using only humans for these types of enquiries.

The motivation would be a saving in operational costs and less waiting and handling times for customers.

Also, agents can be redeployed to more meaningful tasks and the AI can deal with the mundane repetitive tasks that often contribute to lower agent morale, decreased morale and eventually burnout.

With a reduced error rate and an ever spunky, energetic AI agent, the Customer Satisfaction performance can only be impacted positively.

 

The appropriate approach would be to employ the Fine Tuned LLM with flow + prompt-based design and here’s why:

 

This approach fits best because it:

1.      Will utilize existing LLM capabilities and adapt them to customer enquiry handling.

2.      As noted in the BPO industry we deal with high volume transactional, repetitive enquires, this approach provides the flexibility and scalability to handle these volumes.

3.      Because this approach will also allow for personalized responses and is quite efficient, it will definitely lead to improved Customer Satisfaction.

 

 By employing the fine-tuning an LLM with a flow + prompt-based design approach, BPO's can automate customer interactions/enquiry handling, reduce operational costs, and ultimately improve customer satisfaction.

 

AI Impact: Improving Efficiency on content moderation by increasing bot moderation

 

Problem statement:

Majority of the User Generate Content on social networking sites includes harmful, hateful and wrong information. Manual moderation is time consuming and also is prone to error due to the gray areas in policy impositions. The part of content moderation that I would want to automate using AI is to quickly & efficiently moderate the user generated content. 

 

Proposed Solution:

Develop an AI that will review the user submitted content and the evidences available within the system and in open sources, then take a decision of if the content can be allowed or not

 

AI Solution building approach that I would use is: Fine-tuned LLM

 

Rationale to use the Fine Tuned LLM:

• Understanding of the context: Fine Tuned LLMs can better understand the language nittygritties and identify the forms of content in line with policy
• Scalability: Due to large data handling capacity, LLMs are best suited in my industry
• Adaptability: LLMs can easily adapt to the new/changes in policies
• Efficiency: Fine tuning of an existing model to leverage the current knowledge base is easier with LLM

 

Vidhya R

Use Case: Timesheet Compliance Monitoring in Audit Firms

 

Problem Statement:

Audit firms often struggle with timely and accurate submission of employee timesheets. This delays internal reviews, billing cycles, and impacts project profitability analysis. Manually tracking non-submissions across hundreds of employees, identifying patterns, and sending reminders is time-consuming and error-prone.

 

Chosen AI Approach:

Flow + Prompt-Based Design using an LLM

 

Why this approach fits best:

 

• Clarity & Flexibility: Prompt-based design allows you to build a low-code AI assistant that reads structured data (e.g., timesheet records in Excel or Google Sheets), identifies missing entries, and generates natural language summaries for managers.
• Rapid Deployment: No need to retrain or fine-tune models. With smart prompt engineering and basic data structuring, this solution can be operational within days.
• Scalability: You can customize prompts for different departments or partner groups without deep technical skills.
• Practical Example Prompt:
  “Review this dataset of employee timesheets. Identify employees who haven’t submitted entries for more than 3 consecutive days in a given month, summarize their names, dates missed, and location, and draft an email reminder accordingly.”

 

 

Originality:

Unlike rigid rule-based systems or expensive retraining, this approach leverages a general-purpose LLM (like GPT-4) to act as a compliance assistant — adaptive to language nuances, capable of exception handling (e.g., medical leave cases), and scalable across teams.

Scenario: Predicting a Bioequivalence Study for Generic Drug Products

In formulation development, a successful bioequivalence (BE) study is critical for a generic drug product. A successful BE study means that the rate and extent of absorption (pharmacokinetics, PK) are statistically equivalent to the reference (brand/innovator) product. Running a human bioequivalence (BE) study can be quite expensive and time-consuming, but there's also a significant risk of failure. This can lead to project delays and affect the overall business case.

Key Challenges:

1. 1. Pharmacokinetic variability: The success of a BE study depends on various factors, including the design of the formulation, dissolution data, the drug's in vivo effects, and patient-related factors such as pre-existing conditions and geographical differences.
2. Passing criteria: Regulatory agencies like FDA/Health Canada/EMA require strict statistical passing criteria equivalence (90% CI within 80-125% for AUC & Cmax).

An AI that predicts bioequivalence studies could help to customise drug products and reduce the failure probability.

A Conventional AI combined with a fine-tuned pharma domain-specific LLM is the best option

1. Conventional AI (First step)

Prediction of Bioequivalence parameters requires structured data like formulation properties, dissolution profiles, drug substance data, and/or preclinical PK data. Conventional AI approach includes physiological pharmacokinetic (PBPK) modelling, In Vitro-In Vivo correlation, and statistical equivalence creator. PBPK modelling is the basis for bioequivalence prediction. While conventional AI can deliver precise and consistent results, it also offers regulatory clarity.

2. Fine-Tuned LLM (Second step)

A specialized pharma domain LLM can assist in generating reports that meet regulatory standards, extracting relevant study information from existing literature, and outlining a strategy for hypothetical bioequivalence testing. This LLM includes fine-tuning on regulatory guidelines, PK information, and previous pass/fail BE study reports.

Why Not Other Approaches?

• Only Conventional AI: Good for pharmacokinetic data prediction, but lacks explainability and is limited where part of the data is missing.
• Training from Scratch: Bioequivalence prediction does not need to be developed from scratch, as existing PK knowledge can be fine-tuned.
• Pure Prompt-Based LLM: Bioequivalence prediction requires precise modelling, so pure prompt-based LLM is not efficient.

This approach balances scientific knowledge with AI flexibility (LLM augmentation), making it both innovative, accurate and successful.

 

When thinking about ways that technology could improve mental healthcare in South Africa, one possible scenario is the use of therapy robots. Which will improve mental health.

The Problem: The availability and affordability of mental health services are severely limited for many South Africans, especially those living in rural areas. Due to which it impact severely. 
 
Some of the problems with traditional therapy include stigma, long wait times, and an inadequate number of trained therapists. Understanding different cultural contexts and individual needs is crucial for chatbots to be effective in providing initial support, emotional assessments, and useful tools.
My recommendation for fixing this issue is to use a broad language model approach that has been fine-tuned. 

Because of this: By refining a language model, we can better account for South Africa's rich linguistic and cultural diversity. By training on local information, the chatbot can give answers that make sense and are relevant to the situation at hand.

 Emotional Understanding: Through fine-tuning, the model gains a greater understanding of the nuances of emotional language.  This makes it easy to identify indicators of pain and respond appropriately.
Rather than building a model from scratch, fine-tuning uses what is already known. This saves time and money.  The goal of this method is to save time and resources without lowering the quality. 
 More and more people will be able to get help with their mental health because this approach is easy to adapt to other languages and areas.
If we look at alternate  Rule-based systems cannot handle the intricacies of human emotions and cultural contexts.  Training that starts from beginning requires a large amount of data and computer power, so this is not the greatest method to use it.
 Flow and prompt-based design can be effective for structured conversations, but they may be unable to give the empathy and insight required for mental health care.
 Improving an LLM can contribute to the development of a mental health support system that is culturally sensitive, effective, and fulfills the needs of all South Africans.

The performance of rotary tablet press for pharma and Food application, is trigged by the density, moisture and the flowability of the product to be pressed.

 

To define best parameter for high performance and quality, it’s important de follow:

-          The critical Raw Material for binding

-           The temperature and the moisture fort he activation of the binder

In one part.

In the other part record the machine parameters :

-          The pressing force applied to the product to shape it

-          The standard deviation of the pressing force

-          The weight

-          The height of the final product critical for the wrapping

 

And at the end corolate machine  parameter and product parmeter to define the best operational window for highest performance and quality.

 

 

I think, training from cratch AI approach could be helpful to build a model driving a control loop for moisture adjustment

Problem Description: 

 

In a healthcare claims negotiation & arbitration process, when Out-of-network providers feel that they have been underpaid, they raise negotiation requests. In negotiation stage, claims adjustment team reviews the additional records shared by provider and based on that, they pay extra or stick to the initial payment made. They also explain why the payment amount was decided. If the provider is not happy after that, they raise an arbitration request. The arbitration team selects external arbitrators (from a wide-range of available federal govt-certified arbitrators). The arbitrators are needed to be paid a huge amount (between $500-$1500) by the arbitration team. Arbitrators' decision on the payment amount often goes in favor of the provider.

 

Goal:

 

The goal is to create an AI model, which will be able to suggest a claim payment amount along with an explanation behind selecting the amount in the negotiation stage itself, so that the process can save arbitration fee costs. Historical data from claims adjustment (non-negotiated claims, negotiated claims and arbitrated claims) will be used to build the model.

 

Recommended AI Solution Building Approach:

 

Fine-tuning an existing LLM will be the best fit.

 

Why Fine-Tuning over other available options?

 

Because, this can help us leverage the advanced features of MED-BERT, which is a contextualized embedding model designed for healthcare applications and can be fine-tuned to meet the requirement of the process

 

Below are few key advantages of Fine-Tuning:

• Efficiency: Less data and computational resources requirement, given the complexity and volume of historical claims data.
• Performance: When fine-tuned for specific tasks, LLMs can achieve high accuracy and adaptability.
• Domain Adaptability: The model can be allowed to specialize in the domain, leveraging pre-existing knowledge, while adapting to specific negotiation and arbitration contexts.
• Speed: Faster to deploy than training a new model from the scratch, even with iterative feedback-based improvements.

Limitations of other models:

 

Conventional AI-models:

• May struggle with the complexity and variability of healthcare claims data.
• Often lack flexibility and adaptability needed for nuanced decision-making.
• Lower accuracy and capability of generating detailed explanations.

Training a New AI model:

• Requires extensive computational resources, large datasets.
• Requires significantly high time and cost investment.
• High risk of overfitting and poor performance.

Flow and Prompt-based Design:

• May not achieve the required level of accuracy and depth in understanding complex claims data.

 

Challenge:
Demand Forecasting for New Product Launch in the FMCG Industry

 

Problem Description

In the FMCG industry, forecasting demand for new product launches can be quite challenging. Unlike established products, there may not be historical data for the proposed new product. Accurate forecasting is a must and if missed, may lead to excess inventory and inevitably cause wastage

Traditional demand forecasting methods, which rely on time dependent data, struggle with the "cold start" problem. Because there is no history of past sales, machine learning models which require large, labelled datasets may not meet up to standard requirements and this may need to market loss.

 

Proposed AI Approach:

Flow + Prompt-based Design (Leveraging Generalist LLMs)

 

Why This Approach Works Best

1.       Handling Unstructured Contextual Data: A flow + prompt-based LLM solution can incorporate different unstructured information e.g., product reviews from similar items, similar products from competitors, feedback from retailers & distributors. A general purpose LLM can use this information as a basis to provide solution.

2.       Guiding the AI using step by step prompts:
The AI is giving step by step prompts in a structured format to enable analyse the problem. Using trends, it generates demand estimates which makes it’s results reliable and very easy to adjust whenever there is an inflow of new information.

3.       Easy to Set Up and Use Quickly: No training of any new model is required for this, therefore making it easy to build faster. It can fit into already existing tools that the business already uses and allow for quick improvements, even without high technical requirement.

4.       Originality: This gives room for a fresh and unique approach, because most current forecasting models in FMCG still rely on structured, supervised learning. The use of LLMs in interpreting marketing narratives, perceptions of the consumer and the demand history gives the business a new and unique way for demand prediction.

A Facilities Management (FM) company into B2C business caters to requests from customers related to facility maintenance including Plumbing, Carpentry, Civil, HVAC, MEP, Housekeeping, Pest Control and Façade cleaning works. The company has a call center with human agents to attend customer calls / messages, understand the nature of the request, evaluate priority, check technicians’ availability, initiate advance payment and schedules for site visit. The workflow continues until the customer request is resolved and customer satisfaction survey is triggered and logged.

Below mentioned is a high-level process flow.

 

The company has the following challenges:

1.       Low customer satisfaction ratings due to

a.        High call waiting time (time taken by human agents to respond to calls / messages)

b.       High response time (time taken to visit site and address the problem)

c.        Communication (language barriers)

2.       Low human agents’ utilization (volume of requests has seasonality, agents are idle during lean period)

3.       Technicians’ utilization (due to improper scheduling and optimization)

Obviously, these challenges impact the company’s financials. The company decided to make improvements in the process using suitable AI solutions thereby addressing the afore-mentioned challenges. Important aspects to consider choosing a suitable solution include the cost of implementation, speed of implementation and scalability.

 

These AI solution options are evaluated:

1.       Conventional AI

2.       Fine-tuned LLM

3.       Training a new AI model from scratch

4.       Flow & Prompt based design

 

Choosing the AI Solution:

I would decide to use “Fine-tuned LLM” as it best fits the objectives and requirements. The reasons why this option is considered are because:

1.       It reduces implementation costs as only fine tuning is required to do specific tasks.

2.       Deployment will be faster as only fine tuning is required.

3.       Fine-tuning requires high quality data which the company has collected and stored in the past years.

4.       The pre-trained LLM already brings in vast capabilities that the company could leverage.

5.       Specialized resources are not required for deployment

6.       Can handle domain specific process very well

 

Why other options were not considered:

Conventional AI: Suitable for simple tasks but has limited scalability options.

Training a new model from scratch: Tedious and time consuming requiring specialized resources.

Flow & Prompt based design: LLM capability limitations. Could provide inconsistent outputs to ambiguous queries. Best for scenarios with data limitations. However, this could have been another option considered.

 

 

 

AI can solve several problems in the BPO recruitment process, including reducing bias, automating tasks, and improving candidate matching, leading to more efficient and accurate hiring. AI can streamline CV/Resume screening, help predict talent needs, personalize job matches, and even manage communication with candidates through chatbots. 

Reducing Bias and Improving Fairness:

·         AI can help identify and eliminate biases in the recruitment process by using objective data and predetermined criteria, ensuring a more fair and inclusive hiring process. 

·         AI-powered tools can analyze applicant data in a more objective way, reducing the risk of unconscious bias affecting hiring decisions.

·         AI can also help with diversity and inclusion initiatives by identifying and addressing potential biases in recruitment practices. 

 

Streamlining and Automating Recruitment Tasks:

·         Automated Resume Screening:

AI can efficiently sort through large volumes of resumes, identifying the most relevant candidates based on specific job requirements. 

·         Scheduling and Communication:

AI-powered tools can automate interview scheduling and communication, freeing up recruiters' time for other tasks. 

·         Talent Pool Development:

AI can help build and manage talent pools, identifying and engaging with potential candidates proactively. 

·         Cost Reduction:

By automating tasks and improving efficiency, AI can help reduce the overall cost of recruitment. 

 

Improving Candidate Matching and Experience:

·         Personalized Job Matching:

AI can match candidates with job opportunities based on their skills, experience, and preferences, improving the accuracy of hiring decisions. 

·         Predictive Hiring:

AI can analyze past hiring data to identify patterns and predict which candidates are most likely to succeed in a role. 

·         Improved Candidate Experience:

AI can personalize communication and provide timely updates to candidates, enhancing their experience throughout the recruitment process. 

 

Addressing Skills Shortages and Talent Acquisition Challenges:

·         Identifying and Attracting Talent:

AI can help identify and attract candidates with in-demand skills, addressing skills shortages in the marketplace. 

·         Predictive Analytics:

AI can analyze market trends and predict future talent needs, allowing organizations to proactively plan for recruitment. 

·         Enhancing BPO Branding:

By streamlining the recruitment process and improving candidate experience, AI can help enhance BPO branding and attract top talent

 

By addressing these problems, AI can significantly improve the efficiency, fairness, and effectiveness of the recruitment process, leading to better hiring decisions and a stronger workforce. 

 

  Problem Statement : We build software solutions for our clients in the HealthCare sector. There are multiple solutions that we have built and delivered to our clients. Our clients use our solutions, raise issues when they feel that the software/solution is not working as per the expectation or as per the needs or requirements. As we have multiple solutions, we generally have a list of client raised issues in our backlog.  These issues can be categorized into functional, performance, usability, security, compatibility and so on. We have a group of individuals who goes through the issue, categorize these issues and assign them to respective persons or teams (speaking of support teams only). The issues can sometimes be complex and sometimes the explanation of the issue can be hypothetical or vague. 

 

Solution : As the organization has vast amount domain specific data, we can use a Fine Tuning LLM model to build our AI solution. We can use a pre-trained LLM to finetune using domain specific dataset/knowledge base and adapt. We can use methods such as PEFT, Prompt tuning or RHLF , to refine and eliminate over-fitting or bias. This helps the AI solution to categorize the issues without any bias derived from the historical data/knowledge base/dataset . As we fine tune the pre-trained LLM we improve the accuracy, efficiency, flexibility, productivity. This way the AI solution can reduce the workload on the team of individuals who then can be used to as resources for some other purpose.

 

 

It was a real tough one to decide the winner. All the answers are excellent and to a great extent cover the various aspects of the question. Hence, it is recommended that one goes through all the answers.

 

The most well rounded answer has been selected as the winner - Sourav Biswas. Well done!