Q 757.  Think of a process in your domain where an AI agent could handle 90% of the work but might occasionally face a situation where it must pause and escalate to a human. Describe one such scenario and clearly mention what criteria or signals the AI should use to decide when to escalate.

The best answer will be selected on the basis of the following: 
- Relevance & Practicality of the scenario  
- Clarity & Structure of the explanation  
- Thoughtfulness & Originality in defining escalation criteria

 

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.

Online loan application.  

1- Give your full name and your account number and the name of the account manager  

2- Define the type of account, savings or current  

3- Choose the desired loan category:  
Habitat Vehicle Social School  
4- Define eligibility based on quality, analysis of entries and exits over the last three months 

5- Transfer the request, get acceptance by the manager and summon the user to come, sign at the agency (handed over to the human)  
6- After confirmation of the loan, send an email to the applicant to notify them of the availability of the loan

23 hours ago, Vishwadeep Khatri said:

Q 757.  Think of a process in your BPO domain where an AI agent could handle 90% of the work but might occasionally face a situation where it must pause and escalate to a human. Describe one such scenario and clearly mention what criteria or signals the AI should use to decide when to escalate.

The best answer will be selected on the basis of the following: 
- Relevance & Practicality of the scenario  
- Clarity & Structure of the explanation  
- Thoughtfulness & Originality in defining escalation criteria

 

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.

In a telecom BPO, an AI agent can handle most customer billing queries like due dates, payment confirmations, and plan details. However, it should escalate to a human when complex issues arise, such as overcharging disputes with uploaded proof, multiple past bill problems, or when the customer shows frustration. Escalation signals include document uploads, repeated failed resolution attempts, and negative sentiment detected in the customer's language.

In many of our accounts AI is integrated with payment systems like processing payments, refunds, creating invoices, vendor creation etc. To take an example for refund, if AI detects any exception (like approval required for higher refund amount, any ambiguity or exceptional scenario) then trigger goes to the supervisor for approvals.  Since the AI handles rule based workload, it also ensures fast service while maintaining consistency and customer satisfaction.

Save the Driver or Save the Pedestrian -Scenarios which are Critical/Security related where the decision can be ambiguous. For ex - If a man suddenly comes in front of driverless driven car, who should AI save the passenger in car or the pedestrian. In similar situation replaces the man with a 5 yr old child. WHAT should AI do in both situations?

Situation: Utilizing Behavioral Analytics to Track Agent Health with AI

Burnout presents a challenge within high-stress BPO environments. An AI system is adept at analyzing 90% of behavioral indicators, such as tone, speed, and sentiment in conversations, login and logout activities, break lengths, and even abrupt declines in productivity, to proactively identify potential wellness issues.

When should artificial intelligence escalate to a human?

• A significant deviation from standard work behavior, such as logging in late unexpectedly after consistently demonstrating punctuality,
• A series of customer calls exhibited a negative or emotionally charged tone.
• Limited awareness of wellness prompts, such as disregarding recommendations to "take a five-minute breather."
• Consistently neglecting breaks or having excessively long durations of post-call tasks.

These subtle signals indicate more profound issues such as disengagement, stress, or burnout that a manager or human resources partner should address.

While AI is capable of recognizing patterns, only humans can facilitate conversations that are empathetic and meaningful.

 

In data entry processes, an AI agent can efficiently manage up to 90% of the tasks. However, in situations where the data is unclear, the AI should pause and escalate the issue to a human. The criteria for escalation could include a low confidence score in OCR output, illegible or ambiguous handwriting, missing required fields, data format errors, inconsistencies with reference data, or multiple failed attempts to correct the data. These criteria enable the AI to intelligently determine when to seek human intervention, ensuring accuracy while maintaining high levels of automation and efficiency in the data entry process.

An AI agent is deployed in the audit industry to streamline the review of financial transactions, handling about 90% of routine tasks. Two key examples illustrate how the system operates:

• Anomaly Detection:

The AI continuously monitors transaction patterns. For instance, if the system identifies a transaction that significantly deviates from the norm—such as a $10,000 entry when typical transactions are around $500, or repeated high-value transactions from unfamiliar vendors—it flags these anomalies for human auditor review.

• Documentation Gaps:

The AI verifies that every transaction is supported by the required documentation (invoices, contracts, purchase orders). If a high-value transaction is found to be missing one of these critical documents, it escalates the case to a human auditor to ensure compliance and mitigate risk.

 

This approach leverages AI to efficiently process standard transactions while ensuring that potential issues or risks are thoroughly examined by experienced professionals.

The Invoice Creation Process can largely be handled by an AI agent; however, it can also escalate if the criteria are not met, as agreed in the contracts.

 

The whole process requires

1. An AI agent to collect the monthly input. A form can be designed to collect the input data from the operations team, such as transactions processed, number of processors, number of working days, etc.
2. Another AI agent will be created to fetch information from the Knowledge base, primarily the contractual requirements, CRs, amendments, pricing, etc.
3. The first AI agent will collect the input and call the second AI agent to validate the input with the information stored in the Knowledge Base and take decision if Invoice to be created or an escalation to be made. 
4. The invoice will be created if the input matches all the contractual requirements.
5. The escalation will be made to Process Owner if the AI agent detects a mismatch of any of the criteria. For e.g., COLA mismatch or FTE mismatch.

 

Bringing AI into the Invoice Creation process will help avoid calculation errors, leakages, and compliance issues.

 

I can think of following scenarios where AI may need to escalate to a human. 

1. Incomplete Knowledge Base: When the query and resolution are not present in the knowledge base, AI may not be able to address the query effectively.
2. Complex Queries: If the query is unclear and AI cannot route it to the appropriate problem category, human intervention may be required.
3. Security Concerns: Queries originating from suspicious or stolen identities (e.g., flagged in the database for fraud or bad debt) may necessitate escalation for security reasons.
4. Sensitive Customer Data: If handling the query involves some sort of sensitive customer data, it may be required to escalate to a human to ensure proper management and compliance.

I can describe an example that falls under the third category - Security Concerns. In the telecom industry, one common type of fraud is Subscription Fraud. This occurs when fraudsters use stolen identities for KYC verification, sign up for services, and then fail to pay for them. Often, these fraudsters approach call centers or online sales channels, which are front-ended by AI chatbots, to request new or additional services. To combat this, additional AI/ML models are integrated into the AI chatbot workflows to screen and weed out suspicious calls and requests. If the model flags a query as a potential fraud case, the AI chatbot escalates it to human operators for further scrutiny and verification.

 

 

I think employee performance evaluation is one process where AI can track the employee performance based on various parameters and assigned targets. However, it might not be able to take decision related to promotions or putting employee on Performance improvement plan or termination as performance might be impacted due to personal problems, family issues or any financial crisis which AI might not understand well as humans do. There could be a situation where a 2nd top performer is in need of promotion more than one who is a top performer, in that case human intervention is required.

 

AI should take into considerations all the personal, medical, financial or mental aspects of an employee while performance evaluation and should pause and consult a human before taking decision. So I feel these parameters should be the criteria for AI to escalate.

As an internal audit function we could use AI in performing the compliance checks as per the contractual agreements, company's standard requirements, regulatory and statutory checks. AI should be able to validate the adherence to these requirements based on the evidences submitted. While checking the compliance to identify a Minor Non conformance and observations as per ISO standards is straight forward, identifying a Major Non conformance is not. This involves identifying the impact of a specific non conformance, it's relevance to the client/supplier's current situation, any financial impact  this may have and if any process breakdown happened or if no process has been created in the first place. Most of these scenarios don't have documented evidences as  these should be understood post discussion with the relevant stakeholders, approvals from the finance teams etc. So below are the possible scenarios to bring a human into picture.

1. In all scenarios where a contractual breach has happened, there should be a trigger to the Auditor, Auditee requesting any further details or approvals taken to prevent the Major NC
2. In cases where a fraud/financial breach has been identified, there should be a trigger to the stakeholders requesting for the amount of impact, any exceptions taken from the clients, action plans put in place

An AI agent can handle 90% of Queries related to Utilities Energy and Gas as these processes rely on taking Meter Reads, Billing and Payments of Customers. On occasion there may be Electrical and Gas Emergencies faced by Customers. The AI Agent may take the initial details of the Emergency but will have to refer to a human to complete the investigation. These scenarios present a unique challenge as the Emergency will have to be investigated by utilizing probing questions to establish if it is a Gas Emergency and what advise has to be provided to the Customer.

Example: Customer advises they have a smell of Gas in the home, they AI agent may not be able to advise of the full next steps -  to Turn off the Gas Meter at the main valve, evacuate yourself and any Family or pets, or if the building will need to be evacuated and the grid provider advised to turn off main gas supply for a Street or Town. I this type of scenario the AI agent should defer to a human.

 

In a Google content moderation process, AI should pause and ask a human in the following situations:

Edge Cases and Ambiguity

• Where the content's intent is unclear.

• When the content falls into gray areas of policy.

• If the user indicates sarcasm, satire, or nuanced language.

Policy Violations with Context

• Where the context of the content is an exception, but the content violates a policy

Appeals and Disputes

• For cases where we receive appeals on bot decision

• For disputes on the basis of bot categorisation on content or error classification during RCA

New Trends and Novel Content Types

• When new forms of non permissible content or spam/abuse trends emerge that the AI has not been trained on.

• If there are new ways to circumvent policies.

Sensitive Topics

• Content is for highly sensitive topics.

In shipping documentation, for eg approximately 90% of booking amendments can be managed by AI. Container size, cargo weights, descriptions, port changes, etc can be easily actioned by AI based on the predefined rules. 

 

Areas where human escalation may be needed could be for dangerous / reefer cargo specifications check or if the change impacts customer contract requirements or if there are changes to customs' requirements that are not input in the system. 

 

Similarly in a Customer Relations process, most changes can be managed well by AI for loyalty programs, updating/ changing member details, updating bookings or requests. Areas of escalation to humans would be when customer is irate or keeps choosing 'Not satisfied' with the response, legal compliance issues, or high value transactions which need further validations. 

One area that i think AI can handle in BPO industry is Creative Thinking & Innovation.

 

AI generates ideas by analyzing existing data and recognizing past patterns. While it can remix or optimize known concepts, it doesn't truly originate novel ideas. Genuine creativity often requires breaking away from established norms, taking risks, or challenging the rules—something AI isn't naturally inclined to do.

 

AI also lacks the deeper cultural, emotional, and market insights that are essential for meaningful innovation. It might offer technically sound suggestions that fall flat in real-world contexts. For instance, an AI chatbot might propose a marketing promotion during a service outage—logically consistent, but clearly tone-deaf from a human perspective.

In the BPO space, creativity is often tied to a broader business vision or brand philosophy—areas where AI falls short in grasping the "why" behind the "what." Humans are better equipped to align innovation with strategic goals such as long-term growth, social impact, or brand identity.

 

So, while AI can potentially handle 90% of the creative and innovative process, human intervention is still crucial to ensure the final output is contextually appropriate and aligned with broader business objectives.

Let's look at a process that a business process outsourcing (BPO) company does in the area of e-commerce: answering customer service questions after a purchase for an online store that sells clothes and accessories.

 

How an AI Agent Could Handle 90% of the Work:

An AI-powered Chabot, integrated throughout the retailer's website, app, and social media channels, can efficiently handle a high volume of post-purchase inquiries by:
• Provide real-time order status updates, retrieve tracking information from carriers, and answer delivery timeline questions.
• Assisting consumers with returns and exchanges, including policy information, automatic label generation, and system workflow.
• Answer basic billing inquiries, provide invoice copies, and process small adjustments, such as applying discount coupons missed during checkout (within established restrictions).
• Answering frequently asked questions (FAQs) about product maintenance, sizing, materials, and past purchases.
• Customers can alter their shipping address, contact data, and payment information prior to dispatch.

 

Scenario Requiring Escalation to a Human Agent:

A customer contacts assistance, saying that they received the incorrect item in their order. The AI agent verifies the order information and admits the discrepancy. It tries to start a typical exchange process, but the system says the initially ordered item is out of stock and won't be back in stock anytime soon.

 

AI agents should escalate interactions to humans based on the following criteria and signals:
1. System Limitation in Processing the Request: When the automated system is unable to satisfy the customer's request using regular workflows (for example, initiating an exchange for an out-of-stock item). The AI recognizes its incapacity to continue with the typical procedure.

2. Policy exceptions or deviations are required. The customer may request a solution that goes beyond the typical return/exchange policy. The AI is built to recognize certain policy exceptions.

 
3. Requires complex problem-solving and negotiation skills. When the circumstance necessitates problem-solving beyond pre-defined scripts (for example, the customer is extremely upset, numerous products in the order are erroneous, the supplied item is drastically different or damaged). This requires human judgment and the capacity to provide individualized solutions.

4. If a customer expresses strong negative emotions or uses aggressive language, the AI should prioritize de-escalation and transfer the interaction to a human agent with strong emotional intelligence and negotiation skills. Sentiment analysis can cause this.

5. If AI is unable to understand a customer's issue despite several clarifying queries, it should escalate to a human for more effective inference.

 

What Happens After Escalation:

If any of these escalation conditions are met, the AI agent should politely advise the consumer that their issue needs to be escalated to a human support representative for further assistance.

• Estimate when a human agent will contact you (e.g., immediate transfer or callback/email answer within a certain timeframe).

• Transfer conversation context, including order details, customer message, and AI troubleshooting procedures, to the human agent. This saves the consumer from having to repeat themselves.

• Categorize reasons for escalation (e.g., "out-of-stock exchange," "policy exception requested," "high customer dissatisfaction," "unclear issue") to assist human agents in swiftly understanding the situation and responding appropriately.

By utilizing these escalation triggers, AI can quickly handle the bulk of common post-purchase queries, freeing up human agents to focus on more complicated, exception-based, and emotionally sensitive cases, resulting in increased customer satisfaction and operational efficiency.

 

In one of our account, team has implemented a chat bot to handle customer queries as well as escalations. This chat bot works very efficiently to capture client input, provide tracking number, share it with BPM team members, does follow up and if necessary escalations till the issue is not resolved or provide resolution back to client. Also this chat bot backend data has analysis feature.  The one step where we need AI to escalate to human is for doing a Root cause analysis. While basis previous data may give certain inputs for Root Cause Analysis but it has to involve human to get more human inputs which data base cannot provide.

The most relevant and practical scenario that comes to mind is employing an AI Chabot which is capable of handling Customer Support/Service/General enquiries.

 

The flow of the chat bot should be as follows:

1.      Direct customers to self-help information and FAQ's.

2.      Guide on troubleshooting and

3.      Escalate or route to a specialized/complex/complaints dept. operated by humans to further intervene and resolve.

 

One such scenario is when the AI chatbot is dealing with Gas and power (Energy Utilities) customers that have requested an invoice to be rebilled to an accurate read as it constantly bills to an estimated read resulting in an inaccurate invoice.

 

Post all the rules of elimination via troubleshooting and there is still no satisfaction, the AI chatbot will need to route the error/issue together with the interaction and troubleshooting history to a technical dept./Human so that a physical inspection can be conducted on site to establish if there is any physical impairment or issue with the physical meter.

 

Thereafter a physical meter fix/replacement will need to be actioned, and the human agent can then post their updates and fixes to the customer and then hand back to an AI chatbot once the client/end customer has confirmed they are satisfied with the rebilling to an accurate bill/invoice.

Interesting case studies where AI needs to pass the flow to a human. 

 

The winning answer is provided by Hamid. Well done!

 

Answers from Diop Saliou, Vikas Choudhary, Sundar Nag and Vidhya Rathinavelu are also an interesting read.