Leaderboard

Who Should Be Responsible when AI Recommends? Let see this through a BPO Case Study in Customer Complaints Escalation. The Business Process Outsourcing (BPO) sector is one of the areas in which efficiency and accuracy are the most important factors that explain the reason why AI systems are implemented into the operational workflows to provide advice to human agents. Customer complaint escalation is one of the crucial and representative processes. In this case, AI models can process incoming complaints - through sentiment analysis and detecting keywords and historical data - to prescribe the cases that are to be flagged and sent up the hierarchy to senior specialists or the management as urgent concerns. The recommendation is sent to the agent who makes the final click: "Escalate" or "Do Not Escalate." The issue of accountability is complicated and controversial when results - an over-satisfied customer or a harmful churn - are achieved. Not only is clarity philosophical but it is also operational and a legal requirement. Relevance of the Selected Process. A high stakes BPO process is customer complaint escalation. Poorly handled grievance may result in a loss of revenue, damage of reputation, and regulatory backlash. Within this domain, AI recommendation engines are conditioned with thousands of previous tickets and can tell how severe the target is. Yet they act upon lines, rather than knowledge. They may also fail to notice a low-key threat issued by a high-value client or over-tag a complaint which is angry but mainstream. The process is a perfect example of the paradigm of AI recommends and human decides, which is why it would be better to refer to it when discussing accountability. The initiative under consideration is the implementation of the name Escalate-AI, a platform in use in one of the client service departments of a BPO to minimize the time spent on resolving the issue and avoiding the further escalation of small ones. Setting Responsibility Boundaries: Two-scenario Framework. Scenario A: The Recommendation of AI is adhered to and it goes wrong. The flag of a complaint that is observed by Escalate-AI as High Priority is followed by an agent. The case is hastened to an overworked manager who is distracted and not given any real crisis which is brewing in another department and the initially flagged customer ends up being not satisfied with procedural and robotic response. Who is responsible? AI/Developer Responsibility: The accuracy and clarity of its recommendation is the responsibility of the AI. In case the system was trained on biased information (e.g., giving excessive preference to some trigger words in particular demographics) or the logic of the system is a black box, the creators of the system and the organization, which deployed it, will be responsible to provide a flawed tool. Their design results in its very recommendation. Human Agent/Operator Responsibility: The human agent is left with the responsibility of adopting situated judgment. They possess the background the AI lacks: history of the customer, his tone of voice, and subtlety of the ticket story. They should consider the AI their professional companion, and not their superior. Accepting a recommendation and mindlessly implementing it without applying their human expertise is a lapse of their duty of care. Conclusion: The responsibility is collective, with the culpability biased. This responsibility to abrogate final judgment is placed on the human agent and the chain of his or her supervisors. The organization is liable to any system flaws in the AI that deceived a reasonable agent. Scenario B: The AI Recommendation is Disregarded, and Things Do Go Wrong. Escalate-IA suggests a complaint priority which is Standard Priority. However, the agent notices real distress and tries to amplify it by hand, but is prevented through a workflow rule which demands further reasons to go against the AI. The agent is frustrated but does not give up. The case goes out of hand in social media. AI/Developer Responsibility: In this case, the possible failure of the AI is in the false assurance. Even a Standard Priority tag can form a hazardous cognitive anchor and induce the agent to second-guess his own sound intuition. Moreover, in the event that the system design is pro-actively creating friction, or punitive actions of exceeding its recommendation, then the process, itself, is guilty. Human Agent/Operator Responsibility: The responsibility of the agent was to be the voice of the customer. Though they first of all discovered the risk they finally accepted a system they felt was rigid. It is their responsibility not to use the channels they have (e.g., report to a superior) when they assumed that the AI was erroneous. Process / Owner Responsibility: The third party that emerges in the scenario with criticality is that of process design ownership. They are heavily responsible to the managers that introduced a strict, override-squelching workflow. Their culture of human wisdom as something inferior and systemic risk-aversion superseding it was formed. Conclusion: The burden of responsibility moves a lot to process owners and managers. Their responsibility is to have a decision-support system that does not promote good human judgement. Adaptability of the Proposed Accountability Framework: The Three-layer Model. In order to avoid cross-blame and finger pointing, BPOs are advised to embrace a simple and practical accountability model with regard to the activities like the Escalate-AI. 1. Responsibility Level (Clear Ownership). - AI Providers/Data Scientists: These are in charge of the accuracy and the recall and clarity of the recommendation. They must be capable of providing confident ratings as well as write plain-English justification of each recommendation (e.g., recommend due to phrases X, Y, and such-like combinations in the history which brought about 80 percent of the time). - Process Owners / Operations Managers: Accountable as regards to decision making structure. They must devise processes that encourage deliberate overrides, must leave comments too short to counter AI (so you think, not penal), and must get audit deliverables regularly to optimize AI and human training. - Duty to Informed, Contextual Judgment: Humans - They are the final decision-makers and must be ready to act as an AI tool and not an assistant or an oracle. 2. The Digital Paper Trail: The Transparency Layer: The recommendation of the AI with justification, the action of the agent with justification of the agent (especially of an override) should be recorded in each ticket. This does not need to be blamed but learn. It offers a decision journal that is auditable. 3. The Continuity of Calibration (Refinement Level): They should look at successes and failures of decisions through a cross-functional council (Ops, QA, AI Team, Agents) on a weekly basis. Was the AI wrong? Was this activity by the agent prudent or careless? This loop of feedback will constantly enhance the training programs and AI model itself. Benefits gained though the initiative: Using this systematic approach to the Escalate-AI project, our BPO process of customer escalation shifted the disorganized blame into a systematized learning. Once considered a failure measure, override rates turned out to be an important source of knowledge. They disclosed edge cases not found by the AI, which are being improved in the model. Being granted explicit override rules and cognizant that they would be responsible of their decision made agents interact with the AI more considerably. It was not a zero-fault system, but a clear-lined accountability system as a result. The time taken to resolve went down and what was even more important was the rate of fatal complaint mishandling also went down sharply since the hybrid human-AI system had become robust and its points of failure was known and owned. Finally, even in the BPO sphere and other areas, accountability has to be designed, rather than determined retrospectively once AI makes a suggestion and a human actually makes a choice. It is a collective yet non-uniform burden, in which the clarity of role, openness of process, and a culture of learning systems are forthcoming as opposed to the issue of finding one guilty figure.

Domain : Manufacturing : Oils and Gases Context : In Air separation unit the process is maximum hazardous and very sensitive, always on trigger of process failure(Even in stable condition) even due to small errors and internal noise or external noise, this leads to high risk of safety to the Employee, Environment, Surrounding and assets this is safety concerns due to critical characteristic of products and process, sensitive operations of high pressure and temperatures and other parameters. also the On supply to On site customer is so critical it’s mandatory to be on top and vigilant in managing the process and plant consistently. Intent : The intent was to build stable Artificial Intelligence Predictive and control Model to ensure high ‘’Safety and customer service and performance’’ even from start of the process and then maintaining the End to End stable process parameters which leads to better temperature and flow distribution and pressure ratios to attain the desired cryogenic product out put. HOW and what considerations are made to build the below AI Model : AI has emerged has assistant, guide and consultant to review the present process conditions been operated based on real online data and analyse that in real time in few seconds and make the suitable decision to the bring the process bias, to reflect the process output the intended output for highly safe and to ensure high Safety and customer service and performance To build the trust on AI Model by the operator, process operation team we have involved then in the design, considered all technical details, design conditions of the End to End process Lets say from high capacity Air compressors and high capacity Turbines and communication flow to End to End stake holders and taken all suggestions that would call a need for inclusion in the AI Model as Operators and Process Owners being the face of the process and close to reality they know the process very well. Major methodology Executed. Risk analysis, Brain Storming, Suitable best considerations were made, solutions were identified & Evaluation. The solution was built, tested. Simulations were done with involvement of Shop floor operators and Process Owners Ensured trust building and empowerment to Shop floor operators and Process Owners by involving them in End to End development and till Go-Live Commissioning. Control Measures taken to sustain the implementation and developed Trust While on implementation we had the question ‘’How should responsibility be defined’’ to ensure the recommendations from AI predictive model is followed and ignored, what necessary actions to be taken and by whom, who is responsible, accountable, Concerned and Informed. AI has emerged has assistant, guide and consultant to review the present process conditions been operated based on real online data and analyse that in real time in few seconds and make the suitable decision to the bring the process bias, to reflect the process output the intended output. But the question remains is if ‘’The AI’s recommendation is followed, and things go wrong?’’ How should the plan of action to be taken & who should be responsible for it, What could be best RACI ? & The AI’s recommendation is ignored, and things go wrong? As AI Model doesn’t provide a definitive Fix, it’s not a solution or application which is not affected by any External or Internal Factor and it’s not one time installed or invested and forgotten. It’s to be tracked and treated as element of evolution and a process of Continuous improvement and evolution really holds good for this to assure system should not fall behind when the AI solution keeps merging with new horizons. Below Structured approach was deployed : Tools and System : Below tools and systems were implemented to find the true structure approach so that team is not lost with lots of information, change and new updates and would help to follow the simplified necessary steps, eliminate Non value addition, avoid confusions, blame, or risk-avoidance behaviour or perform the activities with risk. System & Standards : RACI Matrix : RACI Matrix was defined and developed to ensure the structured flow to take the actions, by whom, who is responsible, accountable, Concerned and Informed. This was evaluated with cross functional heads and team members, communicated end to end on RACI so that all are aware and aligned. SOP : Standard Operating Procedure : SOP revisions with new versions : Updated Step by Step SOP are created for each process lines and considered new operating conditions with AI predictive Model mentioning, below important detailed needs are incorporated. When to take actions when things can go wrong with AI Proposal and AI Proposals are Ignored When to Ignore the AI recommendation ? When to override and take the control in manual ? How to treat, What actions to be taken when the Alarm given by AI predictive system ? What to actions to take When the sudden and unexpected internal or External Noise appeared in the system ensure ? Standards : Standards were created Standard Documents creation and approvals as per approval hierarchy like creator, reviewer and validation, approver Review and revisions for new versions with updates so that change is tracked. Tools & Systems : ONE DOC : Application/Software to manage the Standards was implemented, with ‘’Definitive Identification Number’’ this has ensured the live availability of right standards documents in the Library to the Process Owners Alert system & Action Log Books & Reporting : Implemented specific action Log book via Online Share point action log book, in which Operators and Process Owners mention the alarm, deviation from AI system and what actions were taken, remarks for the difficulties faced, also the need of further evolution in the AI program to Nullify the error or Noise,, or to mention the Idea if any ? Share point (Online) : Online Share point action log book, in which Operators and Process Owners mention the alarm, deviation from AI system and what actions were taken, remarks for the difficulties faced, also the need of further evolution in the AI program to Nullify the error or Noise,, or to mention the Idea if any ? MOC : Management of Change Management of Change is ensured and given a clarity and facilitated the Live updated Document at any point of time, this has helped to avoid confusions, blame, or risk-avoidance behaviour. Conclusion : AI Model provides solutions which is probably close the stable process as needed by the standards but there are high chances also that AI suggestions needs to be thought about and questioned to avoid the failures due to possible wrong suggestions provided by AI by not taking Noise in to considerations or additional necessary prompt in it’s background verification and suggestions. So End to End thought out Structured process is to be implemented as mentioned above to ensure no confusions, mistake, blame game, to held accountable and responsible, to keep the risk behaviour in control by the AI system also by human.

I am working in a Scientific Research Publishing company. Our reputation and the Brand value is entirely built on the accuracy with which we are producing the scientific articles, this accuracy then results in acquiring the trust of the authors and as well as the audience from the research community who are reading and consuming our content Today's question is very interesting and for that I would like to explain a scenario we experienced in our organisation. Scenario: currently we have deployed an AI based solution to check the content correctness and completeness as well as to check the Plagarism and any kind of text fabrication which is been done by the author These checks are performed using the AI based automation and AI calls that particular manuscript pass or fail. And after that we send it for the actual even best beer review system where they analyse the scientific content, equations and chemical compositions. They perform those relevant steps before publishing the manuscript but the Plagiarism and the content typesetting and corrections are checked by using AI Recently we were using AI, which is supporting the editors in their work faster and efficient. But in few cases, when we obseved that AI failed to work. The AI failed to capture the plagiarized content and as per the process the manuscript was published. After that few authors or researchers they came back to us mentioning the article they are referring is published by us is having copied data, the research is not unique and it looks like it is the fabricated one. Our changed view towards the AI usage : Here in this case we can clearly say that AI fail to perform the required task and in a way we can say it has lied to us. AI mistakenly accepted those articles which we shouldn't have a accepted if checkes manually. Though AI has given incorrect output that doesn't mean AI could be held hundred percent responsible, because we are in scientific publishing domain and we are the get keepers here who are responsible to analyse the data and the content and we are responsible to publish the right content everytime. Hence, in our case here AI is just an enabler which was helping us and our editor to perform their task effectively. This doesn't mean that whatever answers in produce using AI is 100% accurate. Being the scientific content publisher, we cannot put the entire blame on the AI system. We can consider AI as our helper or junior editor assistant which is there to help us. AI if that commits mistake then we should take the responsibility of that mistake, because it's always the journal manager who is there to sign and accept the article at the end of the review. The final review is always lying with the human touch. To conclude this, we can say, AI can give us evidences but the humans can provide the judgements. While considering the AI based solution, we can delegate the task to AI but we can not delegate the responsibility to AI.