• Written by Muhammed Abdulla NC, CEO

When a customer service team implements an AI-powered workflow, they often expect to see a reduction in costs, but the reality is that the actual cost savings can be elusive, and the budget allocated to the project can be quickly consumed by unforeseen expenses, leaving the team wondering where the money actually went. The culprit behind this discrepancy is often the failure to distinguish between unit cost and total cost of the AI workflow.

The unit cost trap

The unit cost of an AI workflow refers to the cost of processing a single customer inquiry, whereas the total cost takes into account the entire volume of inquiries handled by the system. Many organisations focus on reducing the unit cost, without considering the overall volume of inquiries, which can lead to a situation where the total cost remains high, despite the lower unit cost. This is because the AI system may be handling a large number of inquiries, each at a lower cost, but the cumulative effect is a higher total cost.

A closer examination of the cost structure reveals that the unit cost is often influenced by factors such as the complexity of the inquiry, the accuracy of the AI model, and the level of human oversight required. In contrast, the total cost is driven by the volume of inquiries, which can be affected by factors such as the effectiveness of the AI system in resolving issues, the level of customer satisfaction, and the resulting repeat contact rate.

Real levers for cost optimisation

So, what are the real levers for cost optimisation in AI-powered customer service workflows? The answer lies in caching, batching, and model routing. Caching involves storing frequently accessed data in memory, reducing the need for costly database queries. Batching refers to the process of grouping multiple inquiries together and processing them as a single unit, reducing the overhead of individual inquiries. Model routing, on the other hand, involves directing inquiries to the most appropriate AI model, based on the complexity and type of inquiry.

By applying these levers, organisations can reduce the total cost of their AI-powered customer service workflows, while also improving the efficiency and effectiveness of the system. For example, a shift supervisor can use AsscherAi to analyse the inquiry volume and adjust the caching and batching parameters to optimise the workflow, resulting in significant cost savings.

Dependent data

The effectiveness of these levers depends on the availability and quality of customer service data, including ticket history, knowledge base, product and order records, and escalation history. This data is used to train and fine-tune the AI models, ensuring that they are accurate and effective in resolving customer inquiries. The data is also used to identify patterns and trends, allowing organisations to optimise their workflows and improve customer satisfaction.

For instance, a category manager can use AsscherAi to compare the performance of different AI models, based on their ability to resolve inquiries related to specific products or orders, and adjust the model routing parameters accordingly. This ensures that customers receive accurate and relevant responses to their inquiries, improving their overall experience.

Breaking the cycle

One of the key benefits of AI-powered customer service workflows is the ability to break the cycle of repetitive inquiries and root causes that are never addressed. By analysing the inquiry data and identifying patterns, organisations can feed back root causes to the product development team, allowing them to make targeted improvements and reduce the volume of inquiries. This, in turn, reduces the total cost of the workflow and improves customer satisfaction.

Furthermore, AI-powered workflows can help organisations to move away from a culture of answering the same issue from scratch each time, and instead, focus on providing proactive and preventative support to customers. This can be achieved by using AsscherAi to analyse the ticket history and knowledge base, and identifying areas where customers are experiencing recurring issues, allowing the organisation to develop targeted solutions and improve the overall customer experience.

Measuring success

So, how can organisations measure the success of their AI-powered customer service workflows? The answer lies in metrics such as first contact resolution, time to answer, and repeat contact rate. These metrics provide a clear indication of the effectiveness of the workflow in resolving customer inquiries, and the level of customer satisfaction. By tracking these metrics, organisations can identify areas for improvement and make targeted adjustments to their workflows.

For example, a customer service manager can use AsscherAi to track the first contact resolution rate and adjust the model routing parameters to ensure that customers are directed to the most appropriate AI model, based on the complexity and type of inquiry. This can be done by analysing the workflow data and identifying areas where the AI models are struggling to resolve inquiries, allowing the organisation to make targeted improvements and improve customer satisfaction.

Limitations and wrong choices

While AI-powered customer service workflows can be highly effective, they are not a panacea for all customer service issues. There are situations where the use of AI may not be the best choice, such as in complex or emotionally charged inquiries, where human empathy and understanding are essential. In these cases, organisations should consider using human customer support agents, who can provide a more personal and empathetic response to customers.

Organisations should also be cautious not to over-rely on AI-powered workflows, and instead, focus on developing a balanced approach that combines the benefits of AI with the strengths of human customer support. By doing so, organisations can provide a more comprehensive and effective customer service experience, and improve customer satisfaction. For more information on how to implement AI-powered customer service workflows, please contact us.