In a context where customers expect immediate answers, a proactive approach to customer service is central. Effective deflection is not about avoiding customers, but about providing access to tools that make it possible to find answers oneself – quickly and efficiently. With an intelligent deflection strategy based on Zendesk's AI agents and Flow Builder, the ticket volume can be reduced, agent resources freed up for complex tasks, and the customer experience improved.
What is deflection? A strategic approach
Deflection is the process of resolving the customer's enquiry before it becomes a ticket. It is a deliberate strategy that moves customer service from a reactive model, where problems are solved after they arise, to a proactive model, where enquiries are prevented as tickets. A successful deflection strategy results in:
- Reduced workload: Agents can focus on cases that require human empathy, complex problem-solving and strategic thinking.
- Lower operating costs: Every ticket that is deflected saves time and resources.
- Increased customer satisfaction: The customer gets the problem solved immediately, around the clock, without waiting times or repeated explanations.
- Valuable data insights: Insights into which questions customers ask, and where answers are found (or not found), provide knowledge about products, services and communication.
Tools for deflection: AI agents and Flow Builder
Zendesk provides two primary, but different, tools for effective deflection. Understanding the strengths and differences is crucial for building a robust strategy.
AI agents: the intelligent first responder
AI agents are Zendesk's AI-driven bot that functions as a digital first responder. It uses machine learning to understand the intent behind a customer's enquiry and match it against the knowledge base (Help Center). AI agents are well suited to handling a broad spectrum of unpredictable customer questions.
Core features:
- Automatic reply: Answers the customer's question by suggesting the most relevant article from the Help Center.
- Continuous learning: Improves on an ongoing basis based on feedback (e.g. "Was this answer helpful?").
- Seamless integration: Works together with agents. If the problem is not resolved, the conversation is handed over to an agent with the context retained.
Flow Builder: the structured dialogue guide
Where AI agents are designed for unstructured queries, Flow Builder is used to build structured and predictable conversation flows. Flow Builder can be used to design guided dialogues that lead the customer step by step through a specific process, like a digital receptionist with well-defined tasks.
When Flow Builder is the right choice:
- Specific processes: Tasks such as "Reset my password", "Check the status of your order" or "Book a meeting with a sales consultant".
- Collecting information: Collecting specific data before creating a ticket (e.g. serial number, order number, etc.).
- Full control: A need for full control over the dialogue and ensuring that the customer follows a particular path.
The synergy between AI agents and Flow Builder
The most effective implementation is achieved when the tools work together. A typical strategy is to let AI agents be the first point of contact. If the bot is uncertain or does not find a relevant answer, it can – instead of creating a ticket – offer to transfer the customer to a relevant Flow Builder flow. Example: A customer writes "I have forgotten my password". AI agents recognises the intent and starts the "Reset password" flow.
Configuring AI agents: from activation to optimisation
Implementing AI agents requires configuration and ongoing optimisation to ensure value.
Step 1: Activation and basic settings
- Navigate to Admin Center: Go to Admin Center → Objects and rules → Bots → AI agents.
- Activate AI agents: Select the channels (e.g. Web Widget, email) where AI agents should be active.
- Connect the Help Center: Make sure that AI agents is connected to the correct knowledge base. Article quality is crucial for bot performance.
Step 2: Fine-tuning the response logic
This step is central to balancing helpfulness and disturbance.
-
Confidence Threshold: The percentage of certainty that AI agents must have before an article is suggested.
- Too high a threshold (e.g. 95%): The bot rarely replies, which gives low deflection and more tickets.
- Too low a threshold (e.g. 50%): The bot suggests irrelevant articles, which can frustrate customers and damage credibility.
- Recommendation: Start with a medium-high threshold (e.g. 75–85%) and adjust based on feedback and performance data.
-
Number of suggested articles: Configure whether AI agents should suggest one, two or more articles. More suggestions can increase the likelihood of a correct answer, but can also seem overwhelming.
-
Fallback handling: Handling for when AI agents cannot help. Options:
- Create a ticket: The default setting.
- Show a list of articles: The customer chooses from a broader list.
- Hand over to a Flow Builder flow: An effective option.
- Offer contact with an agent: The option to proceed to an agent.
Step 3: Continuous training and improvement
AI agents require ongoing follow-up.
- Analyse feedback: Regularly review which answers are marked as "helpful" and "not helpful".
- Optimise articles: If an article is often marked as "not helpful", this indicates a need for improvement, updating or better labels.
- Add missing articles: Recurring questions that AI agents cannot answer indicate a need for new articles.
Strategic implementation: a step-by-step guide
A successful rollout requires a strategic plan.
Phase 1: The foundation – a well-functioning Help Center
This is the most important step. AI agents depends on the quality of the available knowledge.
- Article quality: Clear, concise and easy-to-understand articles using the customer's language.
- Search optimisation: Unique and descriptive titles as well as relevant labels and categories, so that AI agents can match correctly.
- Ongoing updating: Regular review and updating to ensure correct and current information.
Phase 2: Implementing AI agents
Start on a smaller scale and expand gradually.
- Choose a channel: Start e.g. with the Web Widget on the website.
- Set a conservative confidence threshold: It is better to start with too few answers than too many incorrect ones.
- Communicate clearly: State that the customer is interacting with a bot, and provide a simple option to proceed to an agent.
Phase 3: Advanced flows with Flow Builder
Once AI agents is working stably, the most repetitive and simple tasks can be identified and supported with dedicated flows. Analysis of ticket data can reveal frequent topics such as "How do I return an item?" or "Where is my order?".
Phase 4: Measurement, analysis and iteration
Improvements require measurement. Define KPIs and follow them on an ongoing basis:
- Deflection Rate: The proportion of enquiries resolved without creating a ticket.
- Containment Rate: The proportion of bot conversations resolved fully by the bot.
- Resolution Time for Deflected Issues: How quickly the customer finds the solution themselves.
- CSAT for Bot Interactions: Customer satisfaction with bot interactions.
Best practices for maximum effectiveness
- Quality over quantity in the knowledge base: 100 strong articles are better than 1,000 mediocre ones.
- Use the customer's language: Avoid internal jargon in articles; write from the customer's perspective.
- Experiment with the confidence threshold: Adjust gradually and follow the effect on deflection and CSAT rates.
- Ensure clear fallback handling: The handover to an agent or flow must be smooth.
- Set up dashboards: Use Zendesk Explore for dashboards that show bot performance over time.
- Involve agents: Agents have insight into the most frequent questions and can help prioritise articles and flows.
Troubleshooting and challenges
Even with a good configuration, challenges can arise.
Problem: Low deflection rate.
- Possible causes: The confidence threshold is set too high. The knowledge base is incomplete, or articles are difficult to find. The customer does not see the bot or the suggestions.
- Solution: Lower the confidence threshold gradually. Review and improve articles and labels. Test the user journey from the customer's perspective.
Problem: AI agents suggests incorrect articles.
- Possible causes: Ambiguous or poor labels. Titles are not sufficiently descriptive. The confidence threshold is set too low.
- Solution: Review and clean up labels. Make titles and content more specific. Raise the confidence threshold.
Problem: Dissatisfaction with the bot interaction (low CSAT).
- Possible causes: The bot comes across as robotic and impersonal. Questions are not understood. There is no simple way through to an agent.
- Solution: Adjust the bot's language to be more friendly and empathetic. Analyse "not helpful" answers for patterns. Make sure that the "Talk to a person" button is visible and easily accessible.
Conclusion: the path towards proactive customer service
Implementing AI agents and Flow Builder is both a technical and strategic effort that can change the approach to customer service. Deflection can contribute to a faster, more efficient and more satisfying experience for customers, while agent resources are freed up for tasks of higher complexity. With a targeted strategy, a self-running system can be built that continuously improves through data and feedback.