How Thunai Automates Customer Support with AI Agents (and What It Means for Your B2B SaaS)

Customer support teams are under a pressure cooker. Ticket volumes rise 20% year over year, yet headcount stays flat. Customers expect instant, personalized answers around the clock. For B2B SaaS companies, every delayed response chips away at retention and expansion revenue. Thunai, a fast-growing technology firm, faced exactly this challenge, until they rearchitected support around AI agents and event-driven data streaming with IBM Confluent.

The result? A support function that scaled 3x without adding headcount, a dramatic drop in resolution time, and a CSAT score that climbed past 94%. But the real story isn’t just about technology, it’s about how connecting real-time data streams to AI decision-making transforms customer service from a cost center into a growth engine. This article unpacks Thunai’s approach and gives B2B support leaders a concrete playbook to replicate the results.

Why Traditional Support Stacks Break at Scale

Most SaaS support teams start with a helpdesk, a knowledge base, and a handful of macros. That works when you have 200 tickets a month. At 2,000, cracks appear. At 20,000, the system collapses. Support leaders face three interconnected problems:

• Data silos: Customer context lives in the CRM, product usage data in the analytics tool, past conversations in a separate ticket system. Agents toggle between five tabs just to understand a problem.
• Reactive workflows: Teams wait for customers to report issues. By the time an agent engages, frustration has already set in.
• Knowledge bottlenecks: The sharpest 20% of agents answer 80% of the tricky questions. When they’re unavailable, resolution times balloon and consistency drops.

Thunai recognized that fixing support at scale required rethinking the data layer, not just the ticket interface. They turned to IBM Confluent to unify real-time events, and built AI agents on top that could both interpret and act on that data instantly.

The Data Backbone: Why Real-Time Streaming Changes Everything

IBM Confluent provides a real-time event streaming platform built on Apache Kafka. Instead of polling databases nightly, every customer action, account changes, product feature usage, billing updates, error logs, flows as an event in milliseconds. For support, this means an AI agent can know that a customer just encountered a payment decline or a broken integration before the customer types a single word.

Thunai implemented the following event-driven architecture:

1. Event producers: Every microservice in their SaaS product and operational system (CRM, billing, monitoring) publishes events to Confluent topics.
2. Stream processing: ksqlDB and custom stream processors enrich events, matching a product error to the customer’s plan, contract status, and recent support history, and output a structured “customer situation” record.
3. AI agent trigger: When a situation meets predefined criteria (e.g., “enterprise customer on Pro plan encounters spike in API errors”), the AI agent is invoked automatically.

This architecture flips the support model from reactive to proactive. Instead of waiting for a ticket, the system detects a problem and initiates a resolution sequence.

AI Agents in Action: The Thunai Workflow

With a rich, real-time picture of the customer, Thunai’s AI agents operate across three tiers:

Tier 1: Automated Detection and Outbound Engagement

When a customer’s payment fails, the AI agent triggers an in-app message or email containing a secure link to update details, before the subscription is suspended. For known technical glitches, it sends a personalized message: “We noticed three timeout errors from your account in the last hour. Our engineering team is aware; here’s a workaround you can apply now.”

“We went from firefighting tickets to preventing them. AI agents now intercept 43% of issues that would have become support tickets.”, Thunai’s Head of Customer Operations

Tier 2: Conversational Resolution with LLM-Based Reasoning

If a customer does reach out, the AI agent is already context-aware. It has access to the customer’s recent events, full conversation history, product documentation, and past solutions that worked for similar profiles. Using a retrieval-augmented generation (RAG) pattern, the agent surfaces a precise answer or performs a sequence of API actions, resending a license key, updating a configuration setting, or granting temporary access, while keeping the customer informed in natural language.

Tier 3: Intelligent Escalation with Full Context Handoff

When the AI agent determines an issue requires human judgment (e.g., complex legal compliance questions or nuanced troubleshooting), it hands off to a specialist. But the handoff is game‑changing: the human agent receives a complete briefing, summary of the situation, steps the AI already took, relevant account details, and a suggested next action, all distilled into a concise panel. Average time-to-orientation drops from 8 minutes to under 30 seconds.

Measurable Business Impact: The Numbers That Matter

The proof is in the metrics. Within six months of deploying AI agents powered by real-time event streams, Thunai documented:

• 61% ticket deflection: A majority of inquiries were resolved through automated outbound actions or self‑service driven by AI‑triggered proactive guidance.
• Average first response time dropped from 8.2 hours to 2.1 minutes. This includes tickets that eventually needed human assistance, because AI agents provided an instant acknowledgment and often partial resolution.
• CSAT climbed from 78% to 94%. Customers consistently cited the “anticipatory service” as a reason for upgraded satisfaction scores.
• Cost per ticket fell 64%. By moving Tier‑1 and most Tier‑2 work to AI, human specialists focused on high‑value interactions, driving efficiency and job satisfaction.

The downward trend in ticket volume shows the compounding effect: as AI agents became more accurate and the event‑processing layer matured, fewer issues reached the support queue, and the ones that did were simpler to resolve.

Notably, the CSAT improvement wasn’t a one‑time bump. It kept rising as the system learned from every interaction, further fueling the virtuous cycle. And when you map this gain onto expansion revenue, B2B companies with CSAT above 90% see net revenue retention improve by 10‑15 percentage points, the ROI becomes clear.

The Playbook: How Support Leaders Can Adopt This Model

Thunai’s success wasn’t a “big bang” project. They followed a phased, measured approach that any support leader can replicate. Here is a practical framework:

Phase 1: Map Your Event Trinity (Weeks 1‑2)

Identify the three categories of events that most influence customer experience:

• Product events: error codes, feature usage, uptime anomalies.
• Account events: plan changes, billing failures, new user onboarding.
• Interaction events: recent support tickets, NPS scores, community posts.

Start with just five high‑impact events, like payment failure, critical error on a paid tier, or a customer downgrading. Connect them to a lightweight streaming layer (Kafka, Confluent, or even a webhook‑based bus).

Phase 2: Define Action Triggers and Agent Responses (Weeks 3‑4)

For each event, design an “if this, then that” playbook:

• What condition triggers an AI agent? (e.g., “error rate > threshold on enterprise account”)
• What is the best resolution path? (Send knowledge article, restart service, schedule call with CSM)
• What fallback ensures no customer is left hanging? (Escalate with full context after two failed AI attempts)

Use low‑code AI platforms to build these agents without writing backend code. Modern tools let you drag‑and‑drop data connections, define conversational flows, and integrate directly with helpdesks like Zendesk or Intercom.

Phase 3: Measure, Learn, and Expand (Ongoing)

Track the same metrics Thunai used, ticket deflection rate, first‑response time, CSAT, cost per ticket, in a unified dashboard. Review the top 20 AI‑handled conversations weekly to spot gaps. Expand event coverage and agent sophistication in two‑week sprints.

“The biggest mistake is trying to automate everything at once. Start with the 20% of issues that cause 80% of volume, nail that, then grow.”, Support Ops Advisor

Where Successly Fits In: Accelerating Your AI Agent Journey

Thunai’s architecture required deep Kafka expertise and months of custom development. But the principles don’t require that level of engineering investment. Successly offers a purpose‑built AI support automation platform that already connects to your product analytics, CRM, billing, and helpdesk, ingesting the same real‑time events Thunai relied on, without needing to manage a streaming infrastructure yourself.

Successly’s AI agents are pre‑trained on B2B support patterns, so you get out‑of‑the‑box detection of failed payments, onboarding drop‑offs, and common product errors. And because it integrates with the tools you already use, you can go from pilot to full production in weeks, not quarters. Teams that adopt Successly typically see:

• 50‑65% ticket deflection within the first 90 days
• First‑response times under one minute for AI‑managed interactions
• Over 90% CSAT on automated conversations

Conclusion: Turn Support into a Strategic Moat

Thunai’s story with IBM Confluent and AI agents is a blueprint for what modern support can achieve. By treating every customer signal as an opportunity to serve, they transformed a reactive cost center into an engine of loyalty and growth. For B2B SaaS leaders, the lesson is clear: the foundation of great AI support isn’t a smarter chatbot, it’s a smarter data pipeline. With platforms like Successly, you can deploy that pipeline and the intelligent agents that sit atop it, without the heavy engineering lift.

Your customers are already broadcasting the signals that determine their success and churn. Are you listening?