How a 2-Person Support Team Can Handle 30,000 Monthly Tickets With AI

Summary

• Dutch software company Jortt uses an AI agent to autonomously resolve 92% of its 30,000 monthly customer inquiries, freeing up its two-person team for complex issues.

• The most effective AI support strategy is a hybrid model, where AI handles high-volume, repetitive tickets and directly escalates nuanced cases to human agents.

• Success depends on a four-phase playbook: training the AI on real documentation, setting clear escalation rules, providing for smooth live handoffs, and continuously monitoring analytics to close knowledge gaps.

• A hybrid approach requires a unified platform with both AI resolution and native live chat to prevent context loss during handovers, a core feature of Wonderchat's AI Chatbot Builder.

Jortt, a Dutch accounting software company, has an AI support worker named Femke. Every month, Femke autonomously resolves 92% of 30,000 inquiries — roughly 27,600 tickets — without a single human touching them.

The two-person team behind her isn't drowning. They're not burned out. They're focused entirely on the remaining 8%: the complex, nuanced, genuinely interesting cases that actually require human judgment.

Jortt's founder, Hilco, put it plainly: the work left for humans is "far more interesting." That's not marketing copy. That's the operational reality of a team that built the right system.

If you're running a small support team right now, you've probably heard pitches like this before and thought: "The marketing for this stuff always makes it sound like magic." You've seen the frustrating bot loops. You've watched customers get impatient when a bot keeps asking "did that solve your problem?" when it clearly didn't. You know that chasing complete automation is probably the wrong goal if you care about quality.

You're right on all counts. And this article isn't going to tell you to automate everything.

What it will show you is a four-phase operational playbook — the exact setup that makes results like Jortt's possible — built around a hybrid approach where AI handles the high-volume, repetitive tickets and humans handle everything that genuinely needs them. The goal isn't 100% automation. It's zero-touch resolution for the right tickets, a smooth handoff for the rest, and a continuous feedback loop that makes the whole system smarter over time.

Here's how to build it.

Phase 1: Train the AI on Your Real Docs

Estimated time: 1–3 hours for initial setup

The single biggest reason most AI support agents fail is that they're trained on the wrong data. Generic FAQs. Sanitized help articles written for SEO. Documentation that nobody actually read — or worse, documentation that's six months out of date.

As practitioners who've built these systems from the ground up have noted: "The key was feeding it real past support threads instead of generic FAQs, then routing anything uncertain to a human with full context." That's not a nice-to-have. It's the entire foundation.

Your AI agent for a small support team is only as good as the knowledge you put into it. Here's what to gather before you start:

• Help documentation and knowledge base articles — everything your agents currently reference

• PDFs and policy manuals — the actual source-of-truth documents, not summaries

• Past support conversations — especially resolved tickets from your most common query categories

• Product spec sheets or technical documentation — if your product is complex, your AI needs to be too

• Your entire website — pricing pages, onboarding flows, feature pages

Wonderchat is purpose-built for this kind of knowledge ingestion. It can ingest 20,000+ pages of technical documentation and deliver precise, source-attributed answers — critical when your customers are asking specific questions that require specific answers, not vague deflections to an FAQ page. You can upload files (PDF, DOCX, TXT, CSV), crawl your website directly, or sync with existing helpdesks like Zendesk. It even pulls and displays images and diagrams from PDFs inline in the chat window.

Automatic re-crawling means your AI stays current as your docs change — a detail that matters more than most people realize. An outdated knowledge base is a liability, not an asset.

According to Freshworks, 91% of customers prefer using self-service options before reaching out to support. Your AI is the ultimate self-service tool — but only if it actually knows what it's talking about.

Start here: Audit your top 20 most common support tickets from the last 90 days. Make sure your knowledge base has a clear, specific answer for each one before you go live.

Phase 2: Set Resolution vs. Escalation Rules

Estimated time: 30–60 minutes to configure

Setting system guardrails requires a critical distinction most teams skip: the difference between true resolution and deflection.

Deflection is sending a customer a help article link and closing the ticket. Resolution is the AI running a workflow end-to-end — answering the question, providing the exact steps, confirming the customer is satisfied — with no human involvement required. True automation happens when the system runs workflows and closes the ticket without agent intervention. Routing a ticket or pointing someone to a doc doesn't count.

To set up proper escalation rules, you need to answer two questions:

1. What should the AI always handle?

Focus automation on your highest-volume, lowest-complexity categories first:

• Account access and password resets

• Order status and tracking inquiries

• "How do I…?" procedural questions

• Billing and pricing clarifications

• Standard onboarding questions

These are the tickets where the answer is almost always the same. A well-trained AI handles them faster than any human can — and with just 2 messages to resolution, the customer experience is actually better, not worse.

2. When should the AI hand off?

Configure your AI confidence thresholds so the system escalates automatically when it's uncertain. The rule of thumb: if the AI isn't highly confident it has the right answer, it should say so immediately and offer a human — not guess, loop, or hedge. Nothing destroys customer trust faster than a bot that keeps asking "did that solve your problem?" when the customer has already said no three times.

In Wonderchat's Human Handover settings, you can configure automated triggers including:

• Message count triggers — escalate after N back-and-forth exchanges without resolution

• Confidence-based triggers — escalate when the AI determines it can't answer accurately

• Topic-based routing — send specific issue types directly to the right department, such as:

  • Billing disputes

  • Legal questions

  • Urgent technical issues

This is the operational backbone of the Jortt model. Femke doesn't try to handle everything. She handles exactly what she's confident about — and escalates the rest cleanly.

Phase 3: Configure Live Handoff

Estimated time: 15–30 minutes for setup

The handover is where most AI-only chatbot tools fall apart.

"Customers get impatient fast if the bot can't smoothly hand them over to a real person." That's not an edge case — it's the moment that determines whether your customers trust your support or resent it. The handoff has to be direct, immediate, and context-rich. The human agent who picks up the conversation needs to know exactly what happened before they joined.

This is the core reason a native AI + live chat hybrid matters. If your AI tool and your live chat tool are separate products bolted together with middleware, you will lose context at the seam. You'll also pay significantly more. One Wonderchat customer switched platforms specifically because "you guys have both live chat" — a wedge feature that competitors simply don't offer natively.

Here's the exact setup in Wonderchat:

1. In your dashboard, go to Chatbots → Actions (⋮) → Edit Chatbot

2. Open the Human Handover tab

3. Enable Human Handover — this adds a live chat icon directly to your chat widget

4. Configure your handover settings:

   • Escalation success message — what the customer sees when they request a human

   • Trigger for number of messages — auto-prompt for handover after X exchanges

   • Trigger for fail-to-answer — automatically offer a human when the AI can't respond

   • Contact emails — where the system routes escalations (your helpdesk, your inbox, or Zendesk)

   • Custom form fields — collect name, email, and issue type before the handover, so your agent starts with full context

5. Click Save

The result: when a customer needs a human, the transition is instant. Your agent sees the full conversation history. There's no "can you explain your issue again?" — the single most frustrating sentence in customer support. The human picks up exactly where the AI left off, with everything they need to resolve the issue in one interaction.

For teams already on Zendesk, Wonderchat acts as an AI layer on your helpdesk — the AI deflects Tier 1 before tickets are even created, and only the genuine escalations reach your queue.

Phase 4: Monitor Gaps with Analytics

Estimated time: Ongoing — 1–2 hours per week

A successful AI system requires ongoing monitoring, not a set-it-and-forget-it approach. "You can't expect these systems to be infallible on day 0. They require training and testing to spot the gaps." This is the phase most teams underinvest in — and it's the one that turns a decent AI agent into an exceptional one.

Track these four KPIs from week one:

• Containment rate — what percentage of conversations are fully resolved by the AI without escalation? This is your headline metric. Jortt is at 92%. Most teams start somewhere between 60–75% and improve from there.

• Escalation rate — what percentage are handed to humans? This tells you your AI's ceiling.

• Knowledge base gap rate — which questions is the AI consistently failing to answer? This is your content roadmap. Every unanswered question is a gap in your docs that you can close.

• Customer satisfaction (CSAT) — how do customers rate their AI interactions? Drop below a threshold and you need to investigate immediately.

The most important distinction is between containment rate (the AI fully resolved it) and deflection rate (the AI just kicked the customer away). Teams that track only deflection rate often fool themselves into thinking their AI is performing better than it is.

Wonderchat's analytics dashboard is built specifically to surface knowledge gaps, content quality issues, and KB weaknesses. Keytrade Bank uses it as a "content quality sensor" — every time the AI fails to answer a question, it flags a gap in their documentation.

Over time, the support function becomes a proactive content improvement engine. Your AI gets smarter. Your docs get better. Your containment rate climbs.

Hilco from Jortt described this exact feedback loop: "We're learning how AI and our customers think, and rewriting our help docs accordingly. Everyone sees this as the future — an opportunity, not a threat."

What Your 2-Person Team Looks Like After This

Before this setup: two people manually processing 30,000 tickets per month. Drowning in password resets, status checks, and "how do I…?" questions that eat every hour of every day.

After this setup: the AI handles ~27,600 tickets. Your two-person team handles ~2,400 escalations — with full context already in hand, routed to the right person, and prioritized by complexity. Every ticket that reaches a human is worth a human's time.

That's the operational reality this playbook produces. It's not magic — it's the same model running at Jortt (92% of 30K/mo) and Encompass (75% of 30K/mo), built on a specific feature stack in one product, without middleware:

• AI resolution

• Native live chat

• Analytics

The goal was never to replace your support team. It was to give two people the leverage to do work that actually matters — and to build a system that gets better every single week because you're watching what the AI misses and closing the gaps.

Frequently Asked Questions

What is a hybrid AI support model?

A hybrid AI support model is a system where AI and human agents work together to resolve customer inquiries. The AI handles high-volume, repetitive questions for immediate, zero-touch resolution, while automatically escalating complex or sensitive issues to the human team. This provides customers with the right level of support.

How does an AI support agent learn to answer questions?

An AI support agent learns by ingesting and analyzing your company's specific knowledge sources. This includes help documentation, knowledge base articles, past support conversations, technical manuals, and even your website content. Training on your real, internal documents allows the AI to provide accurate, context-aware answers instead of generic replies.

Will AI replace my human support team?

No, the goal of a well-designed AI support system is not to replace humans, but to augment them. By automating the resolution of common and repetitive tickets (often up to 90% of volume), the AI frees up your human agents to focus on the complex, high-value, and nuanced issues that truly require their expertise, ultimately making their work more interesting and impactful.

What are the best types of customer support tickets to automate with AI?

The best tickets to automate are high-volume, low-complexity inquiries with predictable answers. These typically include questions about account access and password resets, order status updates, basic "how-to" procedural guides, billing and pricing clarifications, and standard onboarding queries.

How do you prevent customers from getting stuck in a frustrating bot loop?

You prevent frustrating bot loops by setting clear and intelligent escalation rules. A modern AI support platform should be configured to automatically hand off a conversation to a human agent when it doesn't have a high-confidence answer, after a certain number of exchanges, or when a customer uses keywords indicating frustration. This direct handover helps the customer get help quickly without getting stuck.

How is this different from a traditional chatbot?

A traditional chatbot relies on pre-programmed scripts and keyword matching, which can be rigid and easily fail. An AI support agent, like the one described in this playbook, uses generative AI to understand intent and context, providing nuanced answers directly from your knowledge base. It also integrates natively with live chat for a smooth human handover, a feature most basic bots lack.

Start Building

The fastest way to test this is to build your first AI agent and run it alongside your team.

Wonderchat's free plan lets you train an AI agent on your help center and docs in under 5 minutes. No middleware. No separate live chat tool. No rebuilding your helpdesk. The same platform that powers Femke at Jortt — AI resolution, native live chat, and analytics — is available today at a fraction of the cost of a single hire.

Train it on your real docs. Set your escalation rules. Configure your live handoff. Then watch the containment rate climb.