Fries with that? Meet Matilda, the AI metahuman taking orders at a Tassie takeaway

Meet "Matilda," the AI metahuman taking orders in Tassie - what support leaders can learn

A local takeaway shop in Tasmania is using an AI "metahuman" named Matilda to take orders and handle quick questions. It's a simple use case with big implications for customer support teams: faster service, consistent answers, and fewer bottlenecks during rush hours.

If you lead support, this isn't sci-fi - it's a practical blueprint. Here's how to turn an on-premise AI assistant into real gains without breaking your operation.

What a metahuman can handle on day one

• Order taking with upsell prompts (sides, sauces, combos).
• FAQ handling (hours, location, dietary info, wait times).
• Basic triage and routing to a human when confidence is low.
• Queue updates and pickup instructions to cut repeat questions.

The value shows up in speed and standardization. Less rework, fewer errors, clearer expectations.

Why this matters for support leaders

• Stabilizes service quality during peak load when teams get stretched.
• Reduces repetitive tickets so humans can focus on edge cases and recovery.
• Creates clean data: intents, objections, common failure points.
• Shortens training time for new hires by turning AI scripts into playbooks.

A 30-day pilot plan (no overbuild)

• Day 1-3: Pick the top 15 FAQs and your standard order flow. Define "must hand off" rules.
• Day 4-10: Draft concise scripts. Keep answers under 2 sentences. Add one upsell prompt per flow.
• Day 11-15: Set confidence thresholds and escalation paths. Human handoff within 10 seconds, max.
• Day 16-20: Soft launch on one counter or kiosk. Shadow with a supervisor. Log every miss.
• Day 21-30: Iterate scripts weekly. Ship only what improves completion rate or reduces handle time.

Conversation design that actually works

• Tone: Friendly, brief, and directive. "I can help with your order. What would you like?"
• Turn length: 1 prompt, 1 action. Avoid stacking questions.
• Confirmations: Repeat key choices: "That's a large burger, no onion. Anything else?"
• Recovery: If stuck twice, escalate: "I'll bring in a teammate to finish this up."

Metrics to track from day one

• Order completion rate: % of orders finished without human help.
• Average handle time: Compare AI vs. human for the same order types.
• Deflection rate: % of FAQs resolved without human intervention.
• Handoff quality: Did the human get a clear summary? Measure time to resolution after handoff.
• Customer sentiment: Quick 1-5 rating post-interaction. Keep it optional and fast.

Risks and how to de-risk them

• Privacy: Post clear signage about data usage. Limit data retention. Offer opt-out. Review the Australian Privacy Principles.
• Accuracy: Lock down menu, pricing, and hours as authoritative sources. Version and audit changes.
• Bias and tone: Use neutral language. Test with diverse accents and speech speeds.
• Outages: Fallback to human immediately. Keep a visible "Talk to a person" option.
• Scope creep: Keep it to orders and FAQs. Expand only when metrics support it.

Team impact (and how to keep morale high)

• Position AI as a queue teammate, not a replacement. Show the data: fewer repetitive tickets, more time for complex cases.
• Rotate agents into "AI coach" shifts to review transcripts and improve scripts.
• Reward recoveries: praise the human save when AI hands off.

Simple escalation rules

• Two misunderstandings in a row → escalate.
• Allergy, refund, complaint, safety → escalate immediately.
• Confidence score below threshold → escalate with a one-sentence summary.

Where to build from here

• Add payment status and pickup notifications to cut "Is my order ready?" queries.
• Connect inventory so the AI never offers out-of-stock items.
• Offer multilingual support starting with your top two languages.
• Feed transcript learnings into training and product decisions weekly.

Tools and training

If you're mapping skills for your team - from prompt writing to AI QA - this curated learning path is a good starting point: AI Learning Path for Administrative Assistants. For process-focused teams, consider structured upskilling: AI automation certification.

Bottom line

Matilda shows how an on-site AI assistant can clean up queues and standardize service without getting in the way. Start small, set firm guardrails, measure relentlessly, and keep humans in the loop. That's how you turn a novelty into measurable support outcomes.