In 30 seconds
- An analyst finds patterns and explains them.
- An assistant answers questions or organises work.
- An autopilot can change the business state or interact with customers.
- The higher the action level, the more permissions, monitoring and recovery matter.
- Ask for the exact current workflow, data boundary, evidence and human checkpoint—not a model name.
Analyst, assistant or autopilot?
Useful for menu structure, aggregate demand, retention signals and campaign learning.
Useful for natural-language exploration, drafting and decision preparation.
Useful when permissions, audit trail and reversal are clear.
Potentially powerful; demands the strongest controls and evidence.
These are not maturity levels where every product should rush toward autopilot. A restaurant may reasonably prefer an analyst that shows its work over an agent that can silently alter a live menu. The right boundary depends on the cost of a wrong answer.
Five restaurant AI models—and where they are actually available
| Product | Best-fit job | Publicly documented behaviour | Australia status | Control question |
|---|---|---|---|---|
| Our review-first pick Zwift Gem | Menu, aggregate-order and campaign decision preparation | Menu Improvement, Order Analysis and Campaign Planner; evidence, assumptions and next moves remain reviewable | Current Zwift workflows; availability confirmed by proposal | Which recommendations are available for this Portal setup, and what data does each use? |
| Toast IQ | Broad operational intelligence inside the Toast ecosystem | Action-oriented AI assistant across restaurant data and workflows | U.S. availability documented; international availability planned | Which functions are live in Australia, under whose permission, and how are changes audited? |
| Square AI | Natural-language business insight for Square sellers | Current help material says it provides insights but cannot update the account | Australian optional open beta | What data scope and beta limitations apply to the restaurant? |
| Lightspeed AI | Intelligence across the Lightspeed retail and hospitality platform | AI-powered intelligence layer across the Lightspeed platform | Announced for Lightspeed Restaurant; confirm local plan | Which hospitality workflows are live in Australia and included in the selected plan? |
| SevenRooms Voice AI | Guest phone interactions and reservation demand | Voice AI within a guest experience and reservations platform | Regional availability not established by the cited page; confirm | How are calls handled, escalated, recorded and connected to the venue’s guest records? |
This is why a single league table would be theatre. SevenRooms Voice AI and Gem Menu Improvement are not trying to do the same job. Square AI’s current non-action boundary differs from an assistant promoted as action-oriented. Compare the job your team actually needs.
How Zwift Gem keeps the human in the loop
Menu Improvement
Examines menu structure, descriptions and aggregate product demand through practical lenses such as clarity, discovery, pricing, bundles and menu flow. The output is a set of reviewable opportunities—not a silent rewrite of the live menu.
Order Analysis
Turns aggregate demand, product, retention and operational signals into prioritised hypotheses. The workflow is designed around aggregate context and does not require customer identities or individual order records.
Campaign Planner
Organises audience, offer, channel sequence, timing, budget direction and measurement across tools such as vouchers, email, SMS and loyalty. It prepares a campaign for review; it does not automatically create the voucher or press send.
Gem’s strength is disciplined proximity. The relevant menu, aggregate-order and campaign context already sits beside the retailer’s working tools, so the output arrives where it can be inspected and used. That is practical hospitality intelligence: specific enough to guide a decision, bounded enough to keep the retailer in control.
Nine questions to ask any restaurant AI vendor
- What exact problem does it solve today? Ask for a live workflow, not a roadmap reel.
- What evidence goes in? Menu, aggregate orders, customer identity, voice calls, payments and third-party data carry different implications.
- Can the operator inspect why? Look for sources, assumptions, confidence or reasoning appropriate to the decision.
- What can it change? Separate insight, draft, approved action and automatic action.
- Who can approve it? Permissions should match the financial and customer impact.
- How is a mistake reversed? Audit trails and recovery matter more than a polished answer.
- Where does data travel? Ask about providers, retention, product improvement and venue controls.
- How is quality measured? A vendor should describe useful task-level evaluation, not only model capability.
- What costs extra? Confirm plan, usage, messaging, setup and service requirements.
AI red flags in a restaurant demo
- A recommendation appears without showing which venue evidence informed it.
- The presenter jumps between live capability, private beta and future roadmap without labelling them.
- “Personalisation” is claimed without explaining consent or customer data access.
- The system can publish, reprice or message customers but cannot show a permission and reversal path.
- A generic chatbot is presented as hospitality intelligence without demonstrating menu, order or service-day context.
- Accuracy is described with one global percentage that has no connection to the actual task.
The best demo prompt is your real decision
Bring one menu category that underperforms, one order pattern your team cannot explain and one campaign question. Ask the vendor to show the evidence, output, human checkpoint and next action for each. A useful system will become more concrete under scrutiny.
Review the evidence
Show me what Gem sees.
Explore Gem’s current Menu Improvement, Order Analysis and Campaign Planner workflows, including the decisions it deliberately leaves with the retailer.
Sources and review notes
AI products change quickly; current availability, beta status and plan access should be reconfirmed.
- Zwift: Gem Intelligence and Business Portal.
- Toast: Toast IQ announcement.
- Square: Australian launch and current help guidance and limitations.
- Lightspeed: Lightspeed AI announcement.
- SevenRooms: Voice AI.
Correction or updated capability? Email marketing@zwift.com.au.