A digital display showing a Lumière AI host welcoming visitors in a premium GCC mall

Buyer's Guide

Eight questions every serious venue operator should ask before deploying an AI character. Including us.

The AI character market is young and not always clearly defined. Vendors range from web chatbot companies adding a face to purpose-built physical AI platforms. Before you brief anyone, including us, here's how to tell the difference.

Read the 8 Criteria ↓ Take the Scorecard →

This page doesn't pitch Performit Live. It gives you the framework to evaluate any vendor, including us. If we're the right fit, the criteria will show it.

The Framework

Eight criteria. Ask every vendor.

Physical environment deployment

AI characters built for websites behave differently in physical environments. The interaction model, the hardware, the latency requirements, and the governance model are categorically different.

Ask "Has this been deployed in a physical environment at commercial scale, not a pilot?"

Good Named deployments or anonymised case studies with real visitor volumes and specific hardware configurations.

Red flag "We're piloting with several enterprise clients" or web-only references.

Behavioural governance

In a flagship mall or museum, an off-script response is a brand incident. You need pre-approved content boundaries, not a general-purpose language model.

Ask "How do you prevent the character from saying something off-brand or factually wrong?"

Good Defined knowledge base, governed response boundaries, no off-base improvisation.

Red flag "The AI is very accurate" or "we have content moderation."

Multilingual capability

Flagship malls and museums have international visitor mixes. A solution that handles two languages is not a solution.

Ask "How many languages, and how does switching work mid-conversation?"

Good Real-time detection and switching across multiple languages, demonstrated live.

Red flag "We support multiple languages" without specifics on the switching mechanism.

Integration capability

An AI character that can't connect to your booking system, tenant directory, or indoor mapping platform delivers a fraction of its potential.

Ask "What integrations do you support, and what does the process look like?"

Good Named integration partners, clear API process, standard integration in deployment fee.

Red flag "We can integrate with most systems" without named examples.

Data ownership and residency

Visitor interaction data is commercially valuable. You need to own it, and it needs to stay in the right jurisdiction.

Ask "Who owns the visitor interaction data? Where is it stored?"

Good You own it. Stays in-region. No selling or aggregation. Data architecture available before contract.

Red flag Vague data policy, unclear ownership, "we comply with GDPR" as the full answer.

Deployment timeline and process

A vendor who quotes 2 weeks to deployment hasn't built anything serious. A vendor who quotes 18 months is overcomplicated. The right answer reveals process maturity.

Ask "Walk me through the deployment process from brief to live."

Good Defined phases with named timelines, typically 8-12 weeks for Performit Live deployments.

Red flag "It depends" with no structure, or suspiciously fast quotes with no integration phase.

Visual quality and brand design

The character represents your brand every day. If it looks like a tech demo, it damages the brand standard your tenants have worked to establish.

Ask "Can we see the character running live, not a video, not a demo reel?"

Good Live demo, immediately, with the actual conversational interface.

Red flag Video demo only, pre-recorded responses, "we can arrange something later."

Commercial model and ongoing costs

Many AI character vendors charge a low day-one fee then accumulate costs through per-conversation pricing, integration add-ons, or knowledge base update fees. Know the full commercial picture before signing.

Ask "What's the full commercial model: deployment, monthly, ongoing? What triggers additional cost?"

Good Clear all-in pricing structure, transparent on what's included and what costs extra.

Red flag "We'll tailor a commercial proposal" with no indicative structure at all.

Vendor shortlist

How Lumière AI fits with what you already run.

Pick the technology in your stack. We'll show you where Lumière AI sits in relation to it.

Including us

Ask us all eight. We welcome it.

Every question on this list applies to Performit Live. We'd rather you ask all eight and decide we're the right fit than sign a contract based on incomplete information.

If we're not the right fit for your property or timeline, we'll tell you that too.

Questions

Frequently asked questions

Does this criteria apply to Performit Live?

Yes. Ask us every question on this list. If we're the right fit, the answers will show it. If we're not, we'll say so.

How long should evaluating an AI character vendor take?

For a flagship deployment, 4-6 weeks of evaluation is reasonable: demo, reference check, integration assessment, commercial review, and legal/data review. Rushing this stage creates deployment problems later.

Should we issue a formal RFP?

For enterprise procurement, yes. We welcome RFPs and have responded to several. If you're building one, the eight criteria above are a strong starting framework for the evaluation criteria section.

How many vendors should we evaluate?

Three is sufficient. The market is small. There aren't 10 credible vendors for physical AI character deployment. Shortlist based on physical deployment proof first; eliminate anyone who can't demonstrate a live, commercial deployment in a public environment.

Ready to put us through the eight criteria?

Book a discovery call. We'll answer every question on this page, live, with the product running in the room.

Book a Discovery Call → Take the Scorecard →