How Shopping Centres Generate Revenue From AI Characters

The question has changed

Shopping centres already collect intelligence. Footfall counters, WiFi analytics, loyalty cards, mobile location signals, tenant POS. The aggregate picture has been improving for a decade. What is new is the layer that talks to the visitor in front of it: a branded AI character that turns a question into a recommendation, a routing event, and a logged intent signal in the same moment. That layer is what we call Venue Intelligence, and it is the part of the stack that owners have not yet learned to price.

The first wave of those deployments has been framed as marketing: an avatar on a screen, a digital concierge bolted on as an experiment, a novelty for the half-term holidays, a press photo for the launch event. Owners paid for it. Nobody asked it to pay rent.

That framing already undersells the asset. US mall traffic is recovering in 2026 (indoor centres up 4.5% in Q1, open-air up 6.4%, per Coresight Research / Capital One Shopping), but anchor tenants are negotiating harder on base rent and the leasing mix is tilting from goods to F&B, services and experiences. Food and beverage now leads US retail leasing activity, per ICSC’s 2025 Retail Resilience report. What centres still own outright is attention inside a physical destination, and that attention is becoming scarcer in the wider economy. An AI character is the most flexible instrument an owner has ever had for monetising it.

The right question is no longer “should we have one?” It is: what is the revenue stack, and which streams move our P&L?

This piece is post 03 in our Venue Intelligence series and sits squarely on the commercial model: seven distinct revenue streams a shopping centre can run through a single AI character, grouped against the three modes that make the system work.

The three modes: advertise when idle, converse when approached, learn from every interaction

Strip away the avatar, the voice, the screen, and Venue Intelligence is a configurable commercial surface, a digital human that earns rent, running in three modes simultaneously:

• Advertise. Idle revenue. The system earns before anyone arrives, through context-aware tenant campaigns on a sponsorable media surface you control. Retail media is now a $140B+ global category, roughly one-fifth of all worldwide digital ad spend, per eMarketer’s worldwide retail media forecast, and still growing. Inside-the-venue inventory is the part most owners haven’t claimed yet.
• Engage. Live interaction. The AI character (Lumière AI, in our case) activates when a visitor approaches and turns dwell time into transactions through conversations, recommendations, routing, and ticketed activations.
• Learn. Live intelligence. Every conversation logs intent, surfaces gaps in the knowledge base, and feeds an anonymised data asset saleable back to tenants and brands.

Every other solution in this market solves one of those three problems. Signage companies sell advertising. AI character firms sell engagement. Analytics vendors sell data. None closes the loop. That gap is the category, and it is also the commercial opportunity: a credible plan needs to be earning across all three modes, not just the easiest one. We’ve written elsewhere about why this matters as the layer above indoor wayfinding, and why a wayfinding alone wont drive mall visitor engagement.

The seven revenue streams

Not every centre will unlock all seven, and not every stream is worth the effort. But a credible commercial plan for an AI character should consider each one explicitly, decide in or out, and price the decision.

Advertise mode

1. Brand sponsorship of the character itself

The cleanest stream, and the natural first step. A brand pays an annual fee to be associated with the character: naming rights (“Brought to you by …”), a co-branded skin, a sponsored seasonal personality, or category exclusivity (e.g. only one telco partner).

• Who pays: national brands looking for always-on presence in a high-dwell environment. CPG and telco are the likely first movers.
• Pricing model: annual flat fee, often with a tiered structure (title sponsor vs. category partner).
• Margin profile: highest of any stream. Once the character exists, an additional sponsor is almost pure margin.
• Watch-out: over-sponsoring kills the character’s credibility with visitors. One title partner plus two or three category partners is usually the ceiling.

This is shopping centre advertising or shopping mall sponsorship, depending on your market but on inventory that did not exist before. It is also the most visible expression of the AI concierge category: a piece of media that can hold a conversation rather than a 15-second loop.

2. Promotional campaign hosting

Tenants and brands buy limited-time activations through the platform: a product launch, a sale, a loyalty push, a back-to-school campaign. Venue Intelligence carries the campaign for two to six weeks across all touchpoints: idle screen states, in-character conversations, branded experiences.

• Who pays: tenants, plus brand marketing teams running national campaigns who want a physical anchor.
• Pricing model: flat campaign fee, often with a media-value rationale (impressions × dwell × engagement compared to OOH benchmarks). Reference point: Quividi’s industry-standard attention research puts average viewer attention on a DOOH screen at roughly 4.6 seconds before the eye moves away. A conversation runs orders of magnitude longer.
• Margin profile: good, but lumpy. This is a calendar business. You sell the Christmas slot once.
• Watch-out: inventory management. Don’t oversell the calendar. Campaign fatigue degrades every other stream.

Engage mode: live retail customer engagement

3. Sponsored conversations and product placement

Where stream 1 sells the character, stream 3 sells what the character says. Brands pay to be surfaced inside conversations: recommended when a visitor asks for a gift, mentioned when a child asks about toys, suggested when someone asks where to eat.

• Who pays: tenants inside the centre, plus relevant non-tenant brands (a coffee brand sold in a supermarket on-site, an FMCG launching a campaign).
• Pricing model: CPM (per thousand surfaced impressions) or CPE (per engaged conversation). Operators looking for a stronger commercial story should move to CPE early.
• Margin profile: high, with a content production overhead.
• Watch-out: disclosure. Sponsored recommendations need to be clearly labelled, both for regulatory reasons and because visitors notice manipulation faster than marketers think.

4. Tenant lead generation and footfall routing

Discovery inside a mall is a real problem: Mappedin’s 2026 Venue Experience Report found 53% of visitors hit navigation problems at large venues, and 24% lose time looking for specific stores instead of engaging with the venue. The AI character is the only piece of centre infrastructure that can direct a visitor to a specific store and measure whether they arrived. It is, in effect, a virtual concierge with a performance media layer built in.

• Who pays: tenants, especially those in lower-traffic locations within the centre, new openings, and pop-ups.
• Pricing model: per qualified walk-in, validated by a follow-up signal (QR scan at the store, voucher redemption, beacon hand-off). Some centres will run it as a monthly retainer with a footfall guarantee.
• Margin profile: healthy, but operationally heavier than streams 1 to 3. You need a clean attribution method or tenants will dispute the invoice.
• Watch-out: this stream changes your relationship with tenants. You are now selling them performance media on top of rent. Price it so it complements the lease, not so it competes with it.

5. Direct commerce and affiliate

The character can close transactions: a recommendation that ends in a click-and-collect order, a QR handoff to a tenant’s checkout, a referral to a delivery partner. Centres take a share.

• Who pays: tenants and e-commerce partners, on a revenue-share basis.
• Pricing model: percentage of sale, usually 3 to 10% depending on category and who fulfils.
• Margin profile: variable. Strong in high-margin categories (beauty, electronics), weak in groceries.
• Watch-out: this stream is real but rarely the headline. It is likely to be the third or fourth stream a centre unlocks, because it requires deeper integration with tenant systems.

6. Experiential and ticketed activations

The character becomes a paid attraction in its own right: ticketed meet-and-greets, kids’ club memberships, partnered IP events (a film tie-in, a sports moment), seasonal experiences with paid entry.

• Who pays: visitors directly, with sponsor stacking on top.
• Pricing model: ticket revenue plus event sponsorship plus F&B uplift attributable to the event window.
• Margin profile: strong on peak events, but staffing and safeguarding are real costs.
• Watch-out: this is the stream most often overestimated in pitch decks. It works brilliantly two to four times a year. It does not work weekly.

Learn mode

7. First-party data and insight licensing

The conversations the character has are, in aggregate, an unusually rich intent signal, materially richer than aggregate footfall or WiFi data because they capture what visitors want, not just where they went. What visitors ask about, what they’re shopping for, what they reject, how that varies by day, weather, and demographic. First-party data and closed-loop attribution are the primary fuel behind retail media’s run to roughly one-fifth of all worldwide digital ad spend. Done properly, this becomes a mall analytics product the centre owns outright and that data has buyers, provided you treat consent and anonymisation properly.

• Who pays: tenants (as a value-add bundled with their lease or sold as a subscription), category brands, and occasionally market research firms.
• Pricing model: subscription, tiered by depth (aggregated dashboards vs. raw category-level intent).
• Margin profile: software-like once the pipeline is built. The first dollar is expensive. The hundredth is nearly free.
• Watch-out: privacy and contract. If your vendor contract does not give you outright ownership of conversational data, this stream is not yours to sell. Renegotiate before you build it.

Learn mode also earns inside the centre, not just outside it. The same data that licences to brands tells the operator which tenants are over-asked-for and under-rented, which categories the centre is missing, and which knowledge gaps the platform itself needs to close. That is what “live intelligence” means in practice.

Stacking the streams: the realistic mix

No centre runs all seven at full intensity. A credible year-one plan for a mid-sized regional centre should look roughly like this:

Stream	Mode	Share of revenue	Share of margin
1. Character sponsorship	Advertise	35%	45%
3. Sponsored conversations	Engage	20%	20%
2. Campaign hosting	Advertise	20%	15%
6. Experiential events	Engage	15%	10%
4. Footfall routing	Engage	5%	5%
7. Data licensing	Learn	3%	3%
5. Direct commerce	Engage	2%	2%

Illustrative. Treat as a shape, not a forecast.

Two things to notice. First, Advertise-mode streams should dominate both revenue and margin in year one. They are easier to sell, easier to deliver, and they earn before a single visitor walks in. Second, the streams with the highest long-term ceiling sit in Engage and Learn (4, 5, 7) and they start small. Centres that get serious results in three years will be the ones that invest in those streams while the Advertise streams pay the bills.

The expected order to unlock them in: 1 → 2 → 3 → 6 → 4 → 7 → 5. If a vendor proposes starting with 7 or 5, ask why. Those are downstream of the conversational volume the earlier streams generate.

What it costs to earn it

1. “For a single character deployed across ten existing display locations in a property with credible advertising potential, the model typically breaks even on the operator’s investment within two quarters. The character runs on infrastructure the mall already owns. The revenue stack in this piece does the rest.”
2. “For operators who would rather not carry the deployment risk at all, the commercial can be structured for shared upside: the vendor takes a position on the revenue, the operator pays from earnings, and the deal is cash-positive from day one. The model bends to the appetite. What does not bend is the underlying economics: the build is not the operator’s problem any more.”

Three questions for the asset manager

Before you sign anything, put these to the vendor or to your internal team:

1. Who owns the audience data? If the answer is anything other than “you do, outright, with the right to license,” the Learn-mode streams are not yours. The contract is the commercial model.
2. How is sponsor inventory priced and capped? A character with no inventory ceiling becomes a billboard and loses visitor trust, and with it, every other stream. Get the cap in writing.
3. Does the platform actually run all three modes, or just one? Signage companies sell Advertise. AI character firms sell Engage. Analytics vendors sell Learn. If you buy three separate systems you get three separate invoices and no closed loop. Decide whether you’re buying a platform or a point solution. (Our own side-by-side comparison walks through the trade-offs.)

Answer those three cleanly and the rest of the commercial model writes itself.

The thesis, one line

An AI character is not a marketing toy with a P&L attached. It is a P&L instrument with a marketing surface. Centres that run it across all three modes (Advertise, Engage, Learn) unlock seven distinct revenue streams against assets they already own. Other vendors solve one mode each. The category opens up when one platform closes the loop.

That is what venue intelligence looks like on the income statement.

Frequently asked questions

Q. How do shopping centres make money from an AI character?

Through seven distinct revenue streams grouped across three modes. Advertise streams (character sponsorship, campaign hosting) earn before anyone arrives. Engage streams (sponsored conversations, footfall routing, direct commerce, experiential activations) earn during the visit. Learn streams (first-party data licensing) earn after, on repeat. Year-one revenue should weight ~70-75% to Advertise; year-three weight tilts towards Engage and Learn as conversational volume builds.

Q. What is a digital concierge for a shopping centre?

A digital concierge is an AI character that greets, advises and routes visitors inside a shopping centre or mall, acting as a virtual concierge across opening hours and languages a human team cannot cover. In the Venue Intelligence model, the digital concierge is the visible front end of a system that simultaneously advertises when idle, engages when approached, and learns from every conversation. It is the layer above indoor wayfinding: it does not just answer “where” it answers “why,” recommends, and measures whether the visitor followed through.

Q. What is Venue Intelligence?

Venue Intelligence is the platform category Performit Live defines: a system that advertises when idle, converses when approached, and learns from every interaction. Where signage sells media, AI character firms sell engagement, and analytics vendors sell data, Venue Intelligence is the only platform that runs all three modes through a single branded character and closes the loop between them.

Q. How is an AI character different from digital signage or indoor wayfinding?

Digital signage broadcasts. Wayfinding answers “where.” An AI character does both, plus conversational discovery, recommendation, intent capture, and live measurement of whether the visitor actually went where it sent them. It is the layer above indoor mapping, not a replacement for it. See Maps vs Missions for the longer argument.

Q. Who owns the audience data the character collects?

It depends entirely on the vendor contract. If the contract does not name the centre as outright owner of conversational data with the right to license it, the Learn-mode streams are not yours to sell. This is the single most important commercial clause in the procurement. Renegotiate before you build.

Next in Venue Intelligence, 04: Operating Model. Who runs the character day-to-day, how the team is structured, and why the staffing question quietly decides whether the commercial model in this post actually lands.

Ready to pressure-test your own readiness? Run the AI Character Readiness Scorecard or book a discovery call.

Related: In-store Retail Media