Tiger Brokers puts 3 AI agents to work across 260-plus screens
A Singapore OOH pilot links AI buyer, seller and creative agents while retaining approval controls and leaving performance gains unproven.
Tiger Brokers has put three AI agents into an actual out-of-home advertising workflow, working with Adzymic and Stellar Ace on a Singapore campaign carried across more than 260 digital screens. The partners disclosed the completed pilot on November 12, following activity from October 21 to November 3 to promote the brokerage's Tiger Trade Experience 2026.
Rather than asking AI to create one more ad, the companies connected inventory discovery, buying proposals and screen-specific creative preparation. In their official account of the pilot, they describe it as an early live demonstration, not a fully autonomous purchasing system. The useful question for marketing teams is where coordination can safely be delegated, and where responsibility still belongs to people.
Table of contents
Jump to each section:
- What the three agents actually did
- Human approval remains part of the buy
- Why this is different from programmatic OOH
- From inventory access to live execution
- What marketers should know about agentic OOH
What the three agents actually did
The campaign divided the advertising process among a buyer agent, a sales agent and a creative agent. Each worked on a different operational problem, using the advertiser's requirements and the media owner's available inventory as inputs.
The buyer agent examined Stellar Ace's screen inventory and proposed a media plan. The sales agent assessed that proposal from the media owner's side. The creative agent then interpreted Tiger Brokers' brand guidelines and individual screen specifications to prepare materials suitable for different placements.
Three AI agents and more than 260 digital screens. Those were the reported workflow components and Singapore campaign scale, according to the campaign coverage.
That division matters because a campaign does not move directly from a brief to an impression. It moves through a chain of interpretation: what the advertiser wants, which locations and formats are suitable, what the seller will accept, and what creative can actually run. An agent can speed up a handoff, but the accuracy of every handoff becomes part of the advertising product.
In this pilot, coordination was the innovation. Faster creative production alone would not have tested whether buying intent could travel from one system to another without losing its commercial meaning.
Human approval remains part of the buy
Tiger Brokers reviewed and approved the proposed media plan before Stellar Ace accepted the media purchase. Both sides also conducted creative checks and handled the operational steps needed to make the campaign live. That is not a minor footnote. It is the boundary between software-assisted preparation and autonomous spending.
The common assumption is that agentic media buying matters when people are removed from the process. This pilot suggests a more practical starting point: remove repetitive coordination while keeping human authority over budgets, placements and brand suitability.
Ernest Teh, Tiger Brokers Singapore's marketing director, described the pilot as an encouraging experience in simplifying media planning. The companies have not disclosed a measured reduction in planning hours, production costs or campaign conversion outcomes. Their statement that this was Asia's first live agentic OOH campaign should therefore be read as a partner claim, not independent proof of market leadership.
An approval layer is valuable only when people can understand what they are approving. For an agency, the next test is whether a buyer agent can explain why certain screens were selected, what alternatives were available and where uncertainty remains. Without that traceability, faster recommendations can create slower reviews.
Why this is different from programmatic OOH
Digital out-of-home already uses automation. Programmatic platforms can activate campaigns across digital screens using inventory and targeting rules. Agentic systems add a layer that interprets less structured campaign instructions and coordinates tasks between the buyer, seller and creative-production functions.
The distinction is important. Programmatic automation is largely about executing a defined transaction; the Tiger Brokers experiment concerns how a transaction is proposed, evaluated and made ready for execution. The new layer does not make existing ad-serving infrastructure irrelevant.
That also raises a competitive question. A media owner whose inventory is well described for software agents may become easier to discover, but visibility does not automatically prove that its screens are the best choice. Location, time of day, audience context and commercial restrictions still require explicit inputs.
For brands, the value proposition should be tested against concrete outcomes: reduced planning effort, fewer revisions, faster approvals or better campaign performance. None of those improvements can be quantified from the disclosed Tiger Brokers results so far.
From inventory access to live execution
Adzymic had already introduced an AgenX Sales Agent in October, allowing compatible buyer agents to discover and evaluate Singapore publisher inventory, including Stellar Ace's digital out-of-home placements. The system used the Ad Context Protocol and supported other agent-communication standards. That earlier launch made supply accessible; the new pilot provides a named advertiser and a completed campaign using connected agents.
This is a meaningful distinction between product availability and operational evidence. A tool can make inventory machine-readable without proving that a live creative and media approval workflow will hold together. The Tiger Brokers experiment narrows that gap, while leaving broader reliability and performance untested.
Kenniess Wong, Adzymic's co-founder and executive director, described the goal as "intelligent agents working together across the advertising value chain". The remark captures both the opportunity and the constraint. Software may connect more steps, but brand teams still need to decide what acceptable outcomes look like.
Even with a real commercial deployment, one pilot does not establish that agent-based buying will work at the same quality across all OOH networks, formats or markets. The next evidence should concern repeatability, not just whether the first execution was possible.
What marketers should know about agentic OOH
For marketers considering agent-driven planning, the Tiger Brokers pilot offers a useful model: delegate the work of assembling options, not the ultimate responsibility for spending.
Brief quality becomes a system requirement. A buyer agent needs clear objectives, geographic constraints, budgets and brand exclusions to propose a credible plan. An ambiguous brief is not improved merely because AI interprets it quickly.
Seller-side information matters. Screens must be described in ways agents can compare without flattening the real differences among locations and formats. Stellar Ace's role illustrates why the media owner's data and approval rules are central.
Creative checks still belong in the workflow. A file that fits technical screen specifications is not necessarily appropriate for the brand or setting. Tiger Brokers and Stellar Ace retained human review rather than assuming generation equals approval.
Operational metrics should precede marketing claims. Teams can measure time to a usable plan, revision cycles, approval exceptions and final activation accuracy. Audience or sales lift needs separate campaign evidence.
The broader shift is not simply AI entering outdoor advertising. It is the possibility of advertising systems that exchange intent instead of passing only files, insertion orders and availability lists.
That makes accountability more consequential, not less. The teams that gain from agentic media may be those that can specify their requirements clearly enough for software to act on them, and preserve enough visibility for people to say no when needed.
