Most procurement processes for interactive technology follow a predictable pattern. A vendor arrives, delivers a breathtaking demo of a digital character responding to queries in real time, and the room is captivated by the technical “magic.” However, as many CMOs and procurement leads have discovered, the distance between a successful demo and a sustainable, three-year deployment is vast.
According to research by Bain & Company, 86% of store technology transformations fail to meet their original expectations. This failure is almost never because the technology itself was faulty. Instead, it is usually because the wrong questions were asked at the wrong stage.
When evaluating an AI character vendor, the focus must shift from what the technology can do in a controlled environment to how it will survive and thrive in the messy reality of a live venue.
If you are new to Physical AI, the discipline of deploying AI characters into shared physical environments, our category-defining guide covers the foundational framework. This blog builds on that foundation with a practical procurement lens.
The right vendor does not just sell you a digital asset. They help you design the operation that makes it work. To help you separate a long-term partnership from an expensive mistake, we have compiled the essential questions every buyer should ask during the selection process.
The AI Character Procurement Checklist
We have grouped these questions into four critical buckets that address the lifecycle of a digital character beyond the initial launch.
Governance and Accountability
Who owns operational performance once we go live: your team or ours?
Good Answer: A clear definition of shared responsibilities, outlining which technical uptimes are the vendor’s and which on-site tasks belong to your team.
Red Flag: “We handle everything remotely.” No public-facing technology is truly autonomous. Someone on-site always needs a basic understanding of the deployment and the authority to act when something goes wrong.
What happens when the character says something off-brand? Show me your incident response protocol.
Good Answer: A detailed explanation of how the system uses pre-defined content boundaries, often called a “no list”, combined with human-in-the-loop escalation so that a trained operator can intervene in real time when the AI encounters a situation outside its guardrails.
Red Flag: “Our AI is highly advanced and won’t make mistakes.” All AI has the potential to drift. A lack of an escalation plan is a major risk and a sign the vendor has not operated in high-footfall public environments.
What data do you capture, where does it sit, and who has access?
Good Answer: A transparent data architecture that prioritises compliance with applicable data protection regulations such as GDPR in the UK and EU, PDPL in Saudi Arabia, or the relevant framework in your region, and explains how first-party insights are delivered back to your business.
Red Flag: Vague answers about “cloud storage” without specific geographic, jurisdictional, or security details. If the vendor cannot tell you where the data physically resides, they have not thought through the governance layer.
Operational Readiness
How do you support our front-of-house team: training, playbooks, and escalations?
Good Answer: Provision of a staff training module and a clear “playbook” that helps your physical team work alongside the digital character. This should cover day-to-day operations, common visitor questions about the character, and what to do when something goes wrong.
Red Flag: Assuming your floor staff need no information because the technology is “self-service.” Any deployment that operates in a retail environment or public venue requires on-site humans who understand what the character does and does not do.
What does your deployment support look like at 9pm on a Saturday?
Good Answer: A support tier that matches your venue’s operational hours, not just standard 9-to-5 office hours. If the character is live during peak trading, support must be live during peak trading.
Red Flag: Standard office-hour support for a retail or hospitality environment that peaks on evenings and weekends. This is a clear signal the vendor’s operational model was built for corporate software, not public-facing deployments.
What is included in ongoing content and behavioural updates, and what costs extra?
Good Answer: A transparent subscription model that includes regular knowledge base refreshes to prevent content decay, whether that is weekly for retail promotions or seasonally for event spaces.
Red Flag: Low upfront costs followed by heavy “professional service fees” for every minor script change. Content decay is one of the leading causes of deployment failure; if updates are expensive, they will not happen often enough.
Commercial and Measurement
How do you measure commercial outcomes, and will you commit to metrics contractually?
Good Answer: A focus on what the character learns from visitors: the questions they ask, the topics they raise unprompted, and the patterns that emerge across thousands of conversations. A well-instrumented Physical AI deployment gives brands and venue operators something no survey or footfall counter can: unfiltered, first-party conversational insight at scale. What are visitors actually looking for? What questions come up repeatedly that your team never anticipated? Which topics drive the longest interactions? The right vendor will structure this data into actionable reporting and tie commercial metrics to genuine interaction quality, not vanity numbers.
Red Flag: Measuring success solely on “impressions” or “total uptime.” These metrics tell you the screen was on, not that it delivered value. If the vendor cannot explain how conversational data flows back to your brand, marketing, or operations teams as insight, they are treating the character as a display, not an intelligence asset.
What is the realistic total cost of ownership over three years?
Good Answer: A comprehensive breakdown that accounts for the full operational picture, including:
- Hardware: initial installation, maintenance cycles, and refresh or replacement schedules.
- Software: platform licensing, AI model access, and any per-interaction usage fees.
- Content: ongoing knowledge base updates, seasonal script changes, and character behavioural tuning.
- Support: tiered support contracts, out-of-hours coverage, and incident response.
- Human oversight: the cost of trained operators, on-site staff training, and any managed service for human-in-the-loop escalation.
- Integration: connecting the deployment to your existing CRM, loyalty, analytics, or footfall systems.
Red Flag: Only quoting the “software licence” cost while ignoring the integration, content, and human oversight required to keep the deployment running. If the vendor’s quote covers less than half the items above, you are looking at a fraction of the true cost.
Technology and Future-Proofing
How do you update underlying AI models without breaking our character’s personality?
Good Answer: A decoupled architecture where the character design (voice, personality, visual identity, behavioural rules) remains consistent even when the underlying intelligence layer is upgraded to a newer model.
Red Flag: “Updates happen automatically.” This suggests your character’s voice or mannerisms could change overnight without warning. Any model update should go through a staging environment and require your sign-off before it reaches the live deployment.
What happens to our deployment if your company is acquired or changes direction?
Good Answer: Contingency plans, data portability guarantees, and clear contractual protection for your intellectual property, including the character design, knowledge base content, and any interaction data generated during the deployment.
Red Flag: Deflection or a lack of long-term business stability evidence. If the vendor cannot answer this question directly, consider what happens to your investment if they pivot to a different market or shut down.
Three Red Flags Most Buyers Miss
When you are in the room with a potential AI character vendor, listen for these three warning signs. They often signal that the vendor is optimising for a “close” rather than your long-term success.
- “We will figure that out after launch.” Governance, measurement, and escalation are not post-launch problems. If a vendor cannot answer these before you sign, they do not have a proven process for live environments.
- “Our platform handles everything; you won’t need to manage it.” This is a myth. No platform runs itself in a high-traffic public space. If the vendor is not asking who on your team will own the deployment, they are selling a product, not a partnership.
- “Here is what it can do.” If the sales conversation is 80% about capability and 20% about operations, you are buying a pilot that will never scale. The demo is the easy part. The operation is the hard part.
The Questions the Vendor Should Be Asking You
A high-quality AI character vendor does not just answer questions, they ask them. The depth of their inquiry reveals whether they truly understand the complexities of your venue. A good vendor will want to know:
- What does commercial success look like for you, beyond just the technical operation?
- Who in your organisation will “own” this character after the launch team moves on?
- What is your existing escalation path for physical incidents outside of business hours?
- Which existing systems (CRM, loyalty, or analytics) does this need to feed into?
- What does your front-of-house team currently look like, and how will they be introduced to their new digital colleague?
If none of these questions come up in your early conversations, it is a sign that the vendor is focused on a quick installation rather than a durable operational asset.
Buy the Operation, Not the Product
The best AI character deployments are rarely the ones with the flashiest technology. They are the ones where the vendor helped the buyer think operationally long before a single screen was installed in the venue.
The right AI character vendor does not just sell you a digital character. They help you design the operation that makes it work. By asking the above ten questions, you move the conversation away from speculative tech and toward a robust commercial strategy.
Before you make your next procurement decision, it is vital to understand your own operational readiness. We have built a free diagnostic tool that evaluates your project across the five layers of Physical AI and identifies the gaps most likely to derail a deployment before you ever engage with a vendor.
Take the AI Character Readiness Scorecard →
AI Character Vendor Frequently Asked Questions
Why should procurement focus on operations rather than technology?
Technology is increasingly commoditised, but successful deployment in a live venue remains an operational challenge. A vendor might have a great visual engine, but if they lack the governance and support frameworks to handle public interaction during peak hours, the project will likely fail to scale or meet commercial objectives.
What is a red flag when discussing AI character data?
Any lack of transparency regarding where data is stored and who has access to it. For venue operators in any jurisdiction, ensuring compliance with applicable data protection regulations (whether GDPR, PDPL, or regional equivalents) is a non-negotiable governance requirement that must be addressed before signing a contract.
How often should an AI character’s knowledge base be updated?
To avoid content decay, the knowledge base should be treated as a living document. A good vendor will provide a system for regular updates, whether that is weekly for retail promotions or seasonally for event spaces, ensuring the character remains relevant and helpful to guests.
What role does human-in-the-loop oversight play in vendor accountability?
The most accountable vendors build human oversight into their architecture, not as a fallback, but as a core layer. When evaluating a vendor, ask how their system handles situations the AI was not trained for. A platform that combines automated guardrails with the ability for a trained operator to step in during high-stakes moments offers the highest level of accountability for brand reputation.
Is your organisation ready to deploy an AI character?
Take the free AI Character Readiness Scorecard. 2 minutes. Personalised PDF report.