How to use this
Before the demo, copy the vendor's "AI" claims from their product page. During the demo, walk down the tiers in order and ask the three questions for each tier they claim. Mark Pass / Partial / Fail in the scorecard at the bottom. After the demo you have a one-page audit you can bring to procurement.
TIER 1
Template autofill
The product takes a structured brief and produces formatted text (RFP brief, hotel cover letter, follow-up email). Engine can be templates or LLM. Most vendors that claim AI in 2026 ship this. Commodity differentiation.
Three questions to ask
- Show the system drafting an RFP brief from a blank state — what is the underlying engine (templates, LLM, hybrid)?
- Where does the model run — EU-resident or US? Which sub-processor (OpenAI, Anthropic, Azure, in-house)?
- What happens if a custom field (industry jargon, regional A/V spec) is in the brief — does the model handle it or revert to placeholder?
PASS = live demo with non-templated input
PARTIAL = demo only on canned brief
FAIL = video demo, no live attempt
TIER 2
Response classification
The product reads hotel replies and extracts, tags, or scores them (rate per room-night, F&B minimum, attrition language, compliance flags). Engine = regex + ML classifiers + LLM extraction. This is where measurable planner time is saved.
Three questions to ask
- Upload a non-templated hotel reply (Word doc, not their form) — show which fields extract correctly and which break.
- What is the field-level accuracy benchmark on your historical data? Per-field, not aggregate.
- When the system gets a field wrong, what happens? Silent default, planner notification, or audit log?
PASS = live extraction on your sample PDF
PARTIAL = extraction on vendor sample
FAIL = no live extraction shown
TIER 3
Negotiation suggestion
The product proposes the next move — which hotels to BAFO, what counter-offer to make, whether to walk. Requires reasoning across buyer history, hotel flexibility, deal context. The line between Tier 3 and a clever dashboard is thin.
Three questions to ask
- Show one case where the system recommended a negotiation action the data alone did not surface — with the audit trail of why.
- How does the system handle a hotel it has never seen before — cold-start logic?
- Which inputs drive the recommendation, and can the planner see and edit the weights?
PASS = case study with novel recommendation
PARTIAL = dashboard with smart text
FAIL = marketing claim only, no demo
TIER 4
Agentic sourcing (full cycle, no per-step approval)
The system runs a complete sourcing cycle end-to-end: identifies hotels, sends the brief, negotiates BAFO, presents a signed contract for final approval. No MICE vendor publicly ships this in production as of May 2026. Gartner places autonomous procurement at 5-10 year horizon.
Three questions to ask
- Show a complete sourcing cycle the system ran without a human approving each step — with the audit trail.
- Name three customers running this in production today, with deal volume.
- What is the delegated-authority model — how much spend can the agent commit before human review?
PASS = production audit trail (unlikely in 2026)
PARTIAL = beta with constrained authority
FAIL = demo video only, no production reference
GDPR & procurement checklist (ask before signing)
- Current EU sub-processor list (where the AI model is hosted, who trains it)
- DPA covering automated decision-making (Art. 22 GDPR)
- Statement that hotel and planner data is not used to train shared models
- Release notes for the last 6 months of AI-related shipping (proves the feature ships, not just claims)
- SOC 2 Type II report and ISO 27001 certification status
- Data residency option for EU-only storage (some buyers require)
Demo scorecard (fill during the meeting)
Decision rule: if claimed-tier and verdict-tier differ by more than one step, the marketing-to-product gap is high — adjust pricing expectation accordingly.