How to use this checklist
Most alumni-platform RFPs fail in the same way: they ask dozens of feature questions and almost no structural ones. Features can be demoed; structure cannot. The twelve questions below are the ones we find actually discriminate between vendors. They are grouped into four domains — data stewardship, integration depth, AI roadmap credibility, and portability — and each is phrased so that a weak vendor is forced to be specific rather than encouraging.
Data stewardship
1. Who owns the alumni record, and in what format can we export it? You want a contractual answer, not a marketing one: you own the data; a full export (not a reduced "active members" extract) is available in a documented machine-readable format, on demand, without a services fee. Ask for the export specification during evaluation — the file structure and field dictionary, not a promise that one exists.
2. Where is the data stored, and under which jurisdictions? Hosting region, subprocessors list, and whether residency can be pinned. If the vendor cannot enumerate its subprocessors in the DPA, treat that as the answer.
3. What happens to deleted alumni? Deletion workflows are where data-responsibility claims get tested. Ask: when an alum requests erasure, is deletion propagated to backups, analytics stores, and any AI/embedding indexes? Vendors doing AI-native search over the alumni graph should be able to explain how embeddings are invalidated, and many cannot.
4. Who may train models on our data? Any vendor with an AI feature set now holds this question. The acceptable answer is explicit and contractual: your alumni data is not used to train shared models, or if it is, you have an opt-out that is honoured in practise. "We take privacy seriously" is not an answer.
Integration depth
5. Which HRIS and CRM connectors are native, and which are services projects? Make the vendor list them by name and separate "native connector we maintain" from "we can build that with our services team." Both can be fine; the bill and the maintenance burden are very different. Native connectors survive the vendor's own releases; services-built integrations are yours to babysit forever.
6. What breaks when the other system changes? Ask for one incident, honestly told, where an upstream HRIS API change broke an integration — how it was detected, how long the fix took, who paid. Vendors with mature connector estates have these stories ready and are oddly proud of them. Vendors without them improvise.
7. Is there a real API, and can we see the documentation now? Full CRUD on the alumni record, webhooks for engagement events, and documentation published without an NDA. If docs require a sales contact, the API is thinner than the slide suggests.
AI roadmap credibility
This is where 2026 evaluations get theatre. Market leaders are shipping natural-language directory search, AI career roadmaps, and alumni-to-opportunity matching; everyone else's decks describe the same features in the future tense. Three questions separate shipped from aspirational:
8. Which AI features are live with customers today, at what reference account, and can we use them on our data in a pilot? A demo on vendor-curated data proves the demo. A pilot on a bounded slice of your alumni base proves the feature. Insist on the latter as a paid or credited pilot term. A vendor that will not pilot its AI claims on real data is telling you the demo is the product.
9. Show us the eval. Serious AI-feature vendors maintain a live evaluation set — a versioned battery of queries with known-good answers — and can talk about query accuracy the way database vendors talk about benchmarks. Ask to see how they measure natural-language search quality or matching precision. A vendor without an eval discipline cannot tell a regression from an improvement, and neither will you after go-live.
10. What is the minimum data quality these features need to be honest? AI roadmaps and matching over a stale, undeduplicated alumni base produce confident nonsense. A credible vendor will tell you the preconditions — verified employment history, deduplication, current-role coverage — and often that your data isn't ready yet. The vendor that says "our AI works on any data" will sell you a directory wearing an AI costume.
Portability
11. What would leaving look like at the end of year two? Ask them to walk the exit path: export format, media/attachments, engagement history, the timeline, and the cost. In a consolidating market — and industry research describes this category consolidating around a small number of platform providers — the most likely exit is forced, via the vendor's acquisition or wind-down, not chosen. Write the exit terms into the contract while you have leverage, i.e. before signature.
12. If your category position changed, what would our protections be? Ask for continuity commitments: source-code or data escrow, a definition of "end of life" with minimum notice, and a migration-assistance clause at defined rates. Large enterprise vendors will negotiate these. Vendors who call them unnecessary are planning a future in which you have no protection — and no seat at the consolidation table either.
A closing note
None of these questions require a technical team to ask. They require only the discipline to insist on specific answers — connectors named, formats documented, pilots delivered, protections contracted — and to write down who said what. The vendor that answers all twelve crisply may still not be the right choice. But the vendor that answers four of them and pivots on eight has saved you a much more expensive discovery.