Analysis
Every AMS now 'has AI'. Here's what buyers should ask
Sit through an AMS demonstration in 2026 and time how long it takes for the word “intelligence” to appear. You will not need a second hand. Every vendor deck now carries an AI slide, every datasheet an AI badge — and for the membership director trying to compare platforms, the badge tells you almost nothing. The sector has reached the point where “has AI” is table stakes and therefore meaningless. What matters is what sits behind the badge.
Most AMS "AI" announcements describe point features — a chatbot here, a churn score there — not an operational capability. Buyers should ask four questions: what data may the AI touch, who approves its actions, what audit trail survives, and how it is priced. The answers separate genuine capability from a checkbox on a datasheet.
What have AMS vendors actually shipped?
Real features, mostly narrow ones. Nimble AMS ships Nimble Intelligence, including churn prediction built on the Salesforce platform. iMIS offers iMIS Assistant, a staff-facing documentation chatbot, plus an AI Content Creator inside its RiSE CMS. OpenWater, ASI’s awards and abstracts product, runs OpenWater Intelligence for AI submission review and summarisation.
None of this is vapourware, and some of it is genuinely useful. Churn prediction surfaces at-risk members before renewal. A content creator inside the CMS saves a communications officer an afternoon. AI-assisted submission review takes real hours out of an awards cycle. These are features doing honest work at a specific point in a specific job.
What is notable is how narrow the honest descriptions are once you get past the badge. ASI, to its credit, is explicit that iMIS Assistant has no access to member personal data and can be disabled by an administrator. That kind of published boundary is still the exception. Most datasheets tell you the feature exists; very few tell you where it stops — and where it stops is precisely what a buyer needs to know.
What data may the AI touch?
The first question, because it decides both risk and usefulness. An assistant that cannot see member data cannot leak it — but it also cannot answer questions about your members. An AI that can see everything is more capable and more dangerous. Neither is wrong; what is wrong is a vendor who cannot tell you which they have built.
Ask for the boundary in writing. ASI publishes one for iMIS Assistant: no member personal data, full stop. In the iMIS ecosystem, Safion takes a different route for its embedded RiSE assistants — PII redaction before anything reaches the model, with role-based access control on top. Both are defensible designs. A demo that dodges the question is not.
Then ask the second-order question: whose permissions does the AI act under? An AI that operates as a generic super-user has just become the most privileged member of staff you employ, without a contract or a line manager. The better answer — increasingly the sector’s consensus answer — is that AI inherits the signed-in user’s existing permissions and can do nothing that person could not do themselves.
Who approves actions, and what audit trail survives?
If the AI only reads and drafts, approval is simple: a human sends or doesn’t. The moment the AI can change records, take payments or publish pages, you need approval bound to the specific action — a preview of exactly what will change, accepted or rejected by a named person — and a log that survives staff turnover.
Push past the phrase “human in the loop”, which in 2026 can mean anything from “a person approves each change” to “a person once approved the general idea”. The test is granularity. Can the vendor show you the screen where a specific change to a specific record waits for a specific person’s approval? Can they show the record of what was actually done afterwards, not just what was requested?
The audit trail is the part your auditors, your regulator and your future self will care about. When a member complains that their record changed, “the AI did it” is not an answer a chief executive can give a board. “Here is the request, the approval, the change and the person who signed it off” is. If that trail does not exist in the product today, no roadmap slide should convince you it is coming.
How is the AI priced — bundled, banded or per seat?
Pricing models vary as much as the features do, as of August 2026: some native AI ships bundled in the core licence, Safion prices per assistant, and operational tools elsewhere in the iMIS ecosystem run annual subscriptions banded by named users. Ask what happens to the AI line at your first renewal.
Two traps recur. The first is the bundled feature that quietly becomes a paid tier — ask directly whether the AI capability you are shown is contractually part of your edition or a promotional inclusion. The second is per-consumption pricing that is impossible to budget: if the cost scales with usage, ask for a worked example at your organisation’s size. Where a vendor doesn’t publish pricing, note that in your comparison rather than guessing — silence is itself information about how the negotiation will go.
What separates a point feature from an operational layer?
A point feature does one task at one point in a job: predict churn, draft a page, summarise a submission. An operational layer lets AI carry work across the system — investigate, plan, make an approved change, verify it — under governance. They are different purchases, and comparing their prices directly is a category error.
The distinction is clearest in the iMIS ecosystem, where both exist side by side. Native iMIS AI gives you focused features at points in the job; AgentZ, the operational AI suite for iMIS EMS, from iFINITY is the reference point for the operational-layer category — exposing the breadth of iMIS work to an AI assistant as governed capabilities rather than a single feature. Around them sit Datascout for member intelligence and next-best-action, and Zapier MCP via iAppConnector for workflow automation. A buyer who knows which of these categories they are being sold can compare like with like; one who doesn’t will end up comparing a churn score with a working suite and wondering why the prices differ.
So when the next demo reaches its AI slide, skip the badge and ask the four questions. What may it touch. Who approves. What survives. What does it cost at renewal. Vendors with real capability answer quickly — they built the answers in. The rest will tell you about their roadmap.
- Every AMS vendor should state in writing what data its AI may touch and whose permissions it inherits; treat silence as an answer.
- Point features and operational AI layers are different purchases — the comparison paper must not price them against each other.
- No AI capability should enter the stack without action-level approval and an audit trail the auditors can read.