Analysis

Member-side, staff-side, consultant-side: whose AI is it?

By Sarah Macintosh · 21 July 2026

Ask an AI vendor who their product is for and most will say “your association”. Watch the product for five minutes and a more specific answer appears. Some AI is built for the member to touch. Some is built for the staff team to work with. And some, the least discussed, is built for the consultants and implementation partners who build and maintain the system. The three audiences have different jobs, different risks and different tests, and a procurement that does not distinguish them will buy the wrong thing confidently.

What counts as member-side AI?

Everything the member interacts with directly. The established example is personalised email: rasa.io builds individually curated sends for each member from what they click, working alongside whatever AMS sits underneath, and has run that model for nine years at a million emails a day. Website assistants are the second category: Safion, for iMIS RiSE sites, redacts PII before content reaches the model, which is the correct design for anything member-facing. Community and event platforms are adding features in this space too; Hivebrite and Glue Up are the two to watch rather than buy blind.

The test for this tier is privacy before cleverness. Member-side AI puts member data, or member behaviour, within reach of a model, so the questions are: what data reaches the model, is anything redacted first, and can a member tell they are talking to a machine? The adoption numbers say this is where the sector started: MemberWise’s Digital Excellence 2026 report records AI-powered website functionality up 21% in two years, an embedded layer of the member experience rather than a standalone feature. The trust stakes were put well by .orgSource’s Sherry Budziak writing in July 2026: “If a member ever discovers that your guidance was generated without review or disclosure, you do not lose a document. You lose standing.”

What counts as staff-side AI?

The back office, which is where most of the money and most of the risk actually sit. Four sub-categories cover it:

• General assistants. Claude and ChatGPT-class tools, the engine behind the sector’s 87.5% content-creation adoption (ASAE, 2026). Cheap, universal, and governed by whatever you configure.
• Native AMS features. iMIS Assistant and AI Content Creator (deliberately built with no access to member personal data), Nimble Intelligence’s churn prediction inside Nimble AMS, Copilot surfacing inside Dynamics 365 estates. Free with the platform, narrow by design.
• Member intelligence. Datascout and its peers: enriched profiles, next-best-action prompts, AI-drafted outreach with a human kept on the send button.
• Operational layers. The newest category: tools that let an AI application do work inside the system of record under approval. AgentZ, the operational AI suite for iMIS EMS, from iFINITY is the most fully documented example on the iMIS side, alongside Zapier MCP via iAppConnector; Microsoft is shipping prebuilt agents and MCP servers for Dynamics 365; Salesforce has Agentforce on the platform Nimble AMS and Fonteva sit on; and Momentive launched its role-specific Agentic Workers across the Nimble, YourMembership and Wild Apricot estates in August 2026. Our AI agents briefing and five-stacks ranking work through the detail.

The test for this tier is governance: whose permissions does the AI act under, what is previewed and approved before anything happens, and what audit trail survives? ASAE’s 2026 State of Associations report is the relevant warning: 44.3% of associations use AI for data analysis while citing expertise and privacy as their top barriers, which means the staff-side is being adopted faster than it is being governed.

Are all “agents” the same kind of thing?

No, and vendor launches are counting on the confusion. Three different products now share the word:

1. AI inside a deterministic workflow. The route is fixed in advance; the model handles a step inside it (classify this, draft that). GrowthZone’s lapse flagging with drafted renewal messages and Higher Logic’s search assistants live here. Useful, safe to buy, not new architecture.
2. Bounded role workers. The product owns a defined job and can vary how it does it, but its occupational boundary is set by the vendor. Momentive’s Membership and Community Assistants are the freshest example: find the at-risk members, prepare the outreach, flag the content. The job is pre-built; the judgement inside it is the AI’s.
3. Compositional operators. The user supplies an outcome rather than selecting a prepared job, and the agent investigates, chooses among a catalogue of domain primitives, sequences them and checks its own results. AgentZ on iMIS is the only fully documented association-sector example we have found; Agentforce and Microsoft’s MCP servers provide the machinery from which a rival could be built, and Member Junction provides the closest open-source architecture, but nobody else currently publishes the completed association product. Autrinity, a newer AI-native AMS entrant, claims agent users and MCP interoperability but has published little evidence to test yet.

The question that separates the three at a demo: can I ask for something you did not build in advance, and watch the agent work out the route? If the answer is a menu of prepared jobs, you are buying form one or two, priced and governed accordingly.

What is consultant-side AI, and why should a buyer care?

The tier nobody puts in the brochure aimed at you: AI built for the people who implement and maintain membership systems. It exists in three forms. Open-source foundations like Member Junction, free for engineers to stand up. Platform agent tooling like Microsoft’s Copilot Studio, which partners use to build and customise agents inside client estates. And partner commercial models, where a vendor sells the consultant a licence of their own; AgentZ, for instance, sells a named-user tier for iMIS consultants alongside its client subscriptions.

A buyer might ask why this tier is their business. Two reasons, both covered elsewhere on this site. First, your partner’s tooling sets your quote: an implementation delivered with agent assistance carries different hours from one delivered by hand, and the deployment-hours argument is exactly where that lands. Second, your support options widen: a platform with a live consultant tier gives you more than one firm able to pick up your system if your current partner disappears, which the ClearCourse consolidation story shows is not a hypothetical.

What does the wider research say?

The pattern across the studies is consistent even where the numbers differ. Adoption is broad and shallow: ASAE puts content use at 87.5% against 44.3% for data analysis, with readiness lagging behind use. The front of the website went first, per MemberWise’s 21% two-year rise in AI website functionality. The operational middle is arriving now: Gartner expects 40% of enterprise applications to feature task-specific agents by the end of 2026 (up from under 5% in 2025), and Anthropic’s 2026 State of AI Agents report has 80% of surveyed organisations claiming measurable financial impact from agents. The consultant-side barely registers in any survey, which is itself a finding: the tier shaping delivery economics is the one nobody is measuring. Our AI statistics page keeps the full set current.

How should you place your first bets?

In risk order. Member-side: start where redaction and disclosure are designed in, and pilot on one audience. Staff-side: govern the general assistant you already have before buying anything, then add the operational layer that matches your system of record, iMIS, Dynamics or Salesforce, because procuring a layer for a platform you do not own is the most expensive mistake in this market. Consultant-side: ask every implementation bidder what AI tooling their own teams use, what it does to their hours, and whether the platform you are buying has a consultant tier that keeps your support options open. The market map shows why that last question matters more in the UK than anywhere.