They read your courses, your CPD records and your community, find the thing worth acting on, and draft what to do about it. Then they stop and wait for a person. Running with a pilot association on real member data today.
One suite, one dataset, so an agent can see a renewal, a course and a conversation as the same person instead of three exports. Everything on this page runs in the product today. We are deliberately not listing what it will do next, so if it is described here, it works.
Ask about the pilotAI that does not know your members is just a clever chatbot with a confident tone.
You cannot let software message your members without a trail you would be happy to show a board.
The tools that sound most certain are usually the ones that quietly rounded the awkward numbers away.
Each specialist has a deliberately small job and hands off when a question crosses into someone else’s territory. The line that matters is not what each can do, but what each refuses to do.
Aggie
Your chief of staff
Takes the question you actually asked, works out which specialists it touches, and comes back with one answer instead of five fragments.
What sets it apart: Holds no data tools of its own. Every fact in Aggie's reply came from a specialist it delegated to, so the trail back to the source stays intact.
Behind Aggie
Membership analyst
Known as Lumen
Reasons across joins, renewals, lapses and engagement signals to show you who is drifting and what shape that drift has.
What sets it apart: Will not extrapolate from a single data point. When the sample is too small it says so and asks you to treat the pattern as directional.
Revenue strategist
Known as Scout
Cross-references orders, sponsorship records and event financials to point at the tiers and packages worth a closer look.
What sets it apart: Highlights underperformance without prescribing the fix. Scout tells you which tier is soft; the pricing decision stays with your finance team.
Engagement coach
proposesKnown as Beacon
Reads your community and your courses like a room, and finds the exact lesson or thread where people quietly drop off.
What sets it apart: A pattern-spotter, not a scorekeeper. Two hundred views with thirty replies tells a different story from two thousand views and three, and Beacon will say which one you have.
Operations manager
Known as Atlas
Watches support patterns, admin activity and system signals, and tells you what is drifting before it becomes what is broken.
What sets it apart: Calm escalation over alarm. A weekend spike in admin edits comes with a severity hint and the numbers underneath, so you can disagree with the call.
Compliance specialist
proposesKnown as Sentinel
Handles CPD, code of conduct and audit posture, which are the questions that carry consequences for your members and your regulator.
What sets it apart: Cite it or drop it. Every compliance claim resolves to the row that backs it, and asking Sentinel to answer without citations does not move it.
Only the engagement coach and the compliance specialist can propose an action that would reach a member. The other four analyse and report. That is a boundary in the code, not a policy we ask them to respect.
See them in conversation
This is the chat surface inside your dashboard, not a channel in your community. Watch what each agent does when the honest answer is smaller than the one you asked for.
A question that crosses learning and compliance. Aggie owns no data of its own, so it has to ask.
Agents
Scroll in, or press Restart, to start the conversation.
Every numbered marker resolves to the row behind it. Aggie carries none of its own: it holds no data tools, so each fact in its reply arrives from the specialist it asked.
See the loop in motion
How an agent goes from noticing something to doing something about it, step by step. Every loop below is a pattern that runs in the product today, including the one that ends with the platform declining to act.
Learners are stalling at one lesson. Beacon names the lesson, drafts the nudge, and waits for you.
Step 1 of 7Observe
Tap any step to read its part.
Beacon is watching
read-onlyEvery pattern has a coverage threshold. If too few members carry the signal, or the records are stale, the agent does not run it and nothing reaches your queue. Messy data makes it quieter, not wronger.
Every number resolves to the record behind it. If the agent cannot cite a row, it does not make the claim.
Every step is logged with the arguments passed and the data returned, so you can reconstruct what happened rather than trusting the summary.
No agent can reach another organisation's data. That is enforced by the database itself rather than by application code, so it holds even if something above it goes wrong.
Stale data, small samples and coverage gaps are stated plainly. Asking the agent to ignore its caveats does not work.
You write your own plan before the agent's proposal unlocks, so you judge it against your thinking instead of being anchored by it.
Agend’s AI runs on Amazon Bedrock. The first question a board asks is where the data goes, and most vendors answer it with adjectives. Here are the four answers, plainly.
Your account is hosted in Australia and the AI runs there too, in the Sydney region. Your members' details do not cross a border to be read, and moving that boundary would be a deliberate decision rather than a default.
Bedrock does not use what we send it, or what comes back, to train any model. Your member data does not become anyone's training set, and there is no setting we could get wrong that would change that.
We run Anthropic's models inside our own Bedrock account. Anthropic is not in the path and receives none of your data. Neither does any other model provider.
An agent sends the rows its question requires, not your database. Everything it touches is logged, so you can go back and see exactly which records were involved.
Where this is up to. We are moving every Agend tool onto this footing, and not all of them are there yet. Ask which capability sits where today and we will tell you straight, in writing, before a pilot starts. Agend is an Australian company and handles member data under the Australian Privacy Principles.
If your board wants the data-flow diagram and the processing terms before a conversation, ask and we will send them first.
The easiest thing for a vendor to do is send every member the nudge, count the ones who came back, and put the number on a slide. We build the opposite in.
30%
held back on every send
When a proposal is approved, a comparable slice of the cohort is deliberately left out of it. They are picked at the same moment, from the same rebuilt cohort, and the selection is hashed so it cannot be quietly reshaped afterwards.
1
honest question it answers
Did the message make the difference, or were those members going to come back anyway? Without a held-back group you cannot tell, and every AI number you have ever been shown has that problem.
The measurement runs on every approved intervention today. Reading it back to you inside the product is the next piece of work, so for now the comparison is something we pull for you rather than a screen you log into.
Members whose CPD deadline is approaching and whose credits sit below threshold, with the cycle window cited, the unknown bucket reported, and the source rows linked.
Not 'completion is down'. The lesson where learners stall and the size of the cohort stalling there. The diagnosis takes seconds; the editorial fix stays yours.
Bounded, consent-filtered, and capped by a thirty-day cooldown so nobody gets nudged twice. If the login signal is too sparse to trust, the pattern does not run at all.
Reports CPD posture in four buckets, compliant, at risk, non-compliant and unknown, with the cycle definition, sample size and coverage gaps named rather than smoothed over.
You approve everything
The default, and where every pattern starts. The agent proposes, you review the cohort and the plan, and nothing touches a member until you say so.
Auto-pilot, one pattern at a time
Once a pattern has earned it, you can let it run without stopping for you. It is opt-in, it is per pattern, and it is never available for compliance messages. Reject one proposal and that pattern drops straight back to asking.
There is no setting that lets an agent decide on its own what a member should receive. Auto-pilot only removes your click on a pattern you have already approved, and any rejection puts that pattern straight back to asking you.
Ask us the hard ones – pricing, migration, the lot. A human from our Australian team replies within one business day.
Ask a questionTalk to us about joining the pilot and we will walk the loop on your members, not a demo set.
