A useful AI starts with well-organized data

The federation brings together several thousand members across the country, organized in local sections and regions. Its leaders came to talk about AI, and specifically about an agent that members could reach on WhatsApp.

On our side, we wanted to understand what the agent would be reading from. An agent answers with what the organization knows. If that knowledge is spread over a dozen tools, the first job is not the agent.

Here is what we picked up during the conversation, rephrased and without any figure:

• One tool per need, so silos. Every committee has its own tool, its own self-assessment, its own database. Nothing flows up into the central application, which has no idea what the others know.
• The same email goes out to every member, with no personalization. Only a minority of recipients open it.
• Forms ask an already authenticated member again for what they have already provided.
• Figures are requested by hand at every level: local section, region, national.
• Company profiles go stale, because the member who should update them sees no immediate benefit.
• Local leaders send every message twice, on WhatsApp and by email, because part of the membership reads only one of the two.

Part of this comes from a legacy tool with no open programming interface, which is why data sits where it was typed and goes no further.

We suggested four steps, in this order. Skipping one means the next one stands on sand.

1. 1 IT diagnostic List the tools, the databases and who uses them.
2. 2 Information system urbanization A shared base, and exchanges between tools.
3. 3 Automation Remove re-entry and manual follow-ups.
4. 4 AI An agent working from clean data.

People come for the AI, but the real subject is the spreadsheet that gets passed around everywhere. Once the figures live in one place and are fed automatically, an agent has something solid to answer from. Before that, it only repeats the confusion.

A membership network has an advantage: it can try something with a few of its own members and then show it. The approach is to show what works once it is done, not to promise it beforehand.

We would start with a test among a few participants, collect their feedback, and only then roll out. The aim is a usage that members discover by themselves, without training. If a tool needs a tutorial, it is not ready yet.

Natalia answers on WhatsApp and in chat. You create your agent yourself and try it on your own cases, with no meeting required.

Deploying an AI agent in a member community: the objections, one by one