Share it

Faced with AI, we sometimes have the impression of ending up in a pharmacy without a pharmacist. The solutions are numerous and the promises are attractive... but it is difficult to know if the solution that we thought was miraculous will not be in vain or even harmful.

To see more clearly, we spoke with Antoine Ortiz and Florian Baron, respectively content manager, and Business Operations Manager at AssoConnect. They don't present themselves as experts in AI, and maybe that's what makes their approach valuable. Their advice is simple: test, doubt, adjust.

‍

Experiment before adopting: the AssoConnect method

AssoConnect is this online tool that makes life easier for thousands of associations by automating time-consuming tasks for non-profits (accounting, management of contributions, etc.). Their objective is clear: to lighten the daily life of associative structures and to allow teams to refocus on their mission. Recently, AssoConnect has been exploring what AI can offer them internally, before seeking to benefit associations. We explain everything to you.

Step 0: Put your foot in the stirrup

Before even launching advanced tools, the AssoConnect teams started by testing ChatGPT, via a Pro account. An essential “step 0" to become familiar with the main language models to understand what they do well or not so well, how to properly build a prompt, and above all, to take a step back on the results. In short: start by keeping things simple, with a tool that everyone knows.

Step 1: identify the AI to be tested

It's hard to know in advance what AI can really do for an organization. AssoConnect therefore chose to test tools without preconceived ideas, and it was through word of mouth that their teams discovered Dust, a tool for creating its own AI agents, connected to its ecosystem (Slack, Notion, Google Drive...) and to several models (GPT-4, Claude, Mistral...) and to several models (GPT-4, Claude, Mistral...).

Step 2: Better 8 curious than 80 lost

‍There is no question of launching a massive deployment. A team of 8 volunteers started by testing Dust on a small scale. Each with its own uses: Florian for database analysis or content reformulation; Antoine for generating YouTube video descriptions. At a rate of 29 euros per user per month, the approach makes it possible to limit licensing costs during the first tests.

Step 3: How do you know if AI is really doing the job?

Each volunteer evaluates responses using a “thumbs up/thumbs down” system, which makes it possible to report reliability statistics at the team level. In addition to this approach, the team meets every two weeks to discuss use cases, limitations, and new features.

And safety in all of this? What we share... and what we keep to ourselves

Even though Dust respects the confidentiality of data on paper, AssoConnect remains vigilant. Florian avoids injecting sensitive data into it. For some requests, he favors Patagonia AI, a macOS application that allows you to run AI models directly on your computer.. The models are slower and limited, but the data is not sent to servers abroad.

Ultimately, the AssoConnect team is even considering hosting open-source models on its own servers: for total control of the data chain.

Finally, starting to use AI is like integrating a new colleague: you have to train, test, supervise him... before trusting him. No miracle tool, but precise uses, to be built step by step. With time, curiosity and the right adjustments, AI can become a lever, provided you choose the right tools and never lose sight of your mission. Since its first tests, AssoConnect now sees Dust as an internal tool, useful for quickly finding the right information and automating analyses. At the same time, each team uses tools specific to their needs, such as ProductBoard for the analysis of customer feedback. Faced with the abundance of solutions, AssoConnect gave up on the miracle tool. The real challenge today is to find the right balance to limit the number of tools while covering the majority of use cases.

AI, at the service of associations

Since its inception, AssoConnect has been designing tools to simplify associative management. True to this philosophy, the organization is now putting AI at the service of its users.

An AI accessible to as many people as possible

1000 documents generated automatically each month: this is the concrete impact of the tool Writing the minutes of the general meeting deployed by AssoConnect. This solution allows associations to use AI on a task with low added value, which is however essential to their operation.

At the same time, the AssoConnect teams also offer a tool that simplifies the creation of associations by generating its statuses for free and in a few clicks.

In short: 2 solutions designed to support associations in their administrative management.

The AssoConnect method for developing an AI project

Take the example of the tool for generating associative statuses, developed with AI. This project, led by Florian, required several weeks of work. Generating statutes in accordance with the law of 1901 represented a real technical and legal challenge. AssoConnect has therefore put the means to list the legal criteria to be respected, test the algorithm in real conditions, and supervise it in a reinforced way to identify the gray areas in the algorithm.

Automation was only deployed once the entire process was fully under control. In terms of tools, the team relied on Retool to build the interface, and on WorkflowAI to compare and select the most suitable models for each task. A precise methodology between technical efficiency and legal rigor.

Even today, the experiments continue. For AssoConnect, AI is not a magic tool, but a colleague in training: you have to support it, supervise it before you can trust it. What's at stake? Know how to delegate without ever losing control.

At Share it, we particularly like the AssoConnect approach: an AI designed, not as an end, but as a pragmatic lever in the service of impact.

—

Do you want to discover other use cases around AI? Discover the old articles from the IA for Good program by clicking On this link.

Have you experimented with AI tools in your organization yourself? Do not hesitate to contact us via the address contact@share-it.io : we are interested in your experience!

‍