Deployment is the halfway point
An AI project is finished only once it is in the user’s hands, being used for the work it was set out to do. It is tempting to put the project together, hand over the keys and hope for the best, which leaves no room for iterations or improvements drawn from how people really interact with the system. Accord delivered a research assistant last week, and we are making sure we keep consistent contact with the users throughout the first month, to make the final tweaks that fit the system to the real way their company works.
What we built
We are working with a national association that was struggling with the research step that comes before any outreach to prospects. We set up a relatively simple ChatGPT project aimed at that research step.
In the first session we went through it together and tested it on two companies they actually wanted to reach out to. On the second, a club they had called the day before without getting anywhere, it named both the person to ask for and the one who really decides. Testing it on their own work rather than our examples showed us exactly what was working, what could be improved and what needed changing. It also showed something we would never have put in a test ourselves: straight away, they wanted to use it for a kind of organisation the instructions didn’t cover yet.
Next step
Iterations with the project in production are fundamental, because some refinements only become apparent once the real people using the system get their hands on it. So we are leaving the research assistant with the client for a couple of weeks before touching it. When we come back, we will review whether the information it presents is actually what helps in the sales process, especially the news articles it finds and the “reason to contact” it gives for each prospect. We have already scheduled the next sessions with this client to work through what we find. Our responsibility doesn’t stop at deployment, but when the people using the system are seeing the results.