I wanted to share a scenario I have been running for a few weeks now, mostly because it solved a problem that was eating way more of my time than it should have. Every lead that came in, whether from a form, an ad, or a DM, was landing in one inbox and getting treated the same way. Someone asking a vague pricing question got the same attention as someone who was clearly ready to buy. I finally sat down and built something to fix that, and it turned out to be one of the more satisfying automations I have put together this year.
The problem I was actually solving
Nothing about my sales process was broken exactly. It was just slow. A lead would come in, I would glance at it whenever I checked email next, decide if it was worth a real reply, and then write that reply from scratch. Multiply that by a dozen leads a day across a few different sources and you get a lot of small delays that add up to lost deals. By the time I got around to some of them, they had already gone quiet or moved on to someone faster.
What the scenario actually does
Here is the rough shape of it, in case it is useful for anyone dealing with something similar.
It starts with a webhook that catches new leads from a few places at once, including a form on my site and a Google Sheet where I log leads that come in through Instagram conversations. I lean on InstantDM for a chunk of that Instagram side, since it captures DM replies against Meta’s official API rather than something sketchier, and I did not want to lose that context once a conversation looked like a real lead.
From there the scenario runs the lead details through an OpenAI module that scores it based on a handful of signals I care about, things like budget mentioned, urgency language, and how specific the request is. That score gets written back into a Google Sheet alongside the lead record, so I have a running log I can actually look at later instead of just trusting the automation blindly.
Next comes a router. High scoring leads get pushed straight to a Slack channel with an urgent tag so I see them immediately. Medium leads go into a follow up queue that I check once or twice a day. Low scoring leads still get logged, but they get routed into a slower nurture path instead of demanding attention right away.
The part I am most proud of is the follow up draft. Once a lead is scored and routed, another OpenAI module drafts a personalized follow up email based on what the lead actually said, not a generic template. I still read every draft before it goes out, because I am not comfortable letting AI send things unsupervised, but having a solid first draft sitting there already has probably saved me an hour a day, easily.
What I would change if I rebuilt it
If I were starting over, I would add a cooldown step before the AI drafts the follow up. Early on it fired the moment a lead hit a certain score, and a couple of times that meant sending a follow up before I had even finished reading the original message myself, which felt a little backwards. I also underestimated how much time I would spend tuning the scoring prompt. Getting it from decent to actually reliable took more iteration than I expected, mostly around teaching it what urgency language looks like in my specific niche versus generic phrasing.
Why I am posting this here
I know a lot of people in this community have built lead routing scenarios that are more advanced than mine, and I would genuinely like to hear how you are handling the scoring step specifically. Are you using a single AI call like I am, or breaking scoring into separate modules for different signals? I have a feeling there is a cleaner way to do this that I have not landed on yet.
Happy to answer questions about the setup if anyone wants specifics on the modules or the prompt structure. This is one of those builds where the last twenty percent took longer than the first eighty, but it has genuinely changed how fast I respond to leads, and that alone made the whole thing worth it.