Case study: how we automated lead processing and cut the time from 2 hours to 5 minutes
A lead came in at 23:40 on a Saturday. The manager saw it Monday morning - if it didn’t get buried in a flood of a hundred other emails. By that point the client had already reached out to a competitor and received a reply. That’s how the B2B company we worked with was losing leads they had paid advertising money to acquire.
Leads came from everywhere: email, a website form, a couple of messengers. Each one had to be read, understood - what does this person want, how urgent is it, who to pass it to - and logged in the database. One lead took a manager up to two hours - not because the work was complex, but because there was a lot of it and all of it was manual. Overnight and weekend leads sat untouched for hours. Some simply got lost.
And leads were scattered across different places. Email in one window, the form went to a different address, messengers were a third app. To not miss anything, the manager kept all of this in their head and manually checked each channel. In practice that doesn’t work: some source always slips through, especially toward the end of the day. It’s not about anyone’s inattentiveness. It’s just that there’s nowhere to look at everything in one place.
The task
We needed to remove manual sorting from the chain and make sure a lead from any channel reached the right person quickly and without loss - including at night and on weekends.
Constraints we kept in mind:
- There are multiple channels and the list will grow. The solution shouldn’t break when a new source is added.
- Client data is sensitive. The client didn’t want it routed through third-party clouds, so we went self-hosted.
- Managers need to stay in the loop. The goal was to take routine off their plate, not to cut them out of the process and hand everything to a machine.
So the task wasn’t “bolt AI onto email.” We needed a working pipeline from lead intake to CRM entry, where AI is one step - not the whole point.
What we built
A pipeline on n8n with an AI model inside. The logic is simple: one entry point for all channels, AI turns text into data, then the lead automatically routes to the right people and gets recorded in the database. The manager engages with an already-processed lead. The lead goes through five steps.
1. A single intake point. A webhook in n8n collects leads from all channels at once - email, form, messengers all come into one place. Before, each channel lived separately and the manager kept several windows open. Now the source doesn’t matter; everything flows into one stream. A new channel appears - we add another input to this same point, everything else stays untouched.
2. AI parses the text. The model reads the lead and extracts structure from it: request type, urgency, the essence of the inquiry, sender’s contact details. Here a live email text gets turned into data that the automation works with downstream. A three-paragraph email with no subject line and typos gets parsed the same way as a short form submission.
3. Enrichment and routing. Based on what AI parsed, the lead goes to the right person or department. An urgent request from a major client and a general pricing question go to different people with different priorities. The manager receives an already-processed lead, not a raw email.
4. Database entry. The lead automatically lands in the CRM. Nothing needs to be copied manually, and the history doesn’t get lost - everything that came in is recorded.
5. Auto-notification to the client. Right after intake the person receives a reply that their inquiry was received and someone will be in touch. Before, the client sat in silence not knowing if it even got through. Now they get a confirmation within the same minute - including overnight.
The entire pipeline runs on n8n plus an AI model, deployed on the client’s own infrastructure. No data leaves to third-party services.
Results
Numbers before and after:
- Processing time per lead: ~2 hours -> ~5 minutes. This is the core result. What used to eat up hours of manual work now takes minutes, and a person at the sorting stage is no longer needed.
- Leads are processed 24/7. Overnight and weekend ones no longer sit waiting until the start of the workday. The pipeline doesn’t take lunch breaks and doesn’t go to sleep.
- Fewer lost leads. A lead from any channel gets recorded in the database and reaches the manager. That Saturday night email scenario is now impossible - it gets parsed and logged the moment it arrives.
- Managers focus on selling, not sorting email. The freed-up time went toward working with clients, not processing the incoming stream.
The difference is most noticeable overnight and on weekends. Where there used to be a gap of several hours, the same pipeline that runs during the day now runs around the clock.
What we learned
A few things worth knowing if you’re thinking about something similar for yourself.
Better to start with one channel. We didn’t try to connect everything at once. We launched one source, confirmed the chain worked from intake to CRM entry, and only then connected the rest. It’s easier to catch problems that way - when there’s one channel, you know where to look.
AI classification had to be trained on real leads. On demo examples everything looks fine. But real emails have typos, cut-off thoughts, two different requests in one message. For the first few weeks we watched where the model got it wrong and tuned the parsing to the client’s actual incoming flow. This is a normal stage - it’s worth building into the plan rather than expecting it to work perfectly from day one.
People need to keep control over edge cases. Not everything should be automated blindly. When AI isn’t sure about the type or urgency of a lead, it goes to a person rather than falling through to some default bucket. This is a safeguard: the machine handles the clear-cut volume, and people decide at the edges. Without this, automation will quietly make a mistake at some point and nobody will notice.
In short: the value here isn’t in AI itself, but in the working process built around it. A single intake point eliminates lost leads. Text parsing eliminates manual sorting. Routing and database entry eliminate copying by hand. AI is one step out of five, and it only works because the other four are in place.
Want the same
If incoming leads at your company are being processed manually and some get lost overnight - this is solvable. We analyze your flow, show what can be automated, and build a pipeline for your channels and your CRM. Self-hosted, no data leaking to third-party clouds.
Write to us: puramind.ai#contacts.