The challenge: a sales team outpaced by its own market
A City of London fintech firm had a strong product and a small, capable sales team, but a lead generation process that couldn't keep pace with either. Prospecting was manual: two SDRs working LinkedIn Sales Navigator and a handful of industry directories, qualifying leads by hand, and logging what they found into HubSpot at the end of the day rather than in real time. The team was good at closing once a genuinely qualified lead reached them. The bottleneck was upstream: not enough qualified leads were reaching them in the first place, and the ones that did often arrived days after the prospect's actual moment of interest had passed.
The firm's leadership had tried a generic lead-scraping tool before working with us, and it made the problem worse, not better: it surfaced volume without qualification, flooding the CRM with contacts that didn't match the firm's actual ideal customer profile and burning SDR time on outreach that never converted. What they needed wasn't more leads. It was more of the right leads, found and qualified continuously rather than in daily manual bursts.
The approach: multiple specialized agents, not one big scraper
We built an autonomous multi-agent pipeline rather than a single monolithic scraping script, because lead generation for this client genuinely broke down into distinct sub-problems that benefited from being solved separately. One set of agents handled sourcing, continuously pulling candidate leads across 14 data sources spanning company registries, funding announcement feeds, hiring signals, and professional network data. A second agent layer handled qualification, scoring each candidate against the firm's ideal customer profile using firmographic and intent signals rather than keyword matching alone. A third layer handled enrichment and routing, attaching contact and context data and pushing qualified leads directly into HubSpot with the reasoning behind the qualification score attached, so SDRs could see why a lead scored the way it did rather than trusting a black box.
Node.js orchestrates the agent pipeline, chosen specifically for how cleanly it handles the kind of concurrent, I/O-heavy polling this system runs continuously across 14 external sources. The system runs around the clock rather than on a schedule, because buying intent signals, a funding round, a new hire in a relevant role, don't wait for business hours, and the firm's competitors weren't waiting either.
Implementation: tuning qualification before scaling volume
The first phase deliberately held sourcing volume down while the qualification agent layer was tuned against the SDR team's actual judgment. Every lead the system scored was checked against what an experienced SDR would have called it, and the scoring logic was adjusted where the two disagreed, rather than shipping a qualification model straight from a generic template. This phase ran for several weeks before we opened up full sourcing volume, a decision made specifically because the earlier failed tool this client had tried made the opposite mistake: maximizing volume before qualification logic was trustworthy, which is how a CRM ends up full of noise.
One constraint worked around during implementation was data-source rate limiting: several of the 14 sources enforce request limits that would throttle a naive continuous scraper. The sourcing agents were built with source-specific pacing and caching logic so the pipeline respects each source's limits without losing continuous coverage, which took more engineering effort than a single generic scraper would have, but was the difference between a system that runs reliably for months and one that gets blocked within weeks.
Results: 310% more qualified leads, real pipeline generated
Once fully live, qualified lead volume reaching the SDR team increased by 310% compared to the prior manual process, measured against the firm's own qualification bar rather than a looser definition of "lead." Because qualification happened before a lead ever reached a human, SDR time spent on outreach that was never going to convert dropped sharply, and the team's actual close rate on inbound-qualified leads held steady rather than diluting, which is the failure mode generic lead-volume tools usually produce.
In the first measured period after launch, the pipeline directly attributable to leads this system sourced and qualified reached £2.1M, tracked through HubSpot's deal attribution against the enrichment data the system attached at the point of handoff. For a firm whose previous bottleneck was lead volume rather than sales skill, that number reflects capacity the sales team already had but couldn't previously reach.
