From 5 scattered forms to a single dashboard with AI classification

Context

Professional services company. Five lead entry channels: main website form, two dedicated campaign landing pages, event registration form, and a generic email inbox (info@) receiving spontaneous queries.

The problem wasn't volume —80-120 leads/month—, it was handling. Web forms emailed a distribution list. Whoever opened the email first got the lead. No traceability. No classification. Leads from enterprise companies landed in the same bucket as students asking about prices. And 30% fell through the cracks.

The challenge

• Centralize without dropping existing forms. Each landing had its own form for tracking reasons. Couldn't unify.
• Auto-classify by intent. "I want general info" vs "we want to hire you for a 50k project" vs "I'm doing my thesis, can you answer questions?".
• Auto-route to correct team. High-value to senior team. General info to marketing. Students to a polite auto-reply.

Approach

Custom solution built in 4 weeks after 1 week of discovery:

1. Discovery (1 week). Mapped the 5 entry points. Analyzed 300 historical leads to find patterns: query types, high-value keywords, profiles that historically converted. 4 categories + 1 "noise" emerged (students, spam, non-customers).
2. Unified capture (week 2). Each form stays where it was, but they all POST to a single API endpoint that records metadata (origin, UTM, timestamp) and stores the lead in a central database.
3. LLM classification (week 3). Every new lead passes through a model (with curated prompt) that assigns category and priority (high/medium/low/noise). Explicit criteria: company present, budget mentioned, urgency detected, target sector.
4. Dashboard and routing (week 4). Panel where commercial sees leads in real time, filterable by source, category and priority. Slack notifications for high priority. For noise, polite auto-reply with no human time spent.

Stack: API receiver, internal database, LLM classification in private architecture (data doesn't leave), web dashboard, Slack and CRM integration to auto-create contacts for high-priority leads.

Results

• 100% leads centralized. Zero leads lost in email inboxes from week 4.
• Hot lead response time: from 4-5 days to <1 hour. High priority reaches senior commercials with immediate Slack notification.
• Hot lead close rate up 40% in the following quarter. Main driver: no longer taking 4 days to respond.
• Marketing recovered ~6 hours/week previously spent triaging info@.
• Full traceability: for the first time they knew which channel brought which quality of lead, changing campaign spend.

Lesson applicable

The lead-capture bottleneck is rarely volume. It's response latency and triage. A hot lead that takes 4 days to hear back has gone cold or gone to a competitor. A cold lead that occupies senior commercial time is cost without return.

The most valuable piece for the client wasn't the dashboard, it was knowing which channel delivered which lead quality. They'd been investing equally across landings for years without knowing what converted. They reallocated budget the next quarter.

Technical lesson: LLM classification works well if the prompt is calibrated with the client's own historical cases, not with a generic "classify this lead". First two weeks the model erred on 18%; after refining the prompt with 50 client examples, down to 4%.

Confidentiality note

Client name omitted by NDA. Figures are real or conservative estimates based on client's internal measurements.