A 10,000-account German energy TAM, scored by AI and routed through one Attio workspace, with calls, cold email, LinkedIn, and meetings all synced back in real time.

The situation

Meteoric sells continuous, stop-loss energy procurement to German industrial companies instead of the annual-RFP model the market is used to. The commercial problem was never “we need a CRM.” It was finding the right companies at the right moment. German Mittelstand manufacturers don’t post “we’re shopping for an energy supplier” anywhere. The signal is buried in hiring patterns, capex filings, news mentions, and registry changes, scattered across a universe of roughly 10,000 candidate accounts.

Three things were missing:

• No way to size or rank the market. Which of 10,000 manufacturers are actually in-ICP, and which of those are close to a buying decision right now versus in 18 months?
• No connected outbound layer. Calls, cold email, and LinkedIn outreach existed as separate efforts with no shared record of who’d been touched, replied, or gone quiet.
• No system turning signal into action. Even if you know a company is in-market, someone still has to notice, prioritize, and write the first message, every day, across thousands of accounts.

What we built

A signal-to-pipeline system with Attio as the spine, AI doing the scoring and first-draft writing, and every outbound channel wired back into one record per account.

TAM mapping and tiering

10,000 German accounts sized and segmented by NACE code (process manufacturing: chemicals, lime, metal fabrication, food processing) and revenue band, validated against two early signed customers: a silo manufacturer and a lime producer, both energy-intensive Mittelstand businesses.

Tier S (Hot)    50-150 accounts    daily signal sweep
Tier A (Warm)   800-1,500 accounts weekly signal sweep
Tier B (Cold)   8,000-9,000        monthly signal sweep

promote: B→A on 1 signal / 90 days, A→S on 2 stacked signals / 60 days
demote:  S→A after 30 days quiet, A→B after 90 days quiet
stack:   3 signals in 90 days = active buying window → Slack alert

Every account starts in Tier B and moves up automatically as real-world signal accumulates: job postings, news, registry filings, capex announcements, pulled from PredictLeads, Firecrawl, the German commercial register, and Bundesanzeiger filings, classified by a Claude model running on the cold tier to keep cost down at 9,000 accounts.

AI scoring and drafting, live on every deal

Five custom fields run on every Attio deal record: priority_score, priority_signal, ai_draft_subject, ai_draft_body, draft_generated_at. A morning batch job scores every active deal and pre-writes outreach drafts for the highest-priority ones. The rep opens their day with a ranked list and a draft already written from the actual signal that triggered the score, not a generic template.

Every outbound channel, one record

Aircall     ── calls logged back to the Attio company/deal record
SmartLead   ── cold email sequences, replies tracked in Attio
HeyReach    ── LinkedIn outreach across 3 senders, replies routed via webhook
Granola     ── meeting notes auto-synced to the matching Attio deal
Linear      ── engineering/ops tickets tracked separately, linked where relevant

No channel operates in isolation. Aircall calls, SmartLead email replies, HeyReach LinkedIn activity, and Granola meeting notes all land on the same record, so a rep can see the full account history in one place, and the AI brief pulls from all of it. Linear keeps the engineering and ops side linked wherever it touches a deal.

Deal Desk: the internal command center

A purpose-built internal tool (Next.js, no database, reads live from the Attio API) gives the sales team a single screen: deals needing attention on the left, full context on the right across four tabs (Granola notes, Gmail threads, AI-generated brief, raw Attio record). A “morning brief” button runs the batch scoring job and surfaces the accounts that moved overnight. Email sends go out through the rep’s actual Gmail account via Composio, not a shared sending domain.

Newsletter and event layer

A parallel Resend-based pipeline handles event invites and an ongoing newsletter to the sales-stage audience: 141 contacts synced live from the Attio deals pipeline, MJML-templated, sent in throttled batches to protect deliverability.

Outcome

• 10,000-account TAM scored and continuously re-ranked without anyone manually reviewing a spreadsheet.
• 5 AI fields live on every deal, turning a cold list into a daily ranked worklist with drafts already written.
• 3 outbound channels (calls, email, LinkedIn) plus meeting notes, all writing back to the same Attio record. Nothing lives in a tool the rest of the team can’t see.
• 2 signed customers (a silo manufacturer and a lime producer) validating the ICP the whole tiering model is built on.
• A morning routine, not a quarterly review: the system finds the 10-20 accounts that moved overnight and tells the team why.

Why this matters for B2B energy and other long-cycle industrial sales

Industrial buyers don’t announce intent. The companies most likely to switch energy suppliers this quarter look almost identical, on paper, to the 9,000 that won’t move for two years, until you stack the right signals: a hiring spike, a capex filing, a registry change, a news mention, all within a tight window. The fix isn’t more data, it’s a system that promotes accounts as evidence accumulates and demotes them when it goes quiet, then hands a rep a ranked list with the reasoning attached. Most industrial sales teams have the data sources already. What’s missing is the architecture connecting signal to score to outreach to one shared record, and that’s the part that has to be built, not bought off a shelf.