CASE STUDY · GTM ENGINEER · SALESPLAYBOOK

Catch every buying
window. Double
the pipeline.

ArchivoX sells SAP archiving into a market of roughly 5,000 DACH accounts. In a market this finite, pipeline doubles through timing: know the day an account enters its buying window, and arrive first with something useful.

For this case study I built that engine. It has run unattended every morning since.

25qualified signals, week one

~$0.10per researched, drafted, routed signal

6-8h → minSDR research per day, before vs after

livedaily autonomous runs on GitHub

Silas Schüttel · silas@automatework.de · repository · live cockpit

01 · GTM Reasoning

The math starts at 300 mails per meeting

Generic cold outbound in comparable enterprise cases books one meeting per ~300 mails. Hold the engine to the same unit and the case decides itself:

VOLUME BURNS THE MARKET

~20 meetings/mo needed for €4M → at 300 mails each: ~6,000 mails/mo, 72,000/yr
The whole market holds ~15,000 relevant people (3 personas × 5,000 accounts) → every one of them in a cold sequence every 9 months, forever
Contacts in one account share the same "no active project" → meetings scale with accounts in a window, extra mailboxes add mails only

ENGINE TARGET: ~40 MAILS PER MEETING

Mails go only to accounts with an open, evidenced window: the biggest failure cause is removed before sending
Right person · their program, named · a prepared give with a yes/no ask
Targets: ~40 mails per Tier-A meeting, ~75 for Tier B · ~1,150 mails/mo total instead of 6,000 · measured from month 3

Tier	Motion	Accts/mo	Mails/mo	Mails/meeting	Meetings
A	Hot signal · ~2 contacts × 2 touches, human-approved	~100	~400	~40	~10
B	Fit + weaker signal · automated sequence	~250	~750	~75	~10
× 37% meeting→opp × €70k ACV			≈ 7.5 opps/mo		≈ €520k/mo

Sensitivity: if the engine needs twice the mails per meeting, the same volume books ~10 meetings (~€260k/mo). The gap closes with levers volume lacks: the existing SDRs call the hot queue (they get 6h/day back for exactly this), expansion signals, inbound routed into the same queue.
Second-order effect: meetings inside a deadline-driven window close above average in a PoC-led motion. Same pipeline euro, more revenue behind it.

02 · The Signal Thesis

Every ECC customer faces a deadline

SAP ends ECC mainstream maintenance in 2027 (2030 extended). Archiving before migration cuts the HANA bill; decommissioning after is mandatory. Detectable, budgeted, deadline-driven: Signal S1.

Tier 1 · buying window open, act within days

S1ERP / S4 transformation projectjob postings, trade press, SI announcements

Companies hiring their own S/4 project leads are inside the window and saying so publicly. The built component runs on this signal: 75 companies in one 7-day window.

S2New CIO / Head of IT / Head of SAPLinkedIn alerts, job-change data

New leaders review legacy contracts in their first 90 days. Same watcher interface, different source.

S3M&A, carve-out, divestitureHandelsregister, Bundesanzeiger, M&A feeds

System separation makes legacy archiving and decommissioning unavoidable.

S4Compliance and audit eventse-invoicing waves, GoBD, NIS2

External deadlines force the retention question onto the agenda.

S5Incumbent ECM distressvendor EOL and pricing news

Displacement windows for the challenger motion and the 40-50% cost story.

Tier 2/3: datacenter exits · SAP Basis/ILM hiring · tech stack · web intent (Dealfront) · marketing engagement routed to outbound. Most live in German-language sources. Signal archaeology (week 1): map the last 20-30 closed-won deals to detectable signals, so priority comes from data.

04 · Proof

Week-one results, inspectable

INPUT · 7 DAYS OF JOB MARKET

150 postings → 25 signals

The first run processed every matching German job posting from the previous 7 days: 150 postings, 75 companies
Filtered out: consultancies, vendors, staffing, and end users lacking transformation context
25 companies remained with an open, evidenced buying window: a truck maker's finance program, a transit operator's own S/4 project leads, a paper group on S/4

COST · PROCESSING THOSE 7 DAYS

~$2 in LLM costs, total

That covers everything: classifying all 75 companies, writing 25 briefs and 25 one-pagers, matching 77 contacts
Per finished signal (researched, drafted, routed): ~$0.10
The same research keeps the case's SDRs busy 6-8 hours a day

DRAFT · TIER A · VARIANT A: QUESTION OPENER

Datenaltlast vor dem S/4-Cutover bei QIAGEN

Hallo Herr H.,

welches Team verantwortet im QIAbase-Programm bei Ihnen die Entscheidung zu Altdaten und alten Dokumenten?

Bei Biotechnologieunternehmen liegen erfahrungsgemäß 40 bis 60 Prozent der ECC-Datenbestände außerhalb der laufenden Nutzung. Wandern sie ungefiltert ins S/4-Sizing, kostet das im HANA-Footprint schnell einen sechsstelligen Betrag pro Jahr.

Wir haben für QIAGEN eine kurze Potenzial-Einschätzung vorbereitet: welche Datenklassen sich vor der Migration archivieren oder stilllegen lassen und was das im Sizing typischerweise spart. Eine Seite, in zwei Minuten gelesen.

Soll ich sie Ihnen schicken? Ein kurzes Ja genügt.

DRAFT · TIER A · VARIANT B: STATEMENT OPENER

Datenaltlast vor dem S/4-Cutover bei BWB

Hallo Herr S.,

im Stabsprojekt SAP S/4HANA beeinflussen lang aufbewahrte Netz-, Anlagen- und Vorgangsunterlagen die Infrastrukturgröße, wenn ECC-Bestand und Archiv nicht getrennt bewertet werden.

Bei Wasser- und Abwasserbetrieben liegen erfahrungsgemäß 40 bis 60 Prozent der ECC-Datenbestände außerhalb der laufenden Nutzung. Wandern sie ungefiltert ins S/4-Sizing, kostet das im HANA-Footprint schnell einen sechsstelligen Betrag pro Jahr.

Wir haben für BWB eine kurze Potenzial-Einschätzung vorbereitet: welche Datenklassen sich vor der Migration archivieren oder stilllegen lassen und was das im Sizing typischerweise spart. Eine Seite, in zwei Minuten gelesen.

Soll ich sie Ihnen schicken? Ein kurzes Ja genügt.

Same reviewed template, one variable slot (the italicized opener). Variant A asks a peer question (here: QIAGEN's digital-solutions VP), variant B opens with a concrete observation (here: the S/4 project lead at Berlin's water utility). Signals alternate 50/50 by signal id; reply rates decide, exported as opener_variant.
The offered one-pager exists: the engine generates it per account, savings ranges computed by a benchmark model in code. The LLM writes prose; the numbers come from arithmetic.

GitHub · PublicThe repository Source, 33 tests, CI, decisions. The daily runs write part of the commit history themselves. Live · GitHub PagesThe signal cockpit 25 signals with briefs, evidence, contacts, drafts. Person data redacted in public artifacts. Docs · DecisionsThe thinking Eight recorded decisions with trade-offs, from "code vs Clay" to "numbers come from code".

05 · Beyond the Cog

The full engine is five layers

The built component is layers 1 to 4 in miniature, for one signal type. The engagement scales it across the taxonomy on the same architecture.

L5 · Feedback

Conversion per signal type → kill or scale. CRO dashboard: pipeline by signal, cost per SQL, SDR time saved.

L4 · Activation

Tier A human-approved 1:1 + LinkedIn · Tier B sequences on warmed domains · Tier C nurture · phone as optional SDR lever.

L3 · Scoring & Routing

Fit × signal → tier, owner, SLA. Both CRMs feed suppression read-only; writes go to HubSpot, already in the stack.

L2 · Signal Detection

Code watchers for German sources (job boards, registers) + Dealfront triggers. Everything normalizes to one signal_event.

L1 · Account Universe

Full TAM (~5k) with fit scores and persisted contacts. Built from ArchivoX's own CRMs first, then Dealfront.

Deliberately out of scope: CRM consolidation (flagged to the CRO as its own workstream), custom sequencing, extra UI. The engine is the brain between market and CRM.
DACH legal reality shapes the motion: UWG §7 pushes Tier A toward individual, evidence-anchored 1:1 messages at low volume, which is also the motion that converts in enterprise.

06 · Deliverable A

From cog to engine in six months

PHASE 0 · WK 1-2

Discovery & signal archaeology

Closed-won deals mapped to detectable signals. CRM audit, ICP workshop, deliverability setup.

EXIT: signal matrix signed off, domains warming

PHASE 1 · WK 3-6

Foundation + MVP

Account universe scored, suppression live, S1 + S2 end-to-end. This phase is what you just saw built.

EXIT: 10+ Tier-A briefs/week, first meetings

PHASE 2 · MO 2-3

Expand & instrument

Signals S3-S6, web intent, Tier B sequences, routing SLAs, dashboard v1.

EXIT: 15+ meetings/mo, research ≤2h/day

PHASE 3 · MO 4-5

Optimize

Feedback per signal type, opener A/B read-out, expansion play on the customer base.

EXIT: €400-500k qualified pipeline/mo

PHASE 4 · MO 6

Industrialize

Runbooks, enablement, handover. Roadmap: partner signals, DSAG events, verticals.

EXIT: engine runs on <0.5 FTE ops

07 · The Questions from the Brief

Fit, track record, and the honest parts

WHO BENEFITS · WHO DOESN'T

Strongest fit: high ACV, finite definable TAM, event-driven purchases (ERP, compliance, leadership change), displacement sales where "why now" beats "what", research-bottlenecked SDR teams. ArchivoX is close to the archetype.
Weak fit: low-ACV velocity motions (volume math wins there), tender-driven public sector, new categories with no detectable triggers, TAMs under a few hundred accounts (human ABM covers them), and any team too slow to work a hot queue: signals decay in days.

BUILT BEFORE · WHAT WORKED, WHAT WAS HARD

Lead universe + evidence-grounded cold email (own GTM stack): Postgres canonical store (~2,400 resolved leads), suppression, AI first lines strictly grounded in scraped evidence, Instantly delivery, HubSpot sync, 38 documented workflows. Worked: the evidence contract lifted reply quality and killed hallucinated personalization. Hardest: entity resolution across sources without silent false merges; solved with match keys, confidence thresholds and a review queue.
LinkedIn / Sales Navigator assist extension (self-built): captures profile and account context, proposes next steps via hotkeys, feeds enrichment and DM drafts. Worked: ~3x throughput per hour while every send stayed a human decision. Hardest: drawing the automation boundary; full automation was easy and wrong. Tier A in this engine draws the same line.

SUPERPOWER · FIRST 3 MONTHS

End-to-end system ownership at AI-native speed: GTM logic, data model, enrichment, CRM integration, production hosting and the AI layer in one person, shipping working v1s in days with coding agents as the daily mode. This case study's engine is itself the sample.
At SalesPlaybook: 2 to 3 client engines shipped to production in the first 3 months, and everything reusable extracted into an internal library (watchers, evidence-contract prompts, routing patterns, HubSpot modules) so each next engagement starts at 60%. Live prototyping in client conversations turns pitches into demos.

KRYPTONITE

My focus binds to the unsolved problem. Once the interesting part is cracked, my attention drops on the last meter, and that is where avoidable careless mistakes can slip through: the typo in a customer-facing line, the stale number in an export.
How I compensate: structurally. Validators and automated checks on everything that leaves a system (this engine's evidence contract exists partly for that reason), checklists for handovers, a second pair of eyes on client-facing work. My own final read is the least reliable check in the chain, so I build chains that survive it.

Catch every buyingwindow. Doublethe pipeline.