Internal

CloseRocket — keep it in the LOOP

A 15-minute read for the whole team — the product told mostly through one lead's journey, and what happens to him after the meeting.

01What we're building

An AI sales operating system: it takes leads from cold to a confirmed meeting — through every channel that works, in the lead's own language, by the rules of the lead's own country — and keeps working after the meeting, turning won deals into expansion and lost deals into future pipeline. The client chooses how much of it to drive: themselves, with us, or fully run by us. The meeting is the only thing the system doesn't do — for now.

Today CloseRocket already lives in one platform — Lead Agent finds prospects, Reach Engine runs outreach, Klara fills the CRM. What's been missing is the brain that connects them: today these still run as separate pieces. The new generation adds the one orchestrating intelligence on top — Klara plans every touchpoint, writes every message, reads every reply, captures what happened after the meeting, and decides what happens next. The first calls are made either by Klara's AI voice or by a real CloseRocket caller. Human salespeople — the client's — enter the story exactly once, with a confirmed meeting in hand.

Market scope: the world. We start in CEE because it's home — and because it's the hardest legal terrain in the world for outreach; a system that runs clean here runs clean anywhere. The architecture is law-aware by design, not Europe-locked: compliance is a per-country rules registry (verified for 11 countries today, incl. non-EU Serbia), and a US client would mean CAN-SPAM/TCPA rows in that registry — not a new product. Languages are configuration, not code: launching with Slovak, Czech, Hungarian, German and English; adding Polish or Romanian is a content task.

5channels designed,
4 already live — the rest earn
their place by performance

5launch languages —
new ones are config, not code

11countries' outreach law
verified & encoded

1touch per lead per day, max —
coordination, not spam

02The problem today

Why this needs to existEvery B2B company runs outreach as disconnected tools: an email tool, a LinkedIn tool, a calling list, a CRM. Nobody coordinates them — so the same lead gets hit from three sides, nobody knows which channel actually converted, and salespeople burn hours on follow-ups and data entry instead of selling.

Channels first → chaos

Same lead contacted by 4 tools at once

No idea which channel converted

Adding a channel = months of work

Salespeople do admin, not selling

A "no" is a dead row in a database

One brain first → channels plug in

Lead experiences one coordinated journey

Every decision is recorded and explainable

New channel = plug in a connector

Salespeople get only hot, prepared meetings

A "no" routes the lead somewhere useful

This "one brain first" idea is the foundation of everything below. We didn't invent it under pressure — the team sketched the same architecture independently a year ago. Now we finally build it.

03Peter's story — from cold to the meeting

One lead, seven daysPeter Novák is CFO of HORNBACH Slovakia. He has never heard of our client. Here is exactly what the system does with him — every step real, taken from the product design.

Day 0

Klara discovers Peter (Lead Agent: public data + Apollo). She checks the legal rules first: email allowed with opt-out, SMS blocked — he never consented. Picks his senders: one mailbox, one LinkedIn profile — and sticks with them, so Peter always talks to the same "person".

Day 0

LinkedIn connection request goes out at 10:43 — not 10:00 sharp; the system adds human-like randomness and respects daily limits so profiles stay safe.

The planned email waits until tomorrow — one touch per day per lead, across all channels. Always.

Day 1

First email: one specific observation about HORNBACH, one question. No "I hope this finds you well."

Day 3

Peter accepts the connection. The system notices instantly and unlocks the next step.

Day 4

LinkedIn message — short, personal, from the same profile that sent the invite.

Day 5

Peter replies: "What would this cost for 200 vendors? Can you connect to SAP?" The system freezes all planned touches for HORNBACH and classifies the reply: a pricing question — high-stakes. Klara drafts an answer; a human approves it in one click; it goes out the same hour.

Day 7

Klara calls Peter. AI voice, introduces herself as an AI, qualifies him (budget, authority, need, timeline) and agrees a meeting: "Thursday 10:00?" — "Works."

Day 7

Hot handoff. The client's salesperson — his first contact with Peter ever — gets the lead in his CRM, an email briefing, and at 17:30 Klara calls him with the day's digest: full history, qualification with quotes, and a 3-point playbook.

Seven days, six touchpoints, three channels, zero spam, zero manual work for the client — and the salesperson walks into the meeting better prepared than he'd ever be on his own. Most tools end the story here. Ours doesn't.

04Act 2 — what happens after the meeting

The part every other tool skipsThursday 10:00, the meeting happens — the one step the system doesn't perform. Friday isn't silence. At the end of the day Klara calls the salesperson: "How did it go with Peter?" Thirty seconds of conversation, and the outcome is captured, the record is written, and the system knows what to do next. No form, no CRM homework.

✓ WonPeter becomes a customer. Four months later his usage hits a milestone — the upsell flow opens, built on his real numbers, not a generic pitch. His colleague from procurement surfaces as a cross-sell candidate for product B. One closed deal quietly seeds two more.

✕ Lost — with a reason"Chose a competitor" → Peter enters a 12-month nurture: useful content, no pitch, and a wake-up when their contract season nears. "Wrong product for us" → 30-day cooldown, then product B opens with a completely different angle. The reason decides the route — a "no" is information, not an ending.

↻ No-showPeter didn't make it. The system reschedules, the salesperson gets the new slot, nobody has to remember anything. Three no-shows would route him to a polite break-up with a re-entry hook — persistence with dignity.

This is the difference between an SDR tool and an operating system: the meeting isn't the finish line, it's one event in a relationship the system keeps managing — in both directions, forever.

05The golden rule

The client's salespeople never chase cold leads. They appear exactly once — at the meeting, with a confirmed slot and a playbook. Everything before it, and everything after it except the meeting itself, is Klara and CloseRocket.

This is the product boundary in one sentence, and it's also the pitch. Calls in the cold phase are made either by Klara's AI voice or by our human callers — never by the client's reps. "Success" has one strict definition everywhere: a confirmed meeting in the calendar. An interested reply is progress; a "maybe" is progress; only a confirmed meeting counts as conversion. That keeps our reporting honest and our promise measurable.

The meeting itself needs no calendar integration to start: Klara agrees the slot in conversation, the salesperson sends the invite and confirms with one click. Simple, works for every client from day one.

06The Learning Loop — the brain of the system

This is what makes CloseRocket a brain, not just a toolA tool does the same thing today it did yesterday. A brain knows whether what it did actually worked, and feeds that back in — so it gets sharper every week. That feedback cycle is the system. Everything else exists to make the loop possible.

Reach out across every channel → listen to whether it landed → learn what worked → adapt the next pass. Every cycle feeds the next, so the system that contacts your lead next week is better than the one that contacted them this week. A loop that just runs is automation. A loop that learns compounds.

What powers the loop

For the loop to work, the system needs four things underneath it — one shared brain instead of four disconnected tools. Channels (email, LinkedIn, phone, SMS) are deliberately dumb: they deliver what the brain decided and report back what happened. All the intelligence lives here:

Memory — so it can listenEvery touch, reply, call outcome and decision, for every lead, forever. This is how the loop knows what already happened — and why the system can say in July: "In May you asked me to come back in Q3 — here I am."

Planner — so it can reach & adaptOne brain decides who gets touched, when, on which channel, with what message — across all channels at once. That's why a lead never gets an email and a call on the same morning, and why pausing one lead pauses everything for them instantly.

Guardian — so it reaches safelyBefore anything leaves: Are we allowed? (the law of the lead's country) · Are we polite? (one touch a day; one conversation per company) · Are we safe? (channel limits, warm-up, business hours). A "no" at any gate stops the action and records why.

Explainability — so learning is trustworthy"Why didn't step 3 go out?" always has a real answer — "the LinkedIn profile hit its daily limit at 13:40; moved to 9:00 tomorrow." Every pass is on the record, which is exactly what makes the learning auditable instead of a black box.

Product grooming — brainstorm

How we know a loop is actually learning

Every loop inside the brain is defined by three things — and those three decide whether it just runs, or actually learns. This is the framework we run every new loop through in grooming — before building, answer all three.

OBJECTIVEWhat "done well" looks likeThe goal of the loop, in one sentence. What is it actually trying to accomplish? If you can't say it plainly, the loop isn't ready to build.

METRICHow the system knowsHow the system itself — not just you reading every output — can tell whether a pass came out better or worse. If you can't answer this one, you've found the real work.

BOUNDARYHow far it goes aloneWhat the loop can do on its own, and where it stops to check in with a human. The line between autonomy and oversight.

The design names five loops. Each is built to either just run, or actually learn — here's the plan for each (nothing here is live yet; this is what we're building toward):

Loop	Objective	Metric the system reads	Boundary	Design intent
Safety governor	Keep accounts healthy — no bans, no throttles	Reply rate + acceptance rate per sender profile	Auto-throttles within defined limits; cannot change limits themselves	LEARNS
Trust dial	Earn the right to act more autonomously over time	% of drafts approved without edits, per category, 30-day window	Suggests Autopilot at ≥95% — human flips the switch, never the system	LEARNS
Reply classifier	Correctly read every inbound reply and route it	Human overrides logged — not yet fed back automatically	Low-confidence → review queue; high-stakes always human	RUNS
Content loop	Send messages that actually get replies and meetings	Reply rate + meeting rate per template variant (A/B instrumented from day 1)	Phase 1: static weights. Phase 2: system suggests reweights, human approves. Phase 3: auto-reweight behind a sample gate	RUNS → LEARNS
Lifecycle routing	Send lost leads to the highest-value next step	Fixed cooldowns and routing order — no signal read-back yet	All routing changes require a human decision	RUNS

Why one client is never enough

A single client's reply data is too thin to learn from — small sample sizes, varying segments, months before a signal is worth trusting. But all clients together write the same play: the same industries, the same objections, the same timing patterns. The content loop aggregates variant performance anonymously across all workspaces and feeds it back into the shared seed templates everyone starts from.

That's the network effect: a DIY client running 200 leads/month can't build this alone. A client who joins later starts from templates that already have 50,000 passes of learning behind them. The system compounds. A competitor can copy the features — they can't copy the data.

One binding rule The system may reweight what a human approved. It may never send what a human hasn't seen. Learning chooses among approved options — it never freelances. Enforced at the instrumentation level, not just in the prompt.

07The channels

Channels plug into the brain — adding one takes days, not monthsBecause all the intelligence lives in the loop, a channel is just a connector that delivers and reports back. A new channel is a connector project measured in days, not a rebuild — and any single channel can be switched off (for one lead, one country, or entirely) without breaking the rest.

LinkedIn — built in via API (~€4.5/account vs $79–99/seat at reseller tools; pilot currently on HeyReach ~€220/mo).
Email — through Instantly (deliverability, warm-up and rotation are its whole job).
Phone — Klara's AI voice on our Vocalls platform, plus human CloseRocket callers for the relationship-heavy calls.
SMS — already live through O2.
WhatsApp — designed and waiting; goes live where consent exists and when performance earns it the slot.

08Loops for leads — neverending

Leads never die, they just change stateEvery other outreach tool runs a straight line: you reach out, you get a yes or a no, the lead disappears. CloseRocket runs a loop. Every exit has a return path, and every return is warmer than the first approach — because the system remembers everything.

"Ask me in Q3." Suspended with a wake-up date. On July 1st: "In May you asked me to come back in Q3 — here I am." No human remembered anything.
"We don't need product A." 30-day cooldown, then product B opens with a completely different angle. Three products = three honest chances.
Ghosted — then the news changes. A silent lead is monitored in the background. When his company gets acquired, Klara re-opens with context: "Congrats on the acquisition — priorities usually shift. Worth a fresh look?"
Won feeds the loop hardest. Customers generate upsell on real usage milestones and cross-sell through colleagues — the warmest pipeline that exists.

One company-level protection: if two people from the same firm are in different campaigns, the system runs one conversation stream per company at a time — the number-one cause of "why is your company spamming us?" eliminated by design.

09Replies handled like a pro

Why this mattersReply handling is where outreach tools quietly fail. An out-of-office mis-read as a real answer corrupts stats and freezes pipelines. We made machine mail a first-class citizen.

Out-of-office: recognized instantly (Slovak, Czech, Hungarian, German, English — and the recognition patterns are data, so adding a language doesn't touch code), everything reschedules to the return date + 2 days, nothing else changes — and nobody calls a person on vacation, on any channel.
"Talk to my colleague Mária" — the most valuable reply in B2B. The system creates Mária as a new lead automatically, opens with "Peter pointed us to you…", and closes Peter's thread politely. A rejection becomes a warmer door.
"Unsubscribe / STOP" — honored immediately and automatically, no queue, no exceptions, fully logged.
An angry reply — stops all outreach to that whole company and goes straight to a human. Reputation first.

10How much control clients keep

The customer is defined by the mode, not by sizeWe don't build "for enterprise" or "for SMB" — we build for anyone with B2B leads. What defines a customer is how much of the wheel they want to hold. Same software in every case; only the seat occupancy changes.

DIYSelf-serveThe client's own team drives. They run campaigns, approve replies, read the dashboards. We provide the platform and get out of the way. The natural home for smaller teams and the hands-on type.

DWYDone-with-youWe set everything up — flows, compliance, senders, first campaigns — then hand over the keys. The client grows into driving it themselves, usually as the trust dial earns their confidence.

DFYDone-for-youOur ops team runs the client's workspace permanently — the O2-style engagement, priced as managed premium. The client just receives hot, prepared meetings; we operate everything behind them.

Inside the product, control is a dial, not a switch. In Copilot mode, every reply Klara drafts waits for one-click human approval. In Autopilot, routine replies go out on their own — but pricing questions and anything sensitive still wait for a human, always. And opt-outs execute instantly in every mode; that's not configurable.

The clever part is the trust dial: the system tracks how often Klara's drafts get approved without edits. When a category runs at 96% for a month, it suggests: "Enable Autopilot for routine replies?" Clients hand over control at the speed the data earns it — which is also exactly how a done-for-you client gradually becomes self-serve.

11Beyond the meeting — the CRM question

"Are you a CRM?" — No. And eventually that's the point.A classic CRM is a database of records that humans must feed. CloseRocket is a system of action where the record is a by-product of the work. Nobody ever wanted a CRM — they wanted customers; the CRM was the tax. Klara has been removing that tax from the start — she fills the CRM after every interaction — and now we finish the thought.

TODAYIntegrateConnect the client's HubSpot or Pipedrive once — hot handoffs then land there automatically, and outcomes captured by Klara flow back into it. There's setup (auth + which field maps to what), but it's done once. The point: nobody migrates their data or changes how they work — we slot in alongside the CRM they already have.

ALREADY TRUEStand-aloneA client without a CRM is fully served: the pipe is the record — accounts, people, every touch and decision, the complete audit trail — and our screens are what a small sales team actually wanted from a CRM, except they fill themselves in.

THE LONG GAMEDisplaceWhen 90% of sales movement lives here, the classic CRM becomes a shrinking attachment someone eventually doesn't renew. We don't fight CRMs; we make them optional.

The discipline that keeps "we're not a CRM" honest: we own movement — every touch, conversation, lifecycle decision, and the capture of what happened after the meeting. We don't build deal administration — pipeline stages, quotes, forecasting, contracts. The boundary is sharp, and it's what lets us integrate with CRMs today while quietly making them unnecessary tomorrow.

12Compliance as a selling point 11 countries verified

Not a PDF in a drawer — a gate in the productOutreach law is encoded per country as a rules registry, verified against primary legal sources for 11 countries (June 2026). Every lead carries its country and legal status; the system literally cannot send an SMS to someone without consent — the step skips itself and logs why. A new market = new rows in the registry, not new code. Serbia, our first non-EU row, proved it.

Country	B2B email (named person)	B2B cold calls	Central register
🇸🇰 Slovakia	Opt-in; published business contacts OK with opt-out	Opt-in + mandatory Robinson check + 0888 prefix	Robinson list (calls)	VERY STRICT
🇨🇿 Czechia	Opt-in; corporate addresses tolerated in practice	Opt-in via directory flag — default is "do not call"	None — opt-in default	VERY STRICT
🇵🇱 Poland	Opt-in, B2B incl. — corporate addresses closed (2024)	Opt-in; asking for consent by phone also banned	None — opt-in default	VERY STRICT
🇦🇹 Austria	Opt-in, no B2B exemption; covers social-media DMs	Opt-in, no B2B exemption	ECG-Liste (email)	VERY STRICT
🇩🇪 Germany	Opt-in, B2B = B2C (competitor lawsuits)	OK with documented presumed consent	None	SPLIT ✗/✓
🇭🇺 Hungary	Opt-in for named people, B2B included	Consent, separately per channel (NAIH)	None	STRICT
🇸🇮 Slovenia	Opt-in + customer exception	Automated: opt-in; live: likely opt-out	None central	STRICT
🇷🇴 Romania	Opt-in (express)	Live human calls arguable; automated: opt-in	None	STRICT
🇷🇸 Serbia	Consumers opt-in; B2B with local counsel	Consumers: opt-in + Ne zovi check	"Ne zovi" (RATEL)	STRICT
🇧🇬 Bulgaria	Opt-out after CPC register check	Consumers opt-in; B2B basis thin	CPC register (B2B email)	MEDIUM
🇬🇧 UK	Corporate subscribers: no consent needed	OK after TPS + CTPS screening	TPS + CTPS (calls)	EASIER

The engine plans differently per country: in Czechia and Poland a new-business play runs email/LinkedIn-first (cold calls effectively need consent); in Germany it runs phone-first (presumed consent with a documented reason); in Austria it's referrals and inbound; in Bulgaria email-first after a register check; in the UK, everything — screened. Same flow logic, different channel availability — automatic.
Five official registers get checked automatically before dispatch: SK Robinson (calls), AT ECG-Liste (email), UK TPS+CTPS (calls), BG CPC (B2B email), RS Ne zovi (consumer calls).
Imports require a signed-off legal basis before a single message goes out; every action lands in a tamper-proof audit log; on calls, Klara discloses she's recording and that she's an AI.
Existing customers stay reachable everywhere (the "similar products" exception exists in all 11 countries) — which is why upsell and cross-sell flows run even in the strictest markets. Slovakia adds a 2-year cap to model.

Why this is a moat: US competitors can't say any of this. When O2's procurement hears "your country's law is built into the dispatch path, here's the audit trail", the conversation changes from risk assessment to rollout planning.

13Competition & why we win

Who	Examples	What they can't do
US enterprise sales platforms	Outreach, Salesloft, Apollo	LinkedIn only as a manual to-do, zero per-country EU legal logic, $100+/seat, a tool — not a system that works for you
Multichannel senders	Lemlist, Skylead	They send; they don't read, decide, or remember. No reply intelligence, no lifecycle
LinkedIn tools	HeyReach, Expandi	One channel, no brain — we can even use them as delivery muscle
AI SDR agents	11x.ai, Artisan	The closest category — but they stop at the meeting, English-first, US-compliance, no European languages or per-country law

Our position in one line: the AI sales operating system for markets with real rules. The SDR agents end where we keep going — the meeting is one event in a loop we run on both sides. CEE is the beachhead because it's the hardest legal terrain; the registry architecture takes the same product anywhere, US included. Local languages, per-country law in the dispatch path, AI voice in the loop, and a lifecycle where leads compound instead of dying. Unit economics help too: LinkedIn at ~€4.5/account through the API vs $79–99/seat tools means margin lives with us, not with middleware.

Honest risks we manage: running LinkedIn ourselves means account-safety is now our responsibility (limits, warm-up and monitoring are built for exactly this); LinkedIn automation is a grey zone for every player in the market — enterprise legal will ask, and our answer is real human profiles, conservative limits, full disclosure; and we're a small team building a big engine — which is why everything ships in strict, testable stages.