• 151

    Projects submitted

  • 109

    Published projects

  • 47

    Comments left

  • 66

    Number of votes

  • 52

    days until voting ends

← Back to list

ZeON: AI Sales Agent for Multichannel B2B Lead Generation

  • Project manager on customer side

    Konstantin Kulenkov

    Kulenkov Group

    CEO & Founder, ZeON Product Owner

  • Category

  • Nomination

  • Goals

    Build an AI-driven sales engine that turns cold B2B prospecting into a predictable, measurable pipeline of qualified meetings with decision-makers.

    Key goals:

    • One North Star metric: 10+ qualified calls per operator per month. Every like, comment, connection and message is only a step toward that call.

    • Replace manual research and outreach of a full SDR team with a disciplined AI agent that works every day and forgets nothing.

    • Personalisation at scale: every touch is relevant, written in the voice of a specific sales leader and backed by real facts. No spam.

    • Full transparency: one source of truth for every company, a live funnel, KPIs and AI cost per text, in tenge and dollars.

    • Zero reputational risk: safety limits as code, human approval for every outgoing message, no CAPTCHA bypass.

    • Turn the internal tool into a multi-tenant product that other B2B companies can launch in days.

  • Timeline

    may, 2026 — october, 2026
  • Project completion year

    2026

  • Project scale

    12 automated workplaces
  • Results

    • 2,909 companies and about 4,400 decision-makers in a single source of truth, with 1,325 structured company dossiers.

    • About 10,000 sales interactions planned, executed and logged, with anti-repeat protection across channels.

    • A Kazakhstan market base of 5,250 companies, AI-segmented into hot, warm and reserve tiers.

    • 792 companies from a client's database scored automatically for $4.15 in total, about half a US cent per company.

    • 12 operator workplaces live across Kulenkov Group and external client companies, onboarded through the self-service ZeON 4.0 wizard.

    • Paper plans replaced by verified plans: only verified, executable actions reach the operator.

    • A daily routine cut from hours of manual research to a 1–2 minute pre-flight check plus one-tap execution.

    • AI cost per text tracked daily in tenge and dollars.

    • The European campaign (DACH, Benelux, UK) and the Kazakhstan launch ran on one platform, with a seamless strategic pivot.

  • Project uniqueness

    ZeON is built on one principle: "Ready = True". Most tools go green when a field is filled in; ZeON goes green only when every action is proven executable. Ten automated controllers check each step of the daily plan and must show evidence of what they rejected and why. The contact must be warmed up, the profile alive, the post read and identified, the link reachable and the tone right. A plan of about 60 actions on paper shrinks to 10–19 that are truly ready, and each of these can be trusted.

    Other differentiators:

    • Warm-first methodology: no cold pitch, and services or prices are never mentioned in the first message.

    • The voice of a real leader: each operator has an approved signature, role and tone of voice.

    • 7 specialised AI agents on a shared RAG knowledge base.

    • Dual scoring: ICP fit × readiness to talk.

    • A Value Bank of 442 verified insights. Every fact the AI uses must have a source.

    • Human-in-the-loop by design: AI prepares, a person approves with one tap.

  • Software used

    • AI: Anthropic Claude models (Opus, Sonnet), including the Batch API for overnight generation. They power 7 agent skills (Scout, Evaluator, Connector, Negotiator, Content Strategist, Keeper, Analyst) and multi-agent workflows with AI judges.

    • Backend: Node.js 20, Express 4 REST API. Data pipelines in Python 3.9+.

    • Frontend: single-page web app (HTML/CSS/JS), mobile-first: operator cockpit, client cabinet, onboarding wizard, funnel dashboard.

    • Data: Markdown company cards with YAML metadata plus JSON event logs. Fully auditable, with backups and a changelog.

    • Knowledge: a local RAG base of about 218 documents in 26 categories, covering voice, pains, cases and ICP.

    • Channels: LinkedIn (read-only research plus the official API for own posts), e-mail (IMAP collector), WhatsApp (a self-hosted reader service) and Telegram.

    • Infrastructure: Linux VPS, Docker, Caddy (HTTPS), systemd timers for nightly autonomous runs, self-hosted GitLab, and separate dev and prod environments.

  • Solution from the Global CIO catalog

    The project does not use solutions from the Global CIO catalog

  • Project implementation complexity

    • Trust in AI output. LLMs hallucinate; in sales one invented fact destroys trust. We introduced "facts only with a source" and text gates. For example, AI cannot thank someone for a reply that never happened. A humanity score and a daily quality line benchmark every text.

    • Account safety. In LinkedIn outreach the account is the single point of failure. Limits, warm-up rules and read-only research are enforced in code.

    • "Green but false" reports. The first version reported a ready plan that broke on execution. We rebuilt every check into an evidence-based controller and proved each one on deliberately bad data.

    • Multichannel orchestration. LinkedIn, e-mail, WhatsApp and Telegram share one memory and one anti-repeat logic.

    • Multi-tenancy. Strict isolation of operators, mailboxes and WhatsApp sessions between client companies.

    • A strategic pivot from Europe to Kazakhstan mid-project without losing data or stopping the daily rhythm.

  • Description

    INITIAL SITUATION

    Kulenkov Group builds AI automation of document workflows for logistics and distribution, a strong product with proven cases such as cutting shipment processing from 30 to 3 minutes. Its growth was limited by sales. Finding the right decision-makers, researching hundreds of companies and writing relevant messages by hand was slow and expensive. It was also risky: one careless campaign can burn an account and a brand. Generic outreach tools produced volume, not conversations.

    IMPLEMENTED SOLUTION

    We built ZeON, an AI sales agent that runs the whole path from "we know nothing about this company" to "the decision-maker agreed to a call":

    1. Find: import and screening of company databases, plus activity checks of decision-makers.

    2. Evaluate: structured dossiers and dual AI scoring (ICP fit × intent).

    3. Warm up: likes and meaningful comments on real, freshly read posts, plus follows.

    4. Connect: requests only to people who were actually warmed up.

    5. First message in the leader's voice: a warm intro, a specific detail about their business, no premature selling.

    6. Dialogue to call: reply classification, follow-ups and handover to a specialist.

    Every night the platform builds a daily plan per operator, with anti-repeat rules, safety limits, territory split and a pre-flight mobile view. The operator reviews the day in 1–2 minutes and executes it with one tap per action.

    IMPLEMENTATION SPECIFICS

    • May–June 2026: core pipeline, 7 AI agents, knowledge base, dashboard. The ZeON AI program was formalised as proprietary software, and production launched on a VPS in June.

    • June–August: European campaign (DACH, Benelux, UK) on LinkedIn and e-mail, with 1,349 companies processed. It produced the "Ready = True" verification layer: 10 controllers, each tested on deliberately bad examples.

    • August: strategic pivot to Kazakhstan, with WhatsApp, Telegram and e-mail added. Then ZeON 4.0: a multi-tenant platform with a self-service onboarding wizard (9 mandatory and 7 optional steps), client cabinets, WhatsApp QR connection inside the cabinet and an isolated mailbox per operator.

    • September–October: an incoming-database flow (a parser delivers a client's list, AI scores it automatically), the Value Bank, and a daily quality line tracking cost, humanity and factual accuracy of every text.

    Engineering discipline: 540+ logged architecture decisions, 3,000+ changelog records, negative tests for every gate and an automated "doctor" health check before each release.

    CHANGES AFTER IMPLEMENTATION

    Sales prospecting became an industrial, transparent process. Decision-maker research, scoring and message drafting happen overnight without people. Operators spend minutes, not hours, on a day's outreach and see the honest funnel at any moment. The company gained a new product: B2B companies in Kazakhstan now run their own sales outreach on ZeON.

  • Project geography

    ZeON was designed from day one as a cross-border platform.

    • Kazakhstan, from July 2026 and the main market since late August: a nationwide base of 5,250 companies, with operators from several client companies working Kazakhstan's B2B market through WhatsApp, Telegram, e-mail and LinkedIn in Russian.

    • Europe, June–August 2026: a LinkedIn and e-mail campaign to logistics and distribution decision-makers in Germany, Austria, Switzerland, the Netherlands, Belgium and the United Kingdom, in English and with territory-based routing between operators. 1,349 European companies were processed.

    • Architecture: the system runs on its own VPS, with the time zone, language, channels and tone of voice configured per operator. Territories (zones) are routed automatically, so British contacts go to one operator and DACH and Benelux contacts to another.

    The platform can be replicated in any country with B2B sales. A new client company is onboarded through a self-service wizard in days.

Comment

Only authorized users can comment.
We suggest you or register.

  • Customer

    Kulenkov Group

We use cookies for analytical purposes and to deliver you the best experience with our website. Continuing to the site, you agree to the Cookie Policy.