Deployment Blueprint: A BDR Agent Team That Runs Top-of-Funnel for Any Industry

This blueprint is an anonymized reference architecture generalized from BDR agent teams we have built and operated for several software companies, and now rebuild for clients on infrastructure they own. No client is named. Prompts, scoring rules and message templates are deliberately omitted.

What the team does

A BDR (business development representative) agent team runs a company's top-of-funnel sales motion end to end, the way a small team of human reps would — but continuously. Given a target list or an inbound lead, it researches the prospect and company, picks the relevant narrative and case studies, drafts and sends a multi-touch sequence across email (and, in some deployments, LinkedIn or WhatsApp), watches the shared inbox for replies, categorizes them, answers questions and objections from the company's own knowledge base, checks the sales team's real calendar, books the meeting once a time is agreed, and keeps the CRM accurate throughout.

A human stays in the loop at the two moments that matter: before a message reaches a prospect, and before a demo lands on a rep's calendar. Both are configurable per deployment rather than hard-coded. Everything else — CRM hygiene, research, drafting, categorization — runs unattended by default.

Team structure

The system is an orchestrator with specialist sub-agents, each calling a small library of single-purpose tools. The orchestrator owns the CRM-of-record actions and the two gated outward actions; it delegates judgment-heavy work to specialists rather than doing everything in one prompt.

Role	Responsibility	Hands off to
BDR orchestrator	Owns CRM task and note lifecycle, contact updates, owner assignment, and the two human-gated actions (send email, create calendar event)	All specialists
Inbox manager	Watches the shared inbox, categorizes each new email, decides whether a reply is needed	Reply specialist, or a hand-to-human path
Sales strategist	Segments the prospect, pulls the right case studies and value proposition, composes the sequence	Prospect researcher, calendar checker
Prospect researcher	Gathers public signal on the person and company — website, LinkedIn, reviews, web search	Sales strategist
Reply specialist	Runs on every inbound reply; extracts questions and objections, flags escalation, drafts the answer	Calendar checker, solution finder
Follow-up manager	Cadence logic for prospects who have not replied; decides which touch is due and drafts it	Orchestrator (send)
Calendar availability checker	Cross-references the prospect's timezone with the team's real calendar; proposes only genuinely open slots	Orchestrator (create event)
Solution finder	Retrieval agent answering prospect questions from the company's FAQ and documentation	Reply specialist
Website-visitor and pre-demo qualifiers	Qualify inbound visitors and demo requests on company size, role, challenge and scope before they enter a sequence	Demo outreach manager

flowchart TB
  subgraph triggers["Triggers"]
    T1["Inbound email"]
    T2["Prospect list"]
    T3["Website / chat visitor"]
    T4["Book-a-demo form"]
  end
  IM["Inbox manager"]
  BDR["BDR orchestrator"]
  SS["Sales strategist"]
  PR["Prospect researcher"]
  MRS["Reply specialist"]
  FUM["Follow-up manager"]
  CAL["Calendar availability checker"]
  SF["Solution finder (knowledge base)"]
  Q["Visitor / pre-demo qualifiers"]
  T1 --> IM
  IM -->|reply needed| MRS
  IM -->|meeting proposed| CAL
  IM -->|unclear| BDR
  T2 --> BDR --> SS --> PR
  SS -->|sequence| BDR
  T3 --> Q --> SS
  T4 --> Q --> CAL
  MRS --> SF
  MRS --> CAL
  MRS --> BDR
  FUM --> BDR
  BDR -->|never-ask| CRM[("CRM: tasks, notes, contact fields")]
  BDR -->|approval-gated| MAIL[("Email: send")]
  BDR -->|approval-gated| EVT[("Calendar: create event")]
  CAL --> CALSYS[("Calendar: free/busy")]
  PR --> ENRICH[("Enrichment: LinkedIn, company data, web search")]
  SF --> KB[("Company knowledge base")]
  BDR -->|channel variants| SEQ[("Sequencer / LinkedIn / WhatsApp")]

The lifecycle of one lead

1. Entry. A bulk prospect list for outbound, a book-a-demo form, an inbound reply to a prior touch, or a qualified website visitor.
2. Enrichment and research. The researcher (outbound) or the demo-enrichment tool (inbound) pulls company and person data into a structured profile.
3. Qualification. Demo requests are scored against challenge, scope and company size and routed, nurtured or disqualified. Outbound prospects are segmented and matched to a value proposition and case studies.
4. Sequence composed. A personalized first touch and follow-up cadence, checked against what has already been sent so a sequence never restarts mid-flight.
5. Send — human gate one. Sending runs under a per-deployment approval mode: fully autonomous, approve every message, or approve the first touch and auto-send the rest. A configuration switch, not a different build.
6. CRM hygiene — always automatic. After any send, a follow-up task is created; on reply it is completed. This bookkeeping runs unattended in every deployment because it is too foundational to gate.
7. Reply handling. The inbox manager categorizes and routes. The reply specialist extracts objections, grounds an answer in the knowledge base through the solution finder, and decides whether to answer or escalate.
8. Availability check. Before proposing or confirming a time, the checker reads the team's real calendar and the prospect's likely timezone.
9. Book — human gate two. Creating the event is the second gated action, often set stricter than sending: a meeting on a real calendar is higher stakes than a draft.
10. No reply. The follow-up manager computes the next due touch, accounting for out-of-office replies and prior escalations, and hands it back under the same gate.
11. Loop or close. The lead cycles until it books, goes cold, or a human takes over, with the CRM record carrying the full history.

How the same team was adapted across companies

The base template has been re-parented many times, each copy adapted for one company's product, ideal customer profile and tooling rather than rewritten:

• A recruiting-software company and a meeting-notes startup each cloned the full sub-agent family under their own brand with their own case-study and value-proposition tools — the same topology stamped onto unrelated verticals.
• A video-creation SaaS ran a lighter build with fewer research tools and a shorter follow-up cadence, suited to a simpler buyer.
• Two companies kept a sandbox copy beside production, a build-test-promote workflow per client.
• One deployment swapped the default email provider for a transactional email service without touching the rest of the graph; others routed the last mile through LinkedIn, WhatsApp, or third-party sequencers.
• A fan-out tool runs a large list through many parallel copies of the orchestrator instead of serializing every lead through one conversation.

What changes per industry, and what does not

Changes: the ideal customer profile and qualification thresholds; which research and enrichment sources are worth querying; the objection and FAQ library; the case studies and value propositions; the delivery channel and booking flow; and the approval mode.

Fixed: the division of labour; the eleven-step lifecycle; the never-ask versus approval-gated split between internal bookkeeping and outward-facing actions; and the rule that every proposed time passes through the availability checker.

Industry	What the inputs become
Construction	ICP narrows to project type and licensing or bonding status; research shifts to permit and project-award data; objections centre on timeline and subcontractor trust
Law	Qualification adds practice-area fit and conflict screening before outreach; approval mode runs stricter given the cost of a misfired message to a referral source
Healthcare	Research sources are constrained by what data is permissible to touch; the human gate widens to reply drafting, not just send and book
Financial services	Compliance disclosures join the knowledge base; stricter audit-trail requirements, which the always-on CRM layer already supports
Manufacturing	Research leans on trade-show, RFQ and distributor signal; cadences run longer — a follow-up configuration change, not a structural one

Why this belongs on infrastructure you own

Across one lead's lifecycle the team handles contact details, full email threads, CRM engagement history, the sales team's free/busy calendar and third-party enrichment data — live correspondence plus CRM plus calendar, flowing continuously through a language model. That is exactly the data a company least wants leaving its infrastructure, and exactly what a platform vendor would most like to learn from.

A self-hosted rebuild needs, as capabilities rather than code: an orchestration layer that calls sub-agents and tools with shared context; a per-action approval gate that flips per deployment without forking the system; connectors to the CRM, an email provider and a calendar provider; swappable enrichment connectors; a retrieval layer over the company's own knowledge, kept separate from copy generation so answers stay grounded; a cadence engine tracking what has been sent per lead; an audit trail of every CRM write and every send or booking; and a fan-out mechanism for batch runs. We deploy it on GLM-5.3-Flash behind self-hosted agent frameworks — see OpenClaw 2.0 vs Hermes Agent — as part of the agent teams solution, typically alongside a lifecycle agent that takes over once the meeting is booked.

Kept private

This blueprint omits the system prompts and internal rules that drive each agent's judgment, the lead-scoring formulas and qualification thresholds, the outreach and follow-up templates, and the identities of the companies the template was built for. What is described is the architecture: roles, hand-offs, tool topology, and the autonomy design.