Replace the "fill out the form and wait three days" inbound queue with a buying agent that answers, qualifies and recommends a plan in minutes, then hands complex deals to a rep with the full conversation attached. Click through the build, then take the operator playbook.
Based on Anthropic's How Anthropic's sales team rebuilt inbound with Claude Managed Agents. Walkthrough screens are an illustrative JayOh recreation with fictional data, not the live product.
Where the form-and-queue model breaks and which inbound questions an agent should own outright.
The three endings every buying-agent conversation should have: checkout, a rep, or an answer.
What you build vs. what Managed Agents runs: prompt, knowledge base and tools on your side; loop, sessions, hosting and versions on theirs.
How JayOh wires it into HubSpot or Salesforce so every hand-off lands as a qualified, routed, measurable lead.
Use Next / Back or your arrow keys. Orange boxes point to what matters on each screen. Illustrative recreation with fictional companies (Brightline Software and its buyers).
Tip: tap the expand button (top right of the guide) and turn your phone sideways for the full screen.
Put a Claude Managed Agents buying agent on your pricing page, contact-sales page, product and emails. It answers pricing, security and fit questions immediately, qualifies the buyer, recommends the right plan and seat count, and ends every conversation in one of three ways: self-serve checkout, a hand-off to a rep with the full transcript and the reason, or a quick answer. The outcome: speed-to-lead measured in seconds, reps who only work educated buyers, and a CRM record for every conversation instead of a queue nobody clears. It's the agent version of our speed-to-lead work.
Head of Sales Development or RevOps owns outcomes and the prompt. One engineer owns the code and tools. Sales and content leads edit the prompt and knowledge base. Sales leadership owns the escalation rules.
Claude Managed Agents (beta) + Claude Console, your website chat widget, CRM (HubSpot or Salesforce) for leads and routing, billing/checkout (e.g. Stripe or HubSpot Commerce), Slack for rep alerts.
Claude API org with Managed Agents access, an approved pricing & security answer set (each answer with a source), a clean lead object with owner routing, a checkout or quote path that works without a human, and a defined escalation policy.
recommend_plan, show_proposal, send_to_checkout, handoff_to_rep. The hand-off tool requires a reason, a summary and the transcript. Done = each tool returns real data from your pricing and CRM APIs in staging.We treat the agent as a new lead source with an SLA, not a website widget. In HubSpot we add properties for Agent conversation ID, Agent outcome, Escalation reason, Recommended plan and Recommended seats, a lead source value of "Buying agent", round-robin or territory routing on hand-off, and a one-business-hour follow-up task. In Salesforce the same fields live on Lead with an assignment rule and a Flow for the task. Guardrails: the agent can quote list price only, discounts and non-standard terms always escalate, regulated or legal questions cite the source or escalate, and every session is logged. QA: 20 scripted buyer personas run against staging before each release, plus a weekly sample of 25 live transcripts scored for accuracy and plan fit.
| Metric | Type | Weak | Acceptable | Strong |
|---|---|---|---|---|
| Time to first answer | Leading | > 1 hour | < 5 min | < 1 min, 24/7 |
| Hand-offs worked within SLA (1 business hour) | Leading | < 60% | 60–85% | > 85% |
| Share of conversations escalated to a rep | Leading | > 40% | 20–40% | < 20% and falling |
| Answer accuracy on weekly transcript sample | Leading | < 90% | 90–97% | > 97% |
| Hand-off lead → opportunity rate vs. old form leads | Lagging | ≤ 1x | 1.3–2x | > 2x |
| Sales cycle on agent-assisted deals vs. baseline | Lagging | No change | 2–4 days faster | 5+ days faster |
Ranges are JayOh working benchmarks for B2B SaaS inbound, set against the improvements Anthropic reports (2x+ conversion to opportunity, ~5 days faster close). Calibrate to your own baseline in the first 30 days.
Building more agent plays on your CRM? See the HubSpot Agent Playbook, the AI-enabled SDR model and the AI agent governance guide.
A 400-line prompt that encodes every BANT branch. It gets brittle, contradicts itself and still escalates the easy stuff.
One goal paragraph, sourced answers, and four tools. Iterate weekly from transcripts.
The rep gets an email address and "wants to talk to sales," then re-asks every question the buyer already answered.
Summary, plan, seats, reason and transcript are written to CRM properties, with an owner and a 1-hour task.
Pushing every buyer to Enterprise. Small teams churn or stall in procurement they didn't need.
The agent recommends the plan that fits. Buyers who land on the right plan stay longer.
A well-meaning wording change breaks pricing answers for a weekend.
Every change is a version, tested in staging by SMEs, promoted on purpose, and reversible in one click.
It's a Claude-powered agent that sits where buyers already ask questions (pricing page, contact-sales page, product, email), answers pricing, security and fit questions, recommends a plan and seat count, and ends in checkout, a rep hand-off with full context, or a quick answer. Managed Agents hosts it and runs the agent loop, sessions and versioning.
No. It removes the queue work: answering questions the docs already cover and chasing buyers for basics. Reps spend their time on hand-offs where a person changes the outcome, plus events and earlier-stage education.
Three things: the prompt (a goal, not a rulebook), a sourced knowledge base of approved answers, and a handful of tools such as recommend plan, show proposal, send to checkout and hand off to a rep. Managed Agents runs the loop, sessions, hosting, Console and versions.
The hand-off tool calls your CRM API: it creates or updates the contact and lead, writes the summary, recommended plan, seats and escalation reason to properties, assigns an owner and creates a follow-up task. JayOh builds this for HubSpot and Salesforce.
Leading: time to first answer, escalation share, hand-offs worked within SLA and answer accuracy. Lagging: hand-off lead-to-opportunity rate vs. old form leads and sales-cycle length on agent-assisted deals.
Anthropic reports one engineer built the first version in a few weeks. With JayOh, plan on a 30-day pilot: week 1 knowledge base and tools, week 2 CRM hand-off and staging, week 3 opt-in launch on the pricing page, week 4 readout.
JayOh designs the agent, wires the hand-off into HubSpot or Salesforce, and runs the 30-day pilot with you.
We engineer growth systems. JayOh builds the RevOps architecture, GTM execution and data infrastructure that turn messy revenue operations into predictable pipeline for B2B SaaS and PE-backed teams.
