Sales Automation & the AI SDR Playbook

Sales automation is any software that takes repetitive, manual work out of the sales motion: building prospect lists, finding and verifying contact details, sequencing multi-touch outreach, personalizing messages, chasing follow-ups, handling replies, booking meetings, and keeping the CRM honest. It spans a spectrum — from a simple email sequencer you feed lists into, to a full AI SDR that runs the entire pipeline itself.

Here is the uncomfortable truth the category glosses over: writing and sending are commoditized. Every tool can now produce a passable personalized email and deliver it on schedule. What separates automation that books meetings from automation that burns domains is targeting — who gets the message, chosen from live signals rather than a static list. That is why most of this guide is about targeting, and why the tool matters less than the list it works.

Think of Sales Automation as a pipeline that runs on autopilot (an hourly engine) or on demand: Discover → Qualify → Enrich → Validate → Research → Write → Send → Reply → Book. Each step feeds the next, and everything streams into a live activity log so you can watch it work.

Discover finds candidates from two sources in parallel (below). Qualify scores every company and person 0–100 for ICP fit and auto-skips anything off-target before spending a cent enriching it. Enrich is internal-first and needs no external vendor — our own people database plus each company’s inferred email pattern, preferring a verified address. (Connect your own Hunter key and it will also verify a pattern-guessed address before anything auto-sends.) Validate makes sure only deliverable emails auto-send. Research pulls the prospect’s recent LinkedIn activity and company news to find a real hook. Write composes a short, human, fact-checked message grounded in your business. Send respects your mode and daily caps. Reply is handled by an inbox agent that answers questions and steers to a call. Book drops the meeting into your calendar.

Everything is cached — companies, contacts, and research are stored and reused — so repeat runs are fast and cheap, and coverage compounds over days rather than being judged on a single run.

First, fill your business context in Settings → Sales — this is the whole game. Three plain-English fields: what the business does, what you sell (the concrete offer and the outcome it delivers), and who you sell to (your ideal customer). Two optional-but-powerful fields — dream accounts and who NOT to target — make your targeting reliable (see the targeting section).

Second, connect the tools the engine uses under Connections: Hunter (optional, bring-your-own-key, and used only to verify an address before it auto-sends — companies come from BusinessMCP’s own 34M-company database, no data vendor needed), Apify (LinkedIn engagement signals + Google-Maps), Tavily (research), your own LLM key if you want frontier-quality writing (otherwise the free tier is used), plus a verified sending domain (Resend) or Gmail for outbound and a calendar (Google/Microsoft/Apple) for booking.

Third, create a campaign with "✦ New campaign" — the AI researches your business and proposes distinct audiences, each sized against a real company database; you pick one or more, preview sample companies, then choose a cadence and a mode: Manual (drafts only), Assisted (you approve every send), or Autonomous (auto-send). Start on Assisted for a few days, read the drafts, then graduate to Autonomous once you trust them.

ICP company discovery builds a target list from your ideal customer: companies that already visited your own site (hottest) → our own 10M+ company database (the primary source) → fresh web-research and Google-Maps when the database runs thin for a niche. It fit-scores each company, keeps the good ones, and finds the buying committee.

LinkedIn intent signals are the sharper, higher-converting source. Add influencer or company profiles — or specific posts — under Add prospects → LinkedIn intent signals. The engine scrapes the people who liked or commented (that engagement is a hand-raise), qualifies them against your ICP, and turns them into prospects. This finds people who are actively interested in your topic right now, which beats any static firmographic list.

Manual drafts and you send from the console; Assisted drafts and you one-click approve; Autonomous sends on its own. Whatever you choose, deliverability is protected automatically: in Autonomous mode only verified (valid) emails auto-send — anything catch-all or unknown is held as a draft for you to review, so you never torch your sending domain’s reputation. Every email carries a compliant footer with sender identity and one-click unsubscribe, as CAN-SPAM requires.

Best practice: use a separate sending domain for cold outreach (not your main company domain), so heavy volume never risks the reputation of the domain your team relies on.

The single most common mistake is selling to people who look like you instead of people who feel the pain. Your ideal customer is defined by the problem you remove, not by how similar they are to you.

Take BusinessMCP as the example. It does data unification, marketing automation, and MCP-gateway management. The tempting-but-wrong instinct is to target AI startups — "because they get it." That is exactly backwards. AI-native startups already stitch their own data, already run their own automations, and would rather build than buy: they get it and therefore don’t need you. The right buyer is a non-technical enterprise drowning in disconnected data with no idea how to implement AI — a mid-market manufacturer, insurer, clinic group, or distributor whose Stripe, ads, CRM, and spreadsheets don’t talk to each other, whose leadership keeps hearing "you need AI" but has nobody who can wire it up. They feel the pain acutely and can’t solve it themselves. That is the buyer.

So the ICP flips. Who we sell to: "Non-technical mid-market and enterprise companies (100–2,000 staff) with fragmented business data and no in-house AI/data team, whose leadership wants AI-driven insight but has no way to implement it." Dream accounts: name a few real such companies — the engine finds lookalikes of them. Who NOT to target: "AI startups, AI-native software companies, dev-tool vendors, agencies that resell this" — these get hard-excluded so discovery never wastes spend on them. Competitors are excluded automatically (you want their customers, not the vendors).

How to express it, and how the AI uses it: you write plain English in Settings → Sales. From that the engine derives the structured ICP — industries, buyer titles, seniority, company-size band — and enforces your dream accounts and exclusions directly. Rules of thumb: lead with the pain and the situation, not your features ("companies where X is broken and nobody can fix it" beats "companies interested in AI"). Name 3–8 real dream accounts — concrete examples teach the engine the shape of a good customer far better than adjectives. Use the exclusions field aggressively — it is the number-one lever; listing "AI startups, agencies, dev-tool vendors" guarantees they are filtered out. Match the buyer title to who actually feels the pain and can sign — for "enterprise with no AI clue" that is often a COO, VP Operations, or Head of Revenue, not a CTO who would just build it. And right-size the company: too small means no budget or pain, too big means procurement gauntlets — mid-market is usually the sweet spot.

For LinkedIn signals, choose sources the same way: monitor the creators your buyer follows, not the ones you follow. For BusinessMCP that is operations, finance, and industry-leadership voices talking about "we’re behind on AI / our data is a mess," not the AI-engineering crowd. The people engaging with "how non-technical companies should adopt AI" are your buyers; the people engaging with "new vector DB benchmark" are not.

Replies are caught across email and LinkedIn/X. An inbox agent answers questions and objections on-context and pushes toward a call — auto-sending in Autonomous mode, or drafting for your approval otherwise. A question keeps the contact a live lead; it is never mistaken for a rejection.

An interested reply auto-books: your personalized calendar link goes out on the same channel and the meeting lands in your calendar. Prospects that go quiet for about 45 days get one fresh-angle win-back automatically — the "circle back months later" motion, without you having to remember. And the CRM pipeline is a real control surface: drag a card to Churned to pause its sequence, or back to Lead to re-queue it.

The market splits into four rough categories. Sequencers (Instantly, Smartlead, Outreach, Salesloft) send and schedule — you bring the lists and the copy. Data and enrichment platforms (Apollo, Clay, People Data Labs, Hunter) find companies, people, and emails — you assemble the workflow. LinkedIn automation tools handle the social side, with the account-safety tradeoffs covered in our LinkedIn guide. And integrated AI SDRs — BusinessMCP among them, alongside tools like 11x and Artisan — run discovery through booking as one pipeline.

Which is right depends on your team, honestly stated. If you have SDRs and a proven playbook, a sequencer plus a Clay-style data stack gives you the most control. If you are founder-led and nobody has time to glue five tools together, an integrated AI SDR wins on time-to-pipeline. Whatever you pick, judge it on guardrails, not features: email validation before sending, per-domain volume caps, suppression lists, and a human-approval mode are what keep the channel alive.

• Sharpen "who you sell to" around the pain, and load up the exclusions — that is your biggest lever.
• Name 3–8 real dream accounts that are ideal customers.
• Add several LinkedIn signal sources your buyer engages with — usually your highest-converting source.
• Connect a verified sending domain (separate from your main domain) and a calendar.
• Run Assisted first, read 10–20 drafts, tune the context, then switch to Autonomous.
• Fill "what you sell" precisely — if it is blank, messages go generic.
• Let it run: caching plus the hourly engine mean coverage compounds over days.