Field notes · Playbook · July 2026
How to do outbound without hiring SDRs: the GTM engineering playbook
An SDR does four jobs: find who to contact, find a reason to contact them, write the message, and keep the machine running. None of these four requires headcount anymore; what they require is an engine with a quality gate. The one-line version of this whole page:
You do not need SDR headcount to run outbound. You need a signal source, a research agent, a QA gate, boring deliverability, and about 15 good sends a day.
The five-part engine
Signals: a real reason to write
Most cold email fails at the reason, not the writing. A signal is a timely, verifiable event that makes your offer relevant this week: a job post for a role your product makes unnecessary, a funding round, a stack change, a new regulation, a leadership hire. Static firmographics (“B2B SaaS, 50-200 employees”) are a filter, not a reason.
Practical rule we hold every build to: if the first line of the email could have been written last quarter, it is not a signal.
Research agents: the SDR's lost hours
This is where AI actually replaces labor. A research agent works a fixed checklist per account: what the company does, what changed recently, what the specific buyer has said publicly, which trigger applies. The non-negotiable: every claim needs a verbatim quote and a source URL, and anything unevidenced is auto-killed. Without the quote-or-kill rule, AI research fabricates confidently.
A cloud infrastructure client of ours cut account research time by 70% with this pattern; the win is not that the AI is smarter than an SDR, it is that it does the two-hour checklist in minutes, every time, without skipping steps.
Personalization behind a QA gate
An LLM drafts the email from the verified research only, in a voice document you actually maintain. Then a second layer, human or automated, checks the draft against the evidence before anything sends. In our published 51-run bake-off of four AI models on a real pipeline, the QA layer caught a reversed fact before it reached a real inbox. Reply rates follow research depth and QA discipline, not the model brand.
Sending infrastructure: boring on purpose
Warmed secondary domains, correct SPF/DKIM/DMARC, staggered sends, and volume that stays low. We cap our own engine at 15 to 20 sends a day per sender. The math still works: at that volume with signal-led targeting, a single-digit reply rate produces conversations every week. Blasting 500 a day produces a burned domain.
The learning loop
Once a week, the engine reads its own results: which signals produced replies, which personas ignored you, which phrasing drew the “how did you know that?” response. Targeting and voice get updated from evidence. This is the part rented tools do worst and the reason we build it as code the client keeps.
Play it: what does renting actually add up to?
And on the day you stop paying, the system is gone. A custom build is a one-time project: the agents, the data, and the code stay yours. That is the trade this whole page is about.
Cost context: verified, sourced pricing for every major vendor lives in our AI SDR Pricing Index, updated monthly.
When this is a bad idea
- Named-account enterprise sales: if ten relationships decide your year, a human beats any engine. Use the research agents, skip the automation.
- Tiny ICP: a few hundred qualified accounts saturate in weeks at any volume. The engine just gets you rejected faster.
- No internal owner: an engine nobody owns decays like any unowned system. Someone on your team must hold the learning loop, even if they never wrote code.
Three ways to get the engine
Rent it as an AI SDR platform, have an agency assemble it from tools like Clay and Smartlead, or commission it as custom code you own. The honest comparison of all three, including our own trade-offs, is in Best AI outbound agencies in 2026. The short version: standard motion and ops capacity in-house, rent; speed on rented tools, agency; weird market, non-obvious signals, or you want the asset, build.
Two questions separate builders from resellers, whoever you evaluate: “who owns the code when we part ways?” and “show me the QA layer that stops a bad email before it sends.”
Want to see a research agent work on something real?
Our free conference ROI calculator is a live demo of the same research agents described in step 02: paste any event URL and AI digs up the real costs and gives you a verdict in about 30 seconds.
Written by Amit Ben Dror, founder of Altitude (ex-Google Cloud SDR, $5M+ pipeline built by hand before automating any of it). Client outcomes referenced above are from real Altitude engagements; vendor pricing lives in the Pricing Index with sources and dates. Yes, we sell the custom-build option. That is why the bad-idea section exists.