Craftsperson at workbench assembling precision machinery, tools organized around them
All articles

Don't Buy an AI SDR. Build One.

Andrew Powers
Andrew Powers·· 5 min read

AI SDR tools charge $2,000/month to do what you can build for $50 in API costs. Same capabilities. You keep the control.

A client asked me last month if they should buy one of the new AI SDR tools — Artisan, 11x, Relevance AI. These are tools that promise to automate sales development: find prospects, research them, write personalized emails, send outreach.

“What’s the price?” I asked.

“$2,000 a month for 1,000 prospects.”

I showed him our n8n setup. Same capabilities. $47 in API costs last month. Same 1,000 prospects.

“Why would I pay 40x more for the same thing?”

He couldn’t answer. Neither could I.

What AI SDR Tools Actually Do

The pitch sounds magical: “Autonomous AI that finds prospects, researches them, writes personalized emails, and handles responses.”

The reality is four API calls in sequence:

  1. Enrichment — Pull company/contact data from Apollo, Clay, or similar
  2. Research — Scrape LinkedIn, news, job postings for personalization hooks
  3. Generation — Use GPT-4 or Claude to write personalized copy
  4. Sequencing — Send emails, track opens, handle replies

That’s it. Four API calls orchestrated in a workflow. None of this requires proprietary technology. None of it requires a $2,000/month platform.

The Architecture

Here’s what we deploy:

flowchart LR
    A([Signal]) --> B([Enrich]) --> C([Research]) --> D([Generate]) --> E([Review]) --> F([Send]) --> G([Monitor])
    classDef default fill:#dbeafe,stroke:#3b82f6,color:#1e40af,stroke-width:2px
StepToolWhat it does
SignalClay, CrunchbaseDetect intent (funding, hiring, tech change)
EnrichApolloPull company + contact data
ResearchProxycurl, PerplexityLinkedIn posts, news, job postings
GenerateClaude / GPT-4Write personalized email
ReviewSlack / AirtableHuman approval (optional)
SendInstantlyDeliver + sequence
MonitorSlack alertsTrack replies, errors

Seven nodes. One n8n workflow. Total monthly cost at 1,000 prospects:

ServiceMonthly Cost
Apollo (enrichment)~$20
Claude API (generation)~$15
Perplexity API (research)~$8
Instantly (sending)~$30
n8n (self-hosted)~$15 VPS
Total~$88

Compare that to $500-2,000/month for an AI SDR tool.

Start with Intent Signals

Don’t start with cold lists. Start with signals that indicate someone might actually want to hear from you:

SignalSourceWhy It Works
New fundingCrunchbase, PitchBookBudget unlocked, growth mode
Leadership hireLinkedInNew VP = new initiatives
Tech stack changeBuiltWith, WappalyzerEvaluating solutions
Job postingLinkedIn, IndeedScaling the team
Competitor mentionG2, socialActively comparing

The trigger matters more than the copy. An average email sent at the right moment beats a perfect email sent to a cold list.

n8n can poll these sources on a schedule, or you can pipe signals from Clay, Ocean.io, or your data warehouse.

The Enrichment Layer

Once you have a signal, enrich the contact:

Company data: Industry, size, revenue, tech stack (via BuiltWith), recent news, funding, hiring.

Contact data: Title, tenure, LinkedIn activity, previous companies.

Apollo handles most of this for $0.02-0.05 per contact. Clay goes deeper but costs more.

Don’t over-enrich. You need 5-6 data points for good personalization. More than that is waste.

The Research Layer

This is where AI SDR tools claim magic. It’s just API calls.

For each prospect, pull:

  • Their last 3 LinkedIn posts (via Proxycurl or PhantomBuster)
  • Their company’s recent news (via Perplexity or news API)
  • Their job posting keywords (if hiring)

Then summarize into a research brief:

Research Brief: Sarah Chen, VP Sales @ Acme Corp

Recent activity:

  • Posted about “scaling outbound without burning the team”
  • Company raised $15M Series A (TechCrunch, Jan 12)
  • Hiring 3 SDRs and 1 RevOps manager

Personalization hooks:

  • Shares concern about SDR burnout
  • Growth phase = pain points are acute
  • RevOps hire = open to automation

That brief feeds the generation step.

The Generation Layer

You don’t need a fine-tuned model. You don’t need prompt engineering courses. You need a solid template and the research context.

One warning: AI defaults to sounding like a butler. It uses phrases like “I hope this email finds you well” or “delighted to connect.”

That kills reply rates. Here’s the prompt structure that works:

Write a cold email to {{name}}, {{title}} at {{company}}.

Context: {{research_brief}}

Rules:

  • 3 sentences max
  • No greetings (“Hope this finds you well”)
  • No corporate speak (“leverage”, “synergy”, “delighted”)
  • Write like texting a busy colleague
  • One specific observation from their LinkedIn/news
  • One clear ask

Output only the email body. No subject line.

The output won’t be perfect. It’ll be 80% there. Which is why you start supervised.

Start with a Human in the Loop

For the first 100 emails, put a human in the loop.

Route drafts to a Slack channel. Your rep reviews, tweaks the tone if it sounds robotic, and clicks “Send.”

You’re not just sending emails — you’re training your eye. After 100 emails, you’ll see the patterns. You’ll know exactly how to tune the prompt.

Once drafts consistently hit your standards, take off the leash. Scale to 500, then 1,000.

The Sending Layer

Don’t send from your main domain.

SetupPurpose
Primary domainInbound, warm replies
Outbound domainCold emails (separate domain)
Sending toolInstantly, Smartlead, or Lemlist
WarmupRun for 2-3 weeks before sending

Instantly or Smartlead handle deliverability, warmup, and inbox rotation. $30-100/month depending on volume.

The n8n workflow pushes approved emails to their API. They handle sequencing, follow-ups, and reply detection.

Monitoring

You need to know when things break and when they work.

MetricAlert Threshold
Reply rate< 2% over 7 days
Bounce rate> 5%
Generation errors> 1%
Daily send volumeVariance > 30%

Pipe these to Slack. Review weekly. Adjust.

The whole point of building this yourself is you can see inside. You know exactly why something failed. You can fix it in minutes. (Just make sure you have proper error handling in place.)

Build vs Buy

FeatureAI SDR ToolDIY with n8n
Prospect identification✓ (via signals)
Data enrichment✓ (Apollo/Clay)
LinkedIn research✓ (Proxycurl)
AI-generated copy✓ (Claude/GPT)
Email sequencing✓ (Instantly)
Reply handling✓ (webhook)
Human reviewSometimes✓ (optional)
Monthly cost$500-2,000$50-100
Control over promptsLimitedFull
Data ownershipTheirsYours

AI SDR tools are convenient. They bundle everything. But you’re paying 10-20x for that convenience — and you lose control.

When to Buy vs Build

Buy an AI SDR tool if:

  • You have zero technical capacity
  • You need to launch in 48 hours
  • Budget doesn’t matter

Build with n8n if:

  • You have anyone who can configure APIs (this is the GTM Engineer profile)
  • You want to iterate on prompts and flows
  • You care about cost at scale
  • You want to own your data and process

Most companies reading this should build.

The 3-Day Build Plan

Day 1: Data flow

  • Set up Apollo or Clay integration
  • Configure trigger (signal source or list upload)
  • Test enrichment on 10 prospects

Day 2: AI layer

  • Connect Claude or OpenAI
  • Build prompt template
  • Add research layer (LinkedIn + news)
  • Test generation on 20 prospects

Day 3: Sending layer

  • Set up Instantly or Smartlead
  • Configure review queue in Slack/Airtable
  • Connect n8n to sending API
  • Send first batch (start with 25/day)

By day 4, you have a working AI SDR. Same capabilities as tools charging $2,000/month. Full control. Full visibility. A fraction of the cost. (For the LinkedIn-specific playbook, see our outreach guide.)

Want to skip the 3-day build? OpenClaw does prospect research, personalized outreach, follow-ups, and booking — as a hosted AI agent. No workflows to maintain. Compare hosting options.