Building a Sales Cadence with AI

A sales cadence is a structured sequence of touchpoints — email, call, LinkedIn, voicemail — designed to move a prospect from cold to qualified. Adding AI to a sales cadence means automating the research, personalizing the messages, and optimizing the timing based on response data. Done right, it multiplies your output without multiplying your headcount. Done wrong, it turns your outbound into a spam factory.

This guide covers exactly how to build an AI-assisted sales cadence from scratch, including the tools, the sequence structure, and the optimization loop.

Step 1: Define Your Cadence Architecture First#

Before you touch any AI tool, get clear on your cadence structure. AI optimizes a cadence — it doesn't create strategy. Answer these questions first:

What's your average deal cycle?

• Under 30 days → 5-7 touch cadence over 2 weeks
• 30-90 days → 8-12 touch cadence over 4 weeks
• 90+ days → 12-16 touch cadence over 6-8 weeks

What channels do your buyers use?

• Email only (common in deeply technical or enterprise contexts)
• Email + LinkedIn (standard for most B2B)
• Email + LinkedIn + phone (required for enterprise or high-ACV deals)

What's your value of getting one meeting?

• Under $5K ACV → low-touch, mostly automated
• $5K-$50K ACV → mixed human + AI, some manual review
• Over $50K ACV → mostly human, AI-assisted research

With these answers, you have your cadence architecture. Now layer in AI.

Step 2: The 7-Touch AI-Assisted Cadence (Standard B2B)#

Here's the full cadence structure that works for most B2B companies with a 30-90 day sales cycle:

Touch 1 — Day 1: AI-Researched Email#

AI's role: Research the prospect (LinkedIn activity, company news, job postings, funding). Generate a personalized first line. Draft the email body based on persona.

Human's role: Review Tier 1 accounts, approve for Tier 2+3.

Subject: [Specific reference to their world]

[AI-generated hook — signal-based, 1-2 sentences]

[Value prop — 2 sentences max]

[Single CTA — one question or one ask]

[Name]

What to optimize: Subject line open rate. A/B test 2-3 subject line variants per segment.

Touch 2 — Day 3: LinkedIn Connection Request#

AI's role: None. Just a connection request, no message.

Human's role: Send from personal LinkedIn account (or use automation tool carefully — LinkedIn is sensitive).

This creates familiarity. When your Day 7 email arrives, they recognize your name.

Touch 3 — Day 5: Phone/Voicemail#

AI's role: Generate a voicemail script based on the prospect's context and the email you already sent.

Human's role: Make the call. Deliver the script naturally, not robotically.

Voicemail script template (AI fills in brackets):
"Hi [Name], this is [Rep] from [Company]. I sent you a note earlier this week about [specific reference]. I'd love to connect — I think [value prop specific to their situation]. I'll send a follow-up email. My number is [number]. Talk soon."

Touch 4 — Day 7: Follow-Up Email#

AI's role: Generate follow-up email that references the original + adds new value (case study, stat, or new angle).

Human's role: Review and approve.

Subject: Re: [original subject]

Quick follow-up to my note earlier this week.

[New data point or case study relevant to their context — AI sources this]

Still worth a short call?

[Name]

Touch 5 — Day 10: LinkedIn Message#

AI's role: Draft a short LinkedIn message (different tone than email, more conversational).

Human's role: Send from LinkedIn, review message first.

"Hey [Name] — connected earlier this week. Sent a couple of emails but inboxes are brutal. Thought I'd try here instead. We've been helping [similar company type] with [problem]. Open to a quick chat?"

Touch 6 — Day 14: Video or Loom#

AI's role: Generate a personalized script for a 60-second Loom video. Include company-specific references.

Human's role: Record the video. This is high-effort but high-conversion for Tier 1 accounts.

For Tier 2+3, skip the video and substitute a text email with a case study link.

Touch 7 — Day 17: "Last Touch" Email#

AI's role: Generate the final email with a closing sentiment and clear opt-out path.

Human's role: Approve and send.

Subject: Closing the loop on [Company]

[Name],

I've reached out a few times over the past couple of weeks — I don't want to be a nuisance. I'll close out my notes on your side.

If timing is off or it's not relevant, just reply "not now" and I won't reach out again for at least 6 months.

If you ever want to revisit [specific value prop], you know where to find me.

[Name]

The "last touch" email reliably generates the highest reply rate of any touch in the sequence. The combination of low pressure + explicit opt-out + long re-engagement gap creates urgency without desperation.

Step 3: AI Prompt Templates for Each Touch#

Here are the exact prompts to feed into your AI system for each touch:

Touch 1 personalized hook:

Prospect: {name}, {title} at {company}
Recent signal: {signal_from_research}
Product category: AI CRM for sales teams
Write a 2-sentence email opener that references the signal naturally and connects to a CRM problem this person likely faces. Do not start with "I noticed" or "I saw."

Touch 4 follow-up:

Prior email subject: {subject}
Prospect role: {title} at {company_type}
Write a 3-sentence follow-up email that adds a new angle or data point, doesn't repeat the original message, and ends with a low-friction question.

Voicemail script:

Rep name: {rep_name}
Prospect name: {first_name}
Company: {company}
Signal: {signal}
Write a 20-second voicemail script that references the signal, states the value prop in one sentence, and ends with a callback ask. Conversational, not salesy.

Store these prompts in your CRM or a connected template library. The consistency of your prompts determines the consistency of your AI-generated outreach.

Step 4: Timing Optimization with AI#

Basic cadences run on fixed schedules. AI-optimized cadences adjust timing based on what the data shows. Here's how to build this feedback loop:

Track send time vs. open rate Run sends at three time windows for 30 days:

• Early morning (6-8am recipient local time)
• Mid-morning (9-11am)
• Late afternoon (3-5pm)

Let your data tell you which window converts for your buyer persona.

Track day-of-week vs. reply rate Tuesday through Thursday typically outperforms Monday and Friday. But this varies by industry. Run the test.

Track follow-up gap vs. response Does waiting 3 days between Touch 1 and Touch 4 outperform 7 days? Or vice versa? This seems granular, but 20-30% more replies from better timing adds up to real pipeline.

With DenchClaw, you can query this data directly: "Which send-time window has the highest reply rate for VP Sales contacts?" or "Does Tuesday outperform Thursday for my Series A target segment?" — the answers are in your DuckDB database. Check what DenchClaw is for how this works.

Step 5: The AI Review Loop#

Build a weekly review into your process. Every Friday, ask:

1. Which touch got the most replies this week? (Is your opener working? Is your last-touch email doing its job?)
2. Which segment has the highest meeting-booked rate? (Are you targeting the right people?)
3. Which AI-generated messages got edited most heavily by the human reviewers? (Signal that the prompt needs improvement)
4. How many prospects moved to "not now" vs. "interested"? (ICP fit signal)

Take this data and feed it back into your prompt templates and sequence structure. This is the compounding effect of CRM-driven cadences: every cycle, you get smarter about what works.

Step 6: Integration Requirements#

An AI sales cadence requires these components to work:

CRM (source of truth) Where contacts live, where activity is logged, where segments are built. This must be connected to every other tool. DenchClaw runs locally on DuckDB — fast, private, no server required.

Sending tool Handles email delivery, open tracking, sequence management. Options: Instantly, Smartlead, Apollo sequences, Outreach. Must connect back to CRM.

Enrichment layer Pulls signals (LinkedIn, company news, funding) for each contact. Either a dedicated tool (Clay, Apollo) or a browser agent (DenchClaw's built-in enrichment uses your existing Chrome sessions).

AI generation Produces personalized openers, follow-ups, and scripts from templates. Either embedded in your CRM/sending tool or a standalone prompt pipeline.

LinkedIn automation Manages connection requests and messages. Be careful — LinkedIn actively enforces anti-automation policies. Use conservative volume limits.

Common Mistakes to Avoid#

1. Starting AI before you have a working cadence manually If your manual cadence doesn't get replies, AI won't fix it. Get to a 3-5% reply rate manually first, then layer in AI to scale.

2. Using one cadence for all buyer personas A VP Engineering reads differently than a VP Sales reads differently than a Founder. Each persona deserves its own message framework and potentially its own cadence structure.

3. Not testing subject lines Subject lines drive 80% of open rate variance. Run A/B tests on every sequence. Let AI generate 3-5 variants and let your data pick the winner.

4. Treating AI output as final AI-generated email drafts should be starting points, not finished products — especially for Tier 1 accounts. Human review catches contextual errors that AI misses.

5. Building cadences without a clear handoff protocol When a prospect replies, what happens? Who owns it? Within how many hours? Without a clear handoff, warm replies go cold while sitting in a queue.

Frequently Asked Questions#

How many touches should an AI sales cadence have? For most B2B scenarios: 5-7 touches over 2-3 weeks. Fewer than 5 leaves pipeline on the table (most replies come after touch 3+). More than 9 in a short window starts to feel harassing. The "last touch" email at the end of the sequence is non-optional — it consistently generates the highest reply rate.

Should I automate all touches or keep some manual? For Tier 3 (volume prospects), full automation works. For Tier 1 (high-value accounts), keep at least Touch 1, Touch 4, and any video touches as human-reviewed. Hybrid approaches — AI drafts, human approves — work well for Tier 2.

What tools integrate AI into sales cadences? Apollo, Outreach, and Salesloft have built-in AI email generation. Instantly and Smartlead are sending-focused tools you pair with external AI (ChatGPT, Claude). Clay specializes in AI research and enrichment. DenchClaw provides the CRM layer with built-in AI enrichment using your existing browser sessions.

How do I avoid being marked as spam? Domain warm-up (4+ weeks), proper authentication (SPF, DKIM, DMARC), conservative sending volumes (under 100/day per domain initially), clean lists (verified emails only), and genuine personalization. Spam filters are increasingly good at detecting template-based mass sends — personalization is now a deliverability strategy, not just a conversion strategy.

How long before I see results from an AI-assisted cadence? Most sequences run 2-3 weeks from first touch to last. Add a week for meetings to be scheduled and shown. Expect to see meaningful data in week 4-6. A full optimization cycle (run sequence → analyze data → improve prompts → run again) takes 8-12 weeks.

Ready to try DenchClaw? Install in one command: npx denchclaw. Full setup guide →