Automated follow-up, reply handling, inbound lead routing, and first-response drafting

Here are 3 must-run GTM plays in 2026. Clay released Workflows last week, so I built all 3 to see how far it goes.

It is a visual canvas where you wire the trigger, the logic, the agents, and data into one play. While a table is linear, a workflow can branch, apply conditional logic, and use code.

𝗦𝗶𝗴𝗻𝗮𝗹-𝗯𝗮𝘀𝗲𝗱 𝗼𝘂𝘁𝗿𝗲𝗮𝗰𝗵

An account enters your ICP audience, a signal fires, and the workflow scores fit, splits by tier, researches the account, picks the contact, assigns an owner and drafts the message.

Tier 1 goes out on all 3 channels. Tier 2 on LinkedIn and email. Tier 3 on email only.

𝗜𝗻𝗯𝗼𝘂𝗻𝗱 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻

The moment a form lands, Clay enriches, scores fit, splits by tier and segment, assigns the best person on the team and drafts the reply.

Slack gets the notification, the rep gets the draft. The human stays in the loop, everything before them is done.

𝗔𝗕𝗠 𝗼𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻

Ad spend and rep time go to top accounts only, and the tier decides how much you spend.

Tier 1 gets ads and a rep by phone or LinkedIn. Tier 2 gets ads plus an email and LinkedIn sequence. Tier 3 gets content nurture.

3 things I learned building these:

1. Double down on what is already working. Start by converting your best-performing tables into sleek workflows. One-click process - thanks, Clay!

2. Conditional nodes consume no credits and no actions. Only enrichment and AI work do. So branch freely, and put the expensive nodes as late in the play as you can.

3. You cannot build search into a workflow yet, so "find me the best 5 people at this account" is not a play today. The contacts have to already be in your audience.

What was the first play you built with Workflows?

HubSpot just made AI agents a standard CRM feature with the launch of Agent Hub and Agent Builder. It's a good move for no-code tools that let you deploy autonomous agents right inside your stack.

If you're on HubSpot, you can now easily spin up agents to handle leads, support, and follow ups without writing a single line of code.

Templates get you started, but custom architecture gets you results. Off the shelf agents work fine for generic workflows, but systems engineered around your exact operations are what turn an AI feature into a competitive moat.

Ready to go beyond out of the box templates? Let's build your custom system together.

Do you use AI to run your sales team's outbound?

I'll die on this hill:

I use AI for:

• Buyer profile research (background, niche, role tenure)
• Signal detection across LinkedIn
• Lead scoring against ICP fit
• Reply triage and conversation state classification

Not everything, though:

Sales teams that fully rely on AI to write every message are the ones with 2% reply rates.
We send thousands of DMs across our team.
Voice consistency at scale needs automation.

I don't use AI for:

• The actual opener phrasing
• The first reply to a curious prospect
• Earning trust on the first message
• Spotting which conversations are worth my reps' time
• Firing off generic "I hope this finds you well" DMs

Don't run a sales team like a script - the relevance layer still needs a human brain on it.

I killed 4 hours/day of manual outreach with one Claude SDR workflow.

Here’s the 6-step setup I run.

Most teams still treat Claude like a smart chat box.

I treat it like an SDR.
With a job description.
With workflows.
With clear targets.

Step 1: Give it a role

Not: “Write me a LinkedIn message.”

I tell Claude:
“You are my AI SDR. Your job is to find warm signals, score leads from 0 to 100, and write personal DMs that show clear buying intent.”

One clear role.
Zero guesswork.

Step 2: Build workflows, not one-offs

I do not re-explain things every day.

I built 3 base workflows:
→ Signal capture: find warm leads from posts, comments, job changes
→ Lead scoring: give each lead a 0–100 score
→ Message writing: write DMs based on the exact signal

Same steps.
Same logic.
Different accounts.

Step 3: Give it memory

Claude needs context.

I store a simple file with:
→ ICP and segments
→ Value prop and offers
→ Tone and style
→ Past winning messages
→ Current campaigns

Claude reads this file before it works.

Step 4: Let it handle low-value work

I did a one-week audit of my time.

Tasks that did not need me:
→ Finding profiles
→ Researching signals
→ Drafting first messages
→ Scoring leads

All of that went to Claude.
I only handle review and replies.

Step 5: Refine weekly

Every Friday, I review:
→ What messages missed the mark
→ What signal logic felt off
→ What small prompt tweak would fix it

The model stays the same.
The system gets sharper.

Step 6: Scale what works

One solid workflow saved 2 hours/week.

So I added:
→ Reply handling
→ Meeting qualification
→ Simple follow-up rules

Now the stack runs most of top-of-funnel for me.

Before: 4 hours/day of manual LinkedIn prospecting.
After: 30 minutes/week reviewing Claude’s work.

Want the exact setup?

Comment SETUP and I will send you:
→ The full 6-step Claude SDR system
→ 5 copy-paste prompts for signals, scoring, and DMs
→ Context file template
→ Weekly refinement checklist

P.S. If you ship this inside your GTM, your outbound operation will not look the same in 30 days.

AI SDRs (Sales Development Reps) are useful. I use them.

But they don't replace the thing most people want to skip.
The copy.

The AI SDR handles the conversation after someone replies. It qualifies. It books calls. It handles objections well enough.

What it can't do is write the opening line that earns the reply in the first place.

That's still a human job.

Specifically, it's the job of someone who understands the avatar deeply enough to write one sentence that couldn't have been written for anyone else.

People are buying AI tools expecting them to solve a copy problem.

They're not copy tools. They're response tools.

Put the time into the copy first. Then let the tool handle what follows.

I've built dozens of AI agents across our GTM operation.

Here’s how I decide which one to build -

Most teams think “adding AI” means dropping ChatGPT into one workflow and moving on.

That is not an agentic system.

Agents are:

→ Always-on
→ Trigger-based
→ Making decisions
→ Taking actions

No human in the loop for 90% of the low-value work.

The real question is not “which tool”.

The real question is “where in your GTM does an agent belong”.

Here is the diagnostic I use across 4 functions.

Sales

→ Time from buying signal to first outreach
→ How reps prep discovery (15 minutes of Google = agent)
→ What happens to overnight replies
→ How you choose which deals to push this week

Marketing

→ How you know which campaigns drive pipeline
→ Where competitive intel comes from
→ Time from idea to campaign live
→ Who decides inbound → sales vs nurture

RevOps

→ How clean your CRM data is
→ How many tools a lead touches before a sequence
→ Which reports still live in spreadsheets
→ What breaks when you 2x volume

Customer Success

→ How you detect churn risk early
→ What signals show expansion readiness
→ How manual onboarding feels for the team

Any answer with “too slow”, “manual”, or “we do not” is an agent slot.

Then match to 5 agent types:

→ Research: signals, company intel, tech stack
→ Enrichment: data waterfall, contacts, email verify, CRM gaps
→ Content: sequences, briefs, reports from data
→ Routing: reply tags, scoring, stage updates, alerts
→ Monitoring: churn, pipeline, anomalies, data drift

Start with one.

Pick the slowest, most repetitive, most time-sensitive workflow.

Team size rule:

→ Under 10: Routing first (reply classification or lead scoring)
→ 10–50: Enrichment first (data quality unlocks all)
→ 50+: Monitoring first (what you cannot see hurts you)

I mapped 20 concrete agent builds across Sales, Marketing, RevOps, and CS into one cheat sheet.

I spent the last 3 months building AI into my actual sales workflow.

Not theory. Not a course. Real calls, real prospects, real results.

Here's what changed:

Before: 2-3 hours writing proposals
After: 20 minutes. AI does the draft. I clean it up.

Before: Forgetting to follow up by day 7
After: 5-touch sequences written and scheduled automatically

Before: Walking into discovery calls cold
After: I know their pain points, budget signals, and likely objections before I dial

Before: Researching prospects by hand
After: 50 qualified leads in under an hour

I documented everything. 14 chapters. Real prompts. Copy-paste ready.

Called it "The AI Sales Consultant's Playbook."

$37. Link in comments.

If you're in commission sales and still doing this stuff manually — this is the shortcut I wish I had.

10 ways I use Claude for sales
Appointments, nurturing, closing, and retention for non-technical founders 👇

I run sales systems not a dev team
I've been building everything with Claude
You describe what you want it builds it

Here's what I actually use it for

1/ Identifying and targeting the right leads
→ Claude builds my full targeting breakdown
→ My team only talks to buyers who fit

2/ ICP from my best 100 clients
→ Claude finds who I should go after
→ Industry, size, role, revenue all mapped out

3/ Full market system built in one day
→ Positioning, value prop, and competitor analysis
→ Plus 12 messaging angles from one brief

4/ Offers backed by 6 months of data
→ Claude structures pricing, packaging, and guarantees
→ Everything built around what already converts

5/ Outbound that books 30% more calls
→ Every email and LinkedIn message personalized
→ No copy paste and no templates ever

6/ Recovers 40% of my no shows
→ Claude sends the right message right time
→ Rebooks the call without my team touching it

7/ Pitch scripts from prospect's 3 biggest problems
→ Not a generic deck nobody reads
→ A conversation framework that actually closes

8/ Objection handling updates every 7 days
→ How AI reads my call notes every week
→ Rewriting responses based on what prospects say

9/ Training built from my best 20 calls
→ Scripts, breakdowns, and scorecards ready to go
→ New hires ramp in 5 days not months

10/ Sales docs nobody has to write
→ Playbooks, SOPs, onboarding, and KPI dashboards
→ Built for a team of 3 or 150

Most of these took under an hour
Some took 20 minutes

It hallucinates sometimes that's real
But for sales ops the ROI is stupid

If your team still does this manually
That's the bottleneck

Comment "Sales" and I'll send the full breakdown
Repost to get early access
Available 48 hours only.

I broke up with ChatGPT. 2 years together. Walked away.

Every session felt good. Warm. Encouraging. But my deals weren't closing faster.

I'd ask for objection handlers. Get 11 seconds of typing. Then a response that validated my approach and left me with more options than clarity. ChatGPT treated me like someone who needed support.

I needed a closer.

5 months ago, I tried Claude.

Pasted a stalled deal. Asked why the prospect went dark.

"Your last email buried the value prop. Here's the exact follow-up that reopens this conversation."

Brutal. Specific. Ready to send.

I don't need AI to validate my outreach. I need it to tell me why the deal stalled and how to unstick it.

Team Claude or team ChatGPT?

A quick and dirty outreach system for founder-led sales. Bonus, it won't make you hate yourself:

1. Create a project in Claude or ChatGPT and call it something like “sales and business assistant.” Drop in a detailed explanation of your ICP, strategy docs, transcripts of sales calls. This becomes your operating system for outreach, planning, and prospecting.

2. Request a data export from LinkedIn (settings -> privacy -> data). Takes them about 24 hours to send you a CSV of your connections.

3. Drop it into your AI project alongside your targeting framework and ask it for a list of people you should reach out to; ask it to include a score for each. Half the results will be wrong. Tell the AI why. You will eventually get your first 10 to 20 people to reach out to, and your AI project will get smarter.

4. Start a chat inside your project called “leads” or “prospects.” When you’re ready to message someone, screenshot their LinkedIn profile, drop it in, and ask the AI to draft a message. Refine heavily because the draft will make you want to claw your eyes out. Again, the AI will get smarter over time. Human edit always.

5. On the message itself: don’t do the fake “I saw your post about X and it really resonated with me” thing. That is insufferable. Just be human and make the ask clear.

6. Send 15-30 LinkedIn connection requests a week to grow your audience. Build lists of people in Clay on the free tier and enrich your lists with their LinkedIn profiles. Or just search manually on LinkedIn.

7. Track who you’ve messaged. Google doc with name + the last action you took. Mine is ugly but it keeps me honest, and my exec assistant uploads it to HighLevel.

8. You can have Claude Cowork scan your list and set reminders on your calendar to follow up with people if appropriate. It can even include a drafted follow-up message in the event notes to keep you from procrastinating when it's time to send a nudge.

9. Write at least one LinkedIn post a week, ideally two. This is fuel for everything else. Your content is what warms people up after they accept your connection request.

10. Audit your LinkedIn headline and about section. Your profile + content is how people decide if you’re worth paying attention to, and it’s important to the new algorithm.

11. The only metric that matters when you start is “did you do your actions this week.” Fifteen to twenty-five. You can’t control who says yes to a demo but you can control whether you did the work.

If you get into a good groove, keep the momentum going. It’s tough to slow down and have to rev the whole engine back up again.

What am I missing?