Inside the AI GTM stacks at HubSpot, Zapier, Clay & Lusha

Tools in this workflow

Key learning

Data quality matters more than prompt sophistication. The biggest CVR lifts came from unstructured behavioral inputs — chat transcripts as an AI input produced +200% CVR in one test. Expect three or more prompt iterations before results turn positive.

— Lucy Alexander, Director of Agentic Prospecting · HubSpot

Lucy’s second workflow — worth knowing about

Across 800+ reps, HubSpot also runs an AI agent that ranks each rep’s book in a four-quadrant fit/intent model. Accounts re-rank dynamically as new signals fire, and accountability mechanisms attach to each quadrant. The system surfaces the top 30% of accounts that drive 95% of revenue — which is what makes the signal-weighted KPI above mean anything in practice.

I have a new inbound lead I need to research before calling back.

Using the Lusha connector, find verified contact data for this person:

Name: [full name]
Company: [company name]

Then pull:
- Current title, seniority, and department
- Verified email and direct dial
- Company headcount, industry, and tech stack if available
- Any recent buying signals on the account

Using that context, draft a short personalized follow-up email I can
send before the call. The email should:
- Reference something specific about their role or company situation
- Connect it to a relevant pain point
- Have a single clear ask

Keep it under 100 words.

Tools in this workflow

Key learning

Standalone tools that needed a new login got abandoned. The workflows that stuck were already inside reps’ tools — signal fires, draft lands in the outbox, MEDDPIC writes back to CRM. No new tab. Context is the whole game: good context creates trust instantly, bad context destroys it faster than no outreach at all.

— Lindsay Rothlisberger, Director of Revenue Operations · Zapier

I need to prep for a call today and draft a follow-up ready to send
right after.

Using the Lusha connector, pull verified data on:

Contact: [name]
Company: [company]
Call context: [discovery / demo / renewal / QBR]

Give me:
1. A pre-call brief — who this person is, their role, company
   snapshot, any recent signals on the account
2. Three questions I should ask given their situation
3. A draft follow-up email I can send within 30 minutes of hanging up,
   with a placeholder [NEXT STEP] where I fill in what we agreed

Keep the brief scannable. Keep the email under 120 words.

Tools in this workflow

Key learning

Don’t defer creative decisions to AI. Every campaign that worked started with a human-designed message. AI scaled the distribution. The other shift: mass cold outbound got scaled back — not killed, but deprioritized against the opportunity cost and the brand cost of being one more cold email in a crowded inbox.

— Davide Grieco, Head of Growth · Clay

I want to build a target account list for an ABM campaign using a
specific signal, not just firmographic criteria.

Using the Lusha connector:

Signal to filter by: [example: companies that recently raised Series B
/ are hiring in RevOps / installed Salesforce in last 90 days]
ICP criteria: [industry, headcount range, geography]
Campaign goal: [pipeline / expansion / competitive displacement]

Give me:
1. A list of target accounts matching the signal + ICP
2. For each account: the signal that qualifies them, company snapshot,
   and recommended entry point (who to contact first)
3. A one-line campaign message hook tailored to the signal — not a
   generic pitch

Return as a table I can copy into a campaign brief.

Tools in this workflow

Key learning

The teams moving fastest didn’t add AI first — they removed the manual step sitting between data and action. AI came after the process was clean. Get that backwards and you’ve automated a broken workflow.

— Elad Uzan, VP of Product · Lusha

I have a buying signal I want to act on today. Help me go from signal
to first touch in one pass.

Using the Lusha connector:

Signal: [example: contact just changed jobs to VP RevOps at [company]
/ [company] just raised Series B]
Target company: [company name]
My ICP: [describe who you sell to]

Do this:
1. Verify the right contact at the account for this signal — title,
   email, direct dial
2. Pull a quick account snapshot — size, industry, any other active
   signals
3. Write a cold outreach email that leads with the signal as the reason
   for reaching out
4. Suggest a follow-up sequence structure (touch 2 and touch 3) based
   on the trigger type

Email tone: direct, no AI opener, under 100 words.

Lucy Alexander · HubSpot

Overweighted prompt sophistication, underweighted data quality. Early prompt iterations all failed — it took three or more rounds before results turned positive. The biggest lift came from unstructured inputs like chat transcripts, but that insight came late.

The lesson: better inputs beat better prompts. If results aren’t moving, the question is what you’re feeding the model — not how cleverly you’re asking.

Lindsay Rothlisberger · Zapier

Built standalone AI apps outside of where people already work. The tools required a separate login and created silos instead of integration. Without managed context shared across the GTM, AI amplified fragmentation instead of reducing it.

The lesson: AI embedded in existing tools beats a new tool that does one thing well. If it needs a new tab, it probably doesn’t ship.

Davide Grieco · Clay

Two misfires. First: vibe-coding custom solutions for problems that were already solved — wasted engineering time and opportunity cost. Second: deferring campaign ideation and creative decisions to AI. Campaigns without a human point of view produced content that didn’t convert.

The lesson: AI can’t replace taste. Use it to scale what works, not to figure out what works.

Elad Uzan · Lusha

The black-box failure. The team built a ready-to-go list that landed in front of reps with no rationale attached — the logic was that reps wanted the work done for them. Reps rejected it. They didn’t want a black box. They wanted to understand why these people were in front of them, and they wanted to be the ones deciding what the machine was optimizing for. Nobody asked to train a model. Everyone wanted to say “these are the signals that matter to me.”

The lesson: technology has to adapt to how each team works, not the other way around. Do the heavy lifting, but leave the controls in the seller’s hands.

The teams pulling ahead in 2026 aren’t running more AI. They got their data clean enough to act on it. Then they removed the manual steps. Then they automated. In that order.