The Instantly.ai + v0 cold outreach stack

The reply-rate lift from sending bespoke v0 prototypes with cold emails covered in an earlier post was real. The operational machinery behind those emails — inboxes, deliverability, sequences, CRM — is a separate and larger piece of work, a

The reply-rate lift from sending bespoke v0 prototypes with cold emails covered in an earlier post was real. The operational machinery behind those emails — inboxes, deliverability, sequences, CRM — is a separate and larger piece of work, and the question we get asked most often after the prototype piece is: "okay, but what's the actual sending stack?" The instantly ai v0 stack we run is three layers: Instantly.ai for deliverability and send infrastructure, v0 for per-prospect prototype production, and Claude for the research between the two — together they support two operators sending around 350 personalized prospect emails per month at an 18–24% reply rate, at a total monthly cost of about €240. This is the full setup.

The stack, end to end

The flow of a single prospect through the stack:

1. List sourcing produces a CSV of 200 names per week. Tools used: LinkedIn Sales Navigator, a scraping-friendly adapter, and manual review.
2. Claude research enriches each record with a one-paragraph observation about the prospect's site or business.
3. v0 prototype builds a bespoke page per prospect, based on the observation.
4. Vercel deploy ships the prototype to a unique subdomain.
5. Instantly sequence sends the email, tracks opens and clicks, handles replies, and routes hot leads to Slack.
6. CRM (Attio) records the conversation and pipeline state.

Each layer is doing what it's best at. No single tool can do all six. The operators who try to do this inside a single platform — usually HubSpot or a specialized Instantly-alternative — get worse results because one or more of the layers is compromised.

Instantly: the layer that sends the mail

Instantly.ai handles the infrastructure of getting email into inboxes. This is a job, not a button. Here's what it covers:

Inbox warmup. Every sending address is gradually "warmed up" by Instantly's peer-to-peer warmup pool, which sends and replies to low-volume messages across the pool. A cold inbox starts at 5 sends/day for two weeks before ramping to 30–40/day. Skipping warmup sends all your mail to spam.

Deliverability monitoring. Instantly tracks inbox-placement across major providers, shows which inboxes are degrading, and pauses them automatically when the score drops. We've had two inboxes auto-paused for four days each in the last year; both recovered fully without intervention.

Sequence management. We run a two-touch sequence (the prototype email, then a hot-opens follow-up only to prospects who opened and didn't reply). Both touches are written by humans; Instantly just schedules them. We don't use its AI copy generator; it's the only feature we actively avoid.

Reply detection. When a prospect replies, Instantly stops the sequence and pushes the email to our Slack inbound channel. This is where the human operator takes over.

We pay Instantly around €97/month for our volume. This is not cheap for a tool, but compared to what we'd spend building and maintaining the equivalent on Mailgun/SES + custom warmup scripts + custom deliverability tracking, it's ruthless value.

Inbox architecture: the quiet load-bearing piece

We send cold email from eight inboxes across three domains. The domain structure is deliberate.

The primary domain is our-agency.com — the agency's main website. This domain is used for in-bound contact only. No cold email leaves from this domain. Ever.

Two sending-alias domains are hello.our-agency-name.com and team.our-agency-name.com. These are dedicated cold-email domains, each with a proper MX record, SPF, DKIM, and DMARC setup. They look like legitimate sub-brands of the main domain (because they are). Cold email is sent from addresses on these domains.

Each domain has four inboxes. Arthur, Jan, Miriam, Tobias — real team members. Each inbox is genuinely monitored. Replies go to actual humans. Nothing on these inboxes is automated past the Instantly sequence.

Why the two sending domains and not one? Because deliverability at scale degrades when you concentrate all sends on a single domain. Eight inboxes on one domain will run into SPF-rate-limiting and spam-folder routing inside three months. Four inboxes on each of two domains distributes the sending weight.

Each inbox sends no more than 40 emails per day, which is well below the deliverability ceiling for a warmed-up inbox but gives us headroom on a spiky sending day. Total daily capacity: 320 cold emails. Actual daily volume: 30–60. We are massively under-utilizing the capacity, by design.

The warmup calendar

Every new inbox runs a strict warmup calendar. We built this as a checklist because we've watched operators skip it and pay for months.

• Days 1–14: warmup only. Instantly's peer-to-peer warmup runs the inbox at 5 sends/day, slowly ramping to 12. No cold email leaves. No exceptions.
• Days 15–21: first real sends, 5/day. Only to prospects we have strong reason to believe will open and reply (usually referrals or lukewarm re-engagement, not pure cold). The purpose is to show the inbox real engagement patterns.
• Days 22–35: scale to 20/day. Actual cold outreach, but kept below 20 sends/day during this window.
• Days 36+: full capacity, 40/day. The inbox is seasoned enough to handle full cold sending.

The two-to-three-week warmup period is where most newcomers to cold email give up. It feels like dead time. It isn't. The inboxes that skip warmup end up in spam inside their first month and never recover; the ones that warm properly send reliably for years.

v0: the prototype layer

The production process for each prospect's prototype runs in v0.

The template we give v0 is a standing prompt that we've iterated twenty times over the last six months. The current version is about 400 words and specifies:

• The visual structure of the page (hero + before/after + CTA in a specific layout)
• The typography stack (Outfit display, Geist sans body)
• The color rules (respect the prospect's accent, don't invent new ones)
• The constraint that nothing fabricated should appear on the page (no fake testimonials, no invented metrics)
• A footer specifying "built in 25 minutes by [agency name]" — the speed disclosure

We drop in the prospect-specific context (company name, accent color, logo URL, the one observation) and v0 produces a first cut in ninety seconds. The operator edits it in Cursor for 8–12 minutes. The final page goes to Vercel.

The average time per prototype: 22 minutes, end to end. Breakdown:

Step	Time
Research (Claude + 2 min of manual site review)	6 min
v0 initial prompt + review	3 min
Cursor edits (copy, colors, spacing)	9 min
Vercel deploy (git push + verify URL)	2 min
Write and send the email (Instantly)	2 min
Total	22 min

Eight hours of operator time produces 20–22 prototypes, which is one full day's outbound capacity for one operator.

Claude: the research layer

The piece that turns "here's a list of 200 names" into "here's a list of 200 names with specific observations" is a Claude agent we've built as an internal tool. It runs in a browser window we open in parallel to Instantly and v0.

The agent does three things:

1. Visits the prospect's website and pulls screenshots, load timings, and a lightweight crawl of the top 10 pages.
2. Runs a prompt-suite over that content, looking for specific classes of observation — broken mobile UX, outdated value propositions, conflicting CTAs, Meta Pixel issues, slow LCPs, typography inconsistency.
3. Writes a one-paragraph observation draft for the operator to review and refine.

About 40% of Claude's observation drafts are good enough to use verbatim. 40% need rewriting by the operator (the observation is valid but the wording is off). 20% are wrong or inapplicable and get tossed.

The agent doesn't replace the operator's judgement. It compresses the research step from 15 minutes per prospect (manual site review) to 3 minutes (review Claude's draft and edit).

The research used to be the bottleneck. Now it's the bottleneck that used to be a wall. Four prospects an hour is the pace; I couldn't do two an hour by myself.

— an operator in our internal review, December 2025

The CRM layer

We use Attio. It's the first CRM I haven't hated. Records sync from Instantly on reply; sync back on meeting booked; integrate with Calendly for scheduling; integrate with Slack for notifications.

Each record has a small set of fields: company, contact name, outreach date, observation used, prototype URL, meeting date if booked, deal stage, ACV if closed. Ten fields. Not fifty. The CRM is a logbook, not a data-science project.

Reporting: we pull monthly numbers by operator — emails sent, replies received, meetings booked, proposals sent, deals closed. We look at the numbers once a month. Anything more frequent is noise.

Deliverability monitoring in practice

Instantly surfaces per-inbox deliverability scores. We check them weekly — a 15-minute ritual on Monday morning. If any inbox drops below 85, we pause it immediately, let it re-warm for 10 days on Instantly's pool, then bring it back online.

We also run a monthly inbox-placement test via GlockApps. One email sent to a seed list of 40 inboxes across major providers, scored on placement (inbox, promotions tab, spam). Our aggregate placement ran at 86% inbox, 10% promotions, 4% spam across 2025. Below 80% inbox, we stop new outreach and investigate.

The factors that most affect placement, in our experience:

• Send volume consistency. Spikes above the norm for an inbox degrade placement for two weeks.
• Subject-line repetition. Reusing the same subject line across too many prospects triggers spam-filter clustering.
• Plain-text ratio. HTML-heavy emails (images, signatures, tracking pixels) score worse than near-plain-text. Our emails are 95% plain text with a single link.
• Link reputation. The Vercel subdomains are a risk here. We rotate the base domain quarterly to avoid any single subdomain getting flagged.

The reply-management playbook

When a reply comes in, Instantly stops the sequence and routes the email to our inbound Slack channel. The routing happens inside 60 seconds. The operator responds within one business hour during working time.

Reply types:

• "Interested, let's talk." Roughly 60% of replies. Route to meeting-booking flow. Send calendar link.
• "Can you tell me more?" 20% of replies. Send a short, structured response addressing their specific question, re-link to the prototype.
• "Not a fit." 12% of replies. Thank them, offer to stay in touch if their situation changes, close the record.
• "Please remove me." 6% of replies. Remove from the list immediately, flag the email address as do-not-contact, send a one-line confirmation.
• "Angry / defensive." 2% of replies. Apologize, remove, don't argue. This has happened four times in 2025. Each time it was because the observation was too pointed or personal. We use these as training material for the operator.

The one-business-hour response SLA is the discipline that makes the stack work. A reply that sits for 24 hours is a reply that cools. A response inside an hour catches the prospect still in the tab.

The monthly cost

Line	€ / month
Instantly.ai (sending platform)	97
Google Workspace (8 inboxes × 3 domains)	48
v0 Pro	60
Claude API (research agent)	22
Vercel	0 (within included plan)
Attio CRM	29
GlockApps (deliverability testing)	15
Domain renewals (amortized)	6
Monthly stack total	277

Actual 2025 average ran closer to €240 because we only paid for GlockApps quarterly. Round up to €280 if you want a conservative budget.

Compared to the single-platform alternatives we've tested (HubSpot's Sequences, Apollo.io, Lemlist): the stack is cheaper by about 40% and produces reply rates 2–3x higher. The trade is that you're running three tools instead of one. For a small team, this is fine. For a large team with a dedicated "RevOps" function, the overhead of running three tools might not be worth it. At our size, it's the right shape.

The weekly operator rhythm

Monday: list review + prototype-plan for the week. 90 minutes. Tuesday–Thursday: build + send, 20 prospects/day, 3 hours morning + 3 hours afternoon. Friday: no sends. Review the week's replies, update CRM, queue next week's list.

Weekly output per operator: 60 prospects, ~13 replies, ~5 meetings booked, ~1 proposal written, ~0.4 deals closed.

Two operators run this rhythm, giving us roughly 2 deals closed per month from cold outbound, which at an average deal of €42,000 first-engagement is €84,000/month of new revenue. The stack costs €280. The operators cost about €14,000/month fully loaded. Net contribution from this channel: €69,720/month at steady state.

This number is high enough that we trust the channel and low enough that we don't over-invest in it. Most of our business comes from referrals, as is typical in the DACH market; cold outbound is the top-of-funnel layer that keeps the referral pipeline warm.

What we're tuning in 2026

More Claude-in-the-loop, less human-in-the-loop on research. Our research agent is producing 40% usable observations now. We think we can get to 65–70% with better prompt scaffolding and better browsing tools. That would free up another 4 minutes per prospect, which is 90 minutes per operator per day.

Multi-language expansion. We run primarily in English and German. Adding Dutch and French to reach more of the EU is on the 2026 roadmap — the template and stack both support it; we need localized operators who can handle replies in language.

Shift more of the proposal step into the prototype. Currently the prototype is the email hook; the proposal is a separate Google Doc that goes out after the first meeting. We're testing a model where the prototype page has a second state — unlocked after the first call — that contains a scoped proposal. Early read is that it shortens the close cycle by a week.

Reduce the sending volume. Counterintuitive. At our current scale, an extra 20% volume produces diminishing marginal replies and measurable wear on the deliverability. We might actually ship less mail in 2026, and channel the freed operator time into the research layer, where the leverage is higher.

The five-line version

• Instantly handles the sending; nothing else comes close on deliverability at this scale.
• v0 handles the prototype layer; 22 minutes per bespoke page, no tool ships faster.
• Claude handles the research compression; 40% of observations usable verbatim is a real productivity multiplier.
• Eight warmed inboxes across three domains; 40 sends/day each; 320 daily capacity used at roughly 20% of ceiling.
• €280/month stack cost; €84,000/month revenue contribution; two-operator team. That's the business, laid out.