How to integrate Instantly with HubSpot for cold outbound

Instantly is the most capable cold email platform on the market right now. It's also one of the most underused. Most teams sign up, connect five mailboxes, paste a sequence they wrote in Notion, and call it an outbound program.

Three months later the inbox is full of OOO replies, the AE doesn't know which deals came from which campaign, and the bounce rate on two domains has quietly killed deliverability for the whole stack. Nobody set up the layer underneath.

The product itself is rarely the problem. Instantly ships with unlimited inboxes, AI warmup, a 450M-record lead database, a unified inbox, and a workflow engine. On paper it's the full stack. In practice, most accounts use maybe 20% of it. The gap is between what the tool can do and what teams actually wire into the rest of their stack. This piece covers what that wiring looks like for a mid-market outbound program (10 to 100 reps), what we build for clients running real volume on Instantly today, and what the ROI tends to look like once the layer underneath is actually in place.

The Reply Chaos Most Instantly Customers Have

When teams plateau on Instantly, it usually shows up the same way. The symptoms are predictable:

• Sending volume keeps climbing but reply rates flatten, because nobody owns deliverability per mailbox.
• The Unibox is full of OOO and unsubscribe replies that someone has to triage by hand every morning.
• Positive replies sit for hours because the AE never gets pinged in Slack, and the CRM doesn't know the deal exists.
• Customers and open opportunities accidentally get cold-emailed, because the suppression list is a CSV someone last updated in February.
• No one can answer "which campaign printed the pipeline this quarter," because Instantly and the CRM live in separate worlds.

Every symptom on that list is fixable. None of them get fixed by sending more email.

Automation Plays We Build with Instantly

These are the four plays we run for almost every Instantly client. Together they take outbound from a blasting motion to a measurable, defensible system.

1. Reply Classification and Hot-Lead Routing

Trigger: every new reply landing in the Unibox. Workflow: an n8n flow pulls the message via webhook, classifies it through Claude (interested, not now, OOO, unsubscribe, wrong person, referral, competitor mention), and acts on each lane. Interested replies create a lead in HubSpot or Pipedrive with the full thread, fire a Slack DM to the assigned AE, and book a follow-up task. OOO replies get parked with a return date and re-queued for that day. Unsubscribes get suppressed across the whole workspace, not just the campaign. Wrong-person replies trigger an enrichment lookup for the right contact at the same company. Competitor mentions get flagged to the product team. Outcome: AEs only ever see conversations worth their time, and response speed on hot replies drops from hours to minutes. We have seen booking rates roughly double when first-touch happens inside the first 10 minutes instead of next morning.

2. Mailbox Health Monitoring and Auto-Rotation

Trigger: a daily cron at 6am. Workflow: hit the Instantly API for every connected mailbox, pull warmup status, bounce rate, spam rate, open rate trend, and reputation score. Anything trending into the red zone (bounce > 4%, spam > 0.3%, open rate dropping more than 20% week over week) gets pulled from active campaigns automatically and put back into warmup-only mode. A summary fires into a Slack #deliverability channel every morning with red, yellow, and green counts. A separate weekly report shows the mailboxes that have been red for 2+ weeks so the ops lead can decide whether to retire the domain. Outcome: domain reputation gets defended before it dies, and you stop running campaigns from mailboxes that are quietly poisoning your inbox placement across the whole stack.

3. Per-Lead Personalisation Pipeline

Trigger: a new list pushed to Instantly. Workflow: before the sequence starts, every contact runs through a Clay table that pulls signals (recent funding, hiring page, tech stack, product launches, LinkedIn activity, podcast appearances). A Claude prompt writes a one-sentence opening line per lead, grounded in one specific signal, with a hard rule that the line has to reference something a human could verify in 30 seconds. The line writes back to Instantly as a custom variable. The send goes out with a first line that actually references something true about that company. Outcome: reply rates typically lift 1.5 to 3x compared to generic personalisation tokens, because the opener reads like a human noticed something specific instead of like a templated mail merge.

4. CRM Sync and Attribution Loop

Trigger: any state change in Instantly (new lead, reply, meeting booked, opportunity created in the built-in CRM). Workflow: a two-way sync into HubSpot, Pipedrive, or Salesforce that writes the source campaign, sequence, variant, mailbox, and personalisation signal onto the deal record. When a meeting books on Cal.com or Chili Piper, attribution gets stamped back to the original send. When a deal closes won in the CRM, the value gets pulled back into the campaign report so closed revenue per campaign is one query away. Outcome: the team can finally answer which campaigns, sequences, and senders actually print revenue, and budget the next quarter off real data instead of vibes. The same loop also tells you which messages are burning the list, which is just as valuable.

How Instantly Should Integrate With Your Stack

Instantly is a sending engine, not a system of record. Treated that way, the integration surface is straightforward:

• CRM (HubSpot, Pipedrive, Salesforce): the system of record for contacts, deals, and revenue. Instantly pushes replies and meetings in; the CRM pushes suppressions out.
• Clay or Apollo: the enrichment and signal layer. Lists get built and enriched there, then handed off to Instantly only when they're ready to send.
• n8n or Make: the orchestration layer connecting Instantly's webhooks and API to everything else. This is where the reply classifier, mailbox monitor, and attribution sync live.
• Slack: where hot replies, deliverability alerts, and meeting bookings surface in real time, not as an email digest.
• Cal.com or Chili Piper: the booking layer. Meetings booked from outbound get attributed back to the source campaign, not lost to a generic "website" channel.
• Claude or GPT (via API): the intelligence layer for per-lead personalisation, reply classification, and reply drafting on warm follow-ups.

What ROI Actually Looks Like

Indicative ranges, not promised. Numbers below assume a team running 20,000 to 50,000 emails a month before the build.

• Reply rates typically lift from 1 to 2% to 3 to 5% once per-lead personalisation runs against real signals.
• Time spent triaging the Unibox usually drops 70 to 90% after reply classification, because the AE only sees the interested lane.
• Speed-to-first-touch on hot replies lands between 2 and 10 minutes instead of 4 to 24 hours, which is where booking rates compound.
• Domain reputation incidents (one mailbox quietly burning) usually go from quarterly to near zero once health monitoring is in place.
• Pipeline attribution from outbound typically goes from "we think it's working" to a defensible weekly number, which is the wedge for getting more budget the next quarter.

Where Teams Go Wrong

Five failure modes we see almost every time we audit an Instantly account:

• Buying more mailboxes before fixing the ones already burning. Volume on top of broken deliverability just spreads the damage.
• Personalising on first name and company only. Every other cold email in the inbox does that. The signal has to be real (recent news, hiring, funding, tech change) or the personalisation tax isn't worth paying.
• Letting the Unibox become a second Gmail. If a human is reading every reply, the system is incomplete. Classification should sit between the reply and the human.
• Skipping the CRM sync because it feels like overhead. The cost shows up six months later when no one can prove outbound generated pipeline.
• Running the AI Sales Agent end-to-end without a human in the loop on the first three replies. The agent is good. It's not yet good enough to handle every nuance of a high-ACV conversation. Use it to triage, not to negotiate.

Where Moonira Comes In

Most teams don't need help running campaigns in Instantly. The product is good enough that anyone can set up a sequence in an afternoon. What they need help with is the layer underneath: the reply classifier, the deliverability monitor, the CRM sync, the attribution loop, the suppression hygiene. That's a 4 to 6 week build for us, depending on what's already in place. After it ships, outbound runs as a system instead of a daily lift, the AE only ever sees conversations worth their time, and the operator gets to see one number every Monday morning: pipeline generated last week, broken down by campaign, sender, sequence, and message. That number is what makes the next round of headcount or budget conversations easy.

If your team is sending real volume on Instantly and the layer underneath isn't there yet, that's the build we do.

If you want the short version: Instantly handles the sending. We handle everything that turns the sending into a revenue system.