In short

• Magi's outbound runs on four tools: Apollo for accounts and contacts, Parallel.ai for enrichment, Magi for the email, and Reply.io for sequencing. Three handoffs, all automated.

• The selection rule is cost-effectiveness plus integrations plus speed for a lean team, and the last one dominates: if two tools don't talk, the gap between them is a job someone now has.

• ZoomInfo, Bombora, Lusha, Lemlist, Instantly, Smartlead and HeyReach were evaluated. None were rejected on quality. Each may join the stack when the motion needs it.

• The email is drafted where the ICP, brand, proof points, and meeting context already live, so the one step that used to need a human re-brief doesn't.

Subhrajit Dutta joined Magi a few weeks ago as the Founding GTM, bringing years of outbound experience and a strong preference against the tech stack that experience usually produces. The mature version of an ABM strategy runs on something like 47 tools. Each does a real job. Together they generate a job of their own: keeping the data consistent as it crosses 46 tools.

For a startup with a lean budget and a lean team, that job is the whole cost. So rather than asking which tools are best, he asked how few boundaries the motion could run on.

The best GTM stack is the one that keeps a lean team moving

This is where the series thread lands in outbound. The first three posts were about removing the transfer tax inside marketing production: re-briefing before a post, copy-pasting into templates, checking every asset by eye. An outbound stack is nothing but transfers. Account data moves from a database to an enrichment layer to a writing tool to a sequencer, and at every boundary a field can be renamed, a record dropped, or a persona lost.

Every tool added is another boundary. If the two sides integrate, the transfer is automated, and the boundary costs nothing after setup. If they don't, someone exports a CSV on Tuesday and reconciles it on Thursday, and the team has hired a data-entry role without a headcount line.

So tool count is a proxy for handoff count, and handoff count is the real cost. The stack below has three handoffs. All three are integrated.

A serious ABM motion needs serious data. It doesn't need 47 handoffs.

The standard pushback: serious ABM strategy needs serious infrastructure. Intent data, technographics, multi-source enrichment, a warmup layer, a sequencer per channel, revenue attribution. A four-tool stack is a toy.

It's right about the data. A motion that targets the wrong accounts fails regardless of how elegant the stack is, and cheap data is usually cheap for a reason.

It's wrong about what that implies for tool count. The data requirement is met by two tools here, and the stack that runs on 47 mostly duplicates capability across them: three sources of firmographics, two enrichment layers, a sequencer for email and a separate one for LinkedIn. Each duplication is defensible in isolation.

Start with the data you need, not every data tool available

Apollo.io is the database. Accounts and contacts, with intent signals layered on, at the coverage a startup needs at a price that doesn't require a board conversation. In Magi's workspace, a saved company search filters on employee count, industry, location, and headcount growth, which is the cheapest reliable signal that a company is about to spend on marketing.

Parallel.ai handles enrichment, which used to be the analyst work nobody wanted: finding the CEO, COO, and CFO of a target account, its annual revenue, employee count, and last funding round, and a working email for the contact. Parallel takes a CSV with company name and website headers and returns the fields. The manual version of this, per account, was the reason ABM programs at small companies quietly stopped after month two.

Both integrate, so the account list that leaves Apollo arrives in Parallel and leaves enriched without anyone touching a spreadsheet.

Write the email where the ICP already lives. Don't re-brief it into another tool.

This is the step where most outbound stacks add a writing tool and lose the plot, because the writing tool knows nothing about the company sending the email.

Magi's Build flow already holds what the email needs. The campaign carries the ICP and persona. BrandOS carries the voice, and the tone selected for outbound. Knowledge carries the meeting transcripts and proof points. The Ideas queue carries angles that the Research Agent assembled during the week from competitive intel and market signals, so the hook in the email is a fact about the buyer's market rather than a compliment about their LinkedIn.

The sequence in Magi: pick Email as the format, pick a template (Cold Outreach Email, Founder Outreach Email, or a custom one), select the voice and tone, attach an idea, and switch on the humanizer skill so the draft doesn't read as generated. Then one prompt. The agent grounds the draft in internal knowledge, sets the email brief, and produces the first version. A draft written to a founder with a lean marketing team opened on their recent funding and the content pressure it creates, cited a customer that built repeatable rituals with Magi and returned content for a new segment in two weeks, and closed on a single question.

There's a second path. Open Claude, and through the Magi MCP ask for the same email. Magi drafts it with the same context, and the result lands in the Magi editor's Chat, Design and Audit tabs like anything else. Subhrajit uses both, and the choice is about where he's already working rather than about output quality.

Either way, the thing Harshini's post described for LinkedIn is what happens here for email: the brief that used to need a person is inherited from the campaign.

Use LinkedIn first. Let prospect behaviour decide the next channel

Reply.io is the fourth tool, chosen for cost-effective multichannel sequencing and built-in domain warmup. Everything Magi drafts goes in as variants.

The sequence itself is a tree rather than a list. Day one sends a LinkedIn connection request. Reply then monitors the connection status for two days. If the request is accepted, the next steps are LinkedIn-native: a profile view after a short wait, then liking a recent post, before any message goes out. If it isn't accepted by day three, the sequence switches channel to email, with a call as a later fallback. The branch is decided by a field Reply already has, so no one has to check who connected.

That design does two things a linear email cadence can't. It uses the warmest available channel first, and it stops sending LinkedIn steps to people who haven't opted into the conversation.

What was evaluated and set aside, and when it comes back

None of these were rejected on quality. Each does something well that the current motion doesn't yet need enough to justify a new boundary.

ZoomInfo and Bombora. Better data and deeper intent than Apollo. The price is the reason they wait, and it's a price a lean startup feels.

Lusha. Direct dials. Valuable the day a phone step becomes core to the sequence; not before.

lemlist. Close to Reply.io in capability, somewhat pricier. If Reply's ceiling is hit, this is the comparison to rerun.

Instantly and Smartlead. Volume email. When the motion moves from targeted ABM to sending a great many emails, one of these joins.

HeyReach. LinkedIn automation at scale. Reply.io's LinkedIn steps cover the current volume; HeyReach is for when LinkedIn becomes the primary channel rather than the opener.

The rule for adding any of them is the same rule that built the stack: it has to integrate with what's already there, or the transfer it saves is smaller than the transfer it creates.

Head over here for the full video.

Frequently asked questions

What account-based marketing tools does a lean startup need? A lean startup needs four account-based marketing tools: a database for accounts and contacts, an enrichment layer, a writing tool that already holds brand and ICP context, and a multichannel sequencer. Magi's outbound runs on Apollo.io, Parallel.ai, Magi and Reply.io, with every handoff between them automated.

What is a good sales tech stack for outbound at a startup? A good sales tech stack for outbound at a startup minimises manual handoffs between tools. Each tool should integrate with the one before and after it, cost less than the headcount it replaces, and do a job the motion needs now rather than later.

What is an ABM strategy for lean teams? An ABM strategy for lean teams targets a small set of accounts with signals that predict spend, enriches them automatically, writes outreach grounded in the company's own context and proof, and sequences across LinkedIn and email based on how the prospect responds. The constraint is people to run it, so tool count is kept to the minimum that integrates.

Why is a bloated tech stack worse than none? A bloated tech stack is worse than none because every tool boundary that isn't integrated becomes a manual data transfer. A team of two or three ends up reconciling exports rather than running outreach, and the stack consumes the capacity it was bought to free.

How does Magi write cold emails? Magi writes cold emails from context that already exists in the workspace: the campaign's ICP and persona, the voice and tone in BrandOS, meeting transcripts and proof points in Knowledge, and angles the Research Agent assembled from market signals. The marketer picks a template, attaches an idea, switches on a humanizer skill and prompts once. The same draft can be requested from Claude through the Magi MCP.

What are the best cold outreach tools for multichannel sequences? The best cold outreach tools for multichannel sequences let a single sequence branch across LinkedIn and email based on prospect behaviour. Magi uses Reply.io, which branches on LinkedIn connection status and includes domain warmup. lemlist is the closest alternative; Instantly and Smartlead suit high-volume email; HeyReach suits LinkedIn-first motions.

Should a startup use ZoomInfo or Apollo? A startup should start with Apollo.io for coverage, intent signals and price, and move to ZoomInfo or Bombora when the motion's data needs outgrow it and the budget supports it. Both were evaluated at Magi; Apollo was chosen for cost, and the others remain options.

How does the Magi MCP fit into an outbound stack? The Magi MCP lets Claude read brand and campaign context and Knowledge from Magi, and draft content through it. In an outbound stack it connects the systems around Magi so account data can flow in and finished, on-brand emails can flow out to the sequencer without a manual step.

About this post. Fourth in The 100x Marketer, a series on how Magi's own team uses the product. Sourced from Subhrajit Dutta's recorded workflow episode, the stack and evaluation notes he shared alongside it, and Magi's outbound configuration. No performance figures are included because none are yet publishable. Published by Magi HQ, an agentic marketing automation platform for lean B2B teams. Customers include 100ms, Payactiv, Lyric and Accuknox.

TL;DR. Subhrajit Dutta, Founding GTM at Magi, runs outbound on four account based marketing tools: Apollo.io for accounts and contacts with intent signals; Parallel.ai for enrichment (executives, revenue, funding, contact details); Magi for the email, drafted from the campaign's ICP, BrandOS voice and tone, Knowledge and the Research Agent's ideas, with a humanizer skill and optionally via Claude through the Magi MCP; and Reply.io for multichannel sequencing with domain warmup, branching on LinkedIn connection status into profile-view and post-like steps or into email and a call. The selection rule is that each tool must integrate with the others and cost less than the headcount it replaces, because each non-integrated boundary is a manual job. ZoomInfo, Bombora, Lusha, lemlist, Instantly, Smartlead and HeyReach were evaluated and may be added as the motion grows. No reply rates are published yet. Magi is building toward more of the chain in-house; the MCP is the current bridge. Customers include 100ms, Payactiv, Lyric and Accuknox.