An outbound sales tech stack is the set of tools that runs cold outreach from a written ICP to a logged conversation in your CRM. Each tool owns one job. There are ten jobs, and they run in a fixed order. The tools matter less than the handoffs between them, because a handoff is where a list breaks.
This page stays on outbound: the order, what each tool passes to the next, what to check at each step and what goes wrong. For the whole stack by company stage, with every layer and the build-or-buy split, start with our guide to the GTM tech stack.
The ten jobs and the tool for each
This is the order we build in for clients. Each row names the tool we default to, what it hands on, and the failure we see most often at that handoff.
| Job | Tool | Hands to the next step | Where it breaks |
|---|---|---|---|
| Define the ICP | Claude | Titles, industries, headcount, tech stack and buying criteria, written as filters | The ICP stays a paragraph, so the database filters are guesses |
| Source and verify | Owned data and BlitzAPI first, then Prospeo | Contacts with a verified email and a mobile number | The filters drift from the ICP and the list fills with near-fits |
| Final validation | BounceBan, DeBounce | Only the addresses that are safe to send to | Catch-all passes are treated as clean and bounce once sending starts |
| Enrich with signals | Clay, with Apify for LinkedIn and job data | Each record with a reason to reach out now | Signals sit in columns the copy never reads |
| Write | Claude | Copy for each contact, built from the enriched record | Empty fields turn into generic lines or broken merge tags |
| Send email | Smartlead or EmailBison | Sends, opens and replies on the email steps | It keeps emailing someone who already replied on LinkedIn |
| Send LinkedIn | HeyReach | Connection requests, messages and LinkedIn replies | Its own interested tag counts a polite thank-you as interest |
| Call | Kixie | Calls logged against the contact | Calls happen in a dialler the other senders never hear about |
| CRM write-back | HubSpot or Attio | One record per person, with every touch on it | Each sender writes its own version and the record splits |
| Reporting | Claude, with alerts in Slack | What moved and what broke, by channel and segment | The numbers come from four dashboards and never reconcile |
Two tools work across the rows instead of inside one. Claude shows up at the start, the middle and the end, because it reads what every other tool produces and decides what happens next. n8n carries records between the steps once moving them by hand costs real time. Before that point it is optional.
Why the handoffs matter more than the tools
Every tool you add to the stack adds a handoff, and every handoff is another place a list can break. Built in the right order, a lean stack is the one with the fewest places to fail. When replies drop, you can point to the exact step that broke.
So each handoff gets a check before the next tool sees the data. The sections below give the check for each job. They follow the way the work splits: the data jobs, the writing, the sending and the shared record.
The data jobs: define, source, validate, enrich
Define the ICP
Claude turns a plain-language description of your buyer into a structured ICP. Titles, industries, headcount, the tools they run and the buying criteria are all mapped before a single contact is sourced. The output is a set of filters a database can actually run.
Check before the handoff: every criterion maps to a field your data source holds. A criterion that cannot be filtered moves to the enrichment step, where Clay can research it account by account.
Source and verify
Prospeo finds the contact, validates the email and returns the mobile number in one pass. Finding and validating in one pass matters the first time you watch a list bounce halfway through a send. Its Starter plan is $49 a month for 2,000 credits, where a verified email costs 1 credit and a mobile number 10 (Prospeo pricing, checked 22 October 2026).
At volume, no single source covers everyone. We call several in a fixed order, cheapest first, and stop at the first clean result. That ordering is waterfall enrichment, and it keeps the paid sources working only on the gaps.
Check before the handoff: read twenty rows against the ICP by hand. A list that has drifted from the filters is cheaper to fix now than after every record has been enriched.
Run a final validation
Every confirmed email goes through BounceBan or DeBounce before launch. An address can pass verification and still be risky, especially on a catch-all domain, where the server accepts mail for any address. BounceBan settles those without sending a message.
This is the check that protects sender reputation, and every later step depends on that reputation. Our email verification waterfall judges every address by the same rule, including one a finder has only just returned.
Check before the handoff: catch-all and unknown results get their own column and their own decision. Mixed in with the clean passes, they hand their bounces straight to the sender.
Enrich with signals
The clean list goes into Clay. Hiring signals, firmographics, intent data and job changes are layered onto every record. Apify adds what only exists on a web page, such as who engaged with a LinkedIn post or which companies are posting jobs this week. Where Clay stops and Claude Code takes over is covered in Clay vs Claude Code.
Before this step you know who to contact. After it you know why they are worth contacting now.
Check before the handoff: every record carries one reason-to-reach-out field that the copy step will read. A signal in a column the copy never uses is spend with nothing to show for it.
Writing: the copy reads the record
Claude reads the enriched record for each contact and writes from what is true about that account. Every email starts from real account context instead of a template full of generic variables.
That makes the writing step only as good as the step before it. A missing signal produces a generic opening line. A missing first name produces a broken merge tag in front of a real buyer.
Check before the handoff: read a sample of finished messages the way the sender will show them, with every variable filled. Then freeze the copy, so the version loaded into the sender is the version that was approved.
Sending: one tool per channel
Start with one sending tool per channel you use, and add channels as your buyers justify them. Picking between email senders is its own comparison, covered in our guide to cold email software.
Smartlead or EmailBison sends the sequence and tracks opens and replies. It also warms up the domains and rotates inboxes as volume grows. That second job is what keeps the email channel running past the first campaign.
HeyReach sends connection requests and follow-ups across several sender profiles, working the same list the data steps built. Its replies collect in one inbox. We pull them out once a day and have Claude classify them, because HeyReach's own interested tag counts a thank-you for connecting as interest.
Calling
Kixie puts a call into the sequence as a step. It suits accounts worth a live conversation and contacts with no valid email. On HubSpot, its integration logs each call to the contact from the dialler itself.
Check before the handoff: load each campaign as a draft and have a person switch it on. Copy, list, senders, schedule and suppression lists are checked first, and a missing one holds the launch.
Every tool holds its own copy of the campaign
This is the part of sequencing that gets underestimated. Each sender keeps its own list, its own step counter and its own idea of who has replied. Smartlead knows about email, HeyReach knows about LinkedIn and Kixie knows about calls. None of them knows what happened in the other two.
Picture a prospect who answers a LinkedIn message on a Tuesday. Unless something tells the email tool, step three of the email sequence still goes out on Wednesday. The prospect now has a live thread and a cold follow-up from the same company.
So something above the senders has to own the sequence. It decides who moves from email to LinkedIn to a call, and it halts all three as soon as a reply arrives on any of them. In the systems we build, that owner is Claude. It assembles the campaign inside each sender and reads every reply, which leaves the senders with a single job.
When a reply comes in on any channel, the coordinating layer does five things:
- Classifies it, so a not-now, a wrong person and an interested reply each get a different next action. Our reply triage skill runs that pass.
- Halts the person everywhere, including in the senders that did not receive the reply.
- Suppresses opt-outs across every campaign, so an unsubscribe on one list holds on all of them.
- Writes the reply to one record, with the channel and the category attached.
- Hands positives to a person, with the thread and the account context travelling with them.
CRM write-back and reporting
On sales calls, buyers ask where all this activity gets logged. The answer is one CRM record per person, written to by every tool. Any CRM that connects through an API can take the write-back, and our builds usually run on Attio or HubSpot.
Agree the fields before the first send: which ones the system fills, what counts as a touch, and who owns the record. Fields decided after launch mean re-mapping every record the campaign has already written.
Reporting is the last job, and it reads from that record instead of each sender's dashboard. Every tool counts opens, replies and interest its own way, so totals pulled from four dashboards do not add up. We build the reports with Claude and post them to Slack, split by channel and by segment.
The outbound sales tech stack in build order
Build it one job at a time, and prove each handoff on a small list before the next tool goes in. The order below is the one we use when a client's outbound tech stack starts from nothing.
- 01Write the ICP down as filtersBefore buying any data, turn the buyer description into fields a database can filter on.
- 02Source and validate one small listRun it through Prospeo and the final validation, then read what comes back by hand.
- 03Enrich and write for the clean records onlyClay and Apify add the reason to reach out now, and Claude writes from it. Records that failed validation never reach this step.
- 04Launch one channel as draftsPick the channel your buyers use most. A person reads the loaded campaign and switches it on.
- 05Wire the reply stop before a second channelConfirm a reply on the first channel lands on the CRM record and halts the sequence. Then add HeyReach or a second sender.
- 06Add calling where an account justifies itKixie goes in when a live conversation is worth more than another email.
- 07Automate the handoffs that cost timeAdd n8n once moving records between steps by hand takes real hours each week.
- 08Report from the CRMOnce every sender writes to one record, build the weekly report from that record.
You do not need all of these tools to start. Different combinations work for different clients, depending on budget, stage and channel, and the jobs matter more than the names. At the founder-led stage, a GTM stack for startups is lighter still: often one channel worked by hand, with Claude doing the research and the drafts.
Whether to build this stack in-house or buy it as a service is a separate build vs buy decision. Every tool named on this page has an entry in our tool directory, with its pricing and how we use it. Our outbound service builds and runs this stack for clients.
Which CRM is best for B2B sales?
For outbound, the best CRM is usually the one your team already works in, provided it has an API the sending tools can write to. Our builds mostly run on Attio or HubSpot. Agreeing the fields and the record owner before launch matters more than the brand.
Where does the outbound activity get logged?
On one CRM record per person. Each sending tool, whether email, LinkedIn or the dialler, writes its touches and replies back through an API. Anyone who opens the record can see whether that person has been contacted and how they answered.
What infrastructure do you use to plan, orchestrate and report campaigns?
Claude plans and orchestrates: it builds the ICP, writes the copy, assembles each campaign in its sender and classifies the replies. Clay handles enrichment and signals, the senders only send, the CRM holds the record, and reports are built with Claude and posted to Slack.
Can we run the tools ourselves later?
Yes. The tools can sit in your own accounts, and the enrichment workspace can be yours. When a team takes the system in-house, we hand over the documentation, recorded walkthroughs and training.

