Outbound sales tech stack: what is broken
You have four subscriptions and outbound still is not working. The useful question is not which tool to add. It is which job nobody is doing, and which layer is lying to you about where the fault sits.
By Kshitij Maheshwari, co-founder · Updated August 2026 · 18 min read
Written by operators who run this stack for seed-stage teams, not by a vendor selling one layer of it.
What a stack actually is, and the part nobody names
Counting tools tells you what you pay for. It tells you nothing about what is running. A stack is a set of jobs, and one of those jobs has no product category at all.
An outbound tech stack is the set of jobs your outbound needs done, plus whatever is currently doing each one: a tool, you on a Tuesday night, or nobody at all.
Ask three vendors what a stack is and you get three shopping lists:
A stack is a database and whatever you plug into it, because the database is what they sell.
A stack is domains, mailboxes and the tool that sends the follow-ups, because that is what they sell.
A stack is one login, because the second login belongs to a competitor.
A stack is not what you own. It is everything that has to happen between "who should we write to" and "somebody replied", and the honest question is which of it nobody is doing.
How few tools you can get away with at the very start is its own question, answered in the minimum viable outbound stack.
Six jobs, and how you know one is missing
Each row is a job, what it has to produce, and the thing you would notice on a Tuesday morning if nobody were doing it.
A rule two people would build the same list from
Not a description of your buyer, a written filter. The symptom when nobody owns it: you can describe your customer out loud and still cannot produce 200 of them by Friday. Depth: writing an ideal customer profile.
An afternoon of your own time.
Real companies and real people, rebuildable
The symptom when nobody owns it: your list is a one-off export nobody could produce again, and it gets older every week you keep it. Depth: building the list.
Apollo Basic is $49 per seat per month billed annually, as of August 2026. Per seat, so it doubles when your co-founder logs in.
One fact per row, from a page you could open
The symptom when nobody owns it: every first line is interchangeable, and you could swap two prospects' emails without either of them noticing. Depth: enrichment, the data layer.
Clay's Launch plan is $167 a month billed annually, or $185 monthly, as of August 2026, more than the sender and data seat together.
Mail that arrives, from a domain you can burn
The symptom when nobody owns it: sends go out and nothing comes back, not even a no. Deliverability is just this job, done or not done. Depth: the sending build.
Smartlead's Base plan is $39 a month, ten Mailforge mailbox slots $30 at $3 each billed yearly, 50,000 MillionVerifier checks $89 once. As of August 2026.
An event with a date on it
Somebody is contacted this week because something happened to them. The symptom when nobody owns it: your good weeks are unexplained and you cannot reproduce them. Depth: signal-based selling.
RB2B is free at company level, then Starter at $79 a month for person-level visitors, as of June 2026.
Every reply seen, sorted and answered today
The symptom when nobody owns it: a good reply from three weeks ago that nobody ever answered. This is the job teams build last and lose the most to. Depth: reply management.
Attio is free up to three seats, then Plus at $29 per seat per month billed annually, as of June 2026.
A tool, you, or nobody
Every one of those six jobs is being done by a tool, by you, or by nobody. Nobody is the default, and it is the only one of the three that never sends an invoice.
A job done badly leaves something you can open and improve. A job done by nobody leaves nothing at all, which is why a budget review never finds one.
So audit the jobs, not the invoices. Put a name against each of the six.
The layer showing the problem is rarely the layer at fault
Layers hand work to each other, so a fault upstream surfaces downstream, wherever a number happens to be visible. These four are the ones that actively mislead.
| What you see | What gets blamed | Where it usually is | The check that settles it |
|---|---|---|---|
| You are landing in spam | The sending tool | The data. Bad addresses bounce, and bounces cost you reputation. | Open twenty addresses from last week's list by hand. |
| Nobody replies | The copy | The targeting. A better email to the wrong list is still the wrong list. | Read ten rows and say why each one should care this month. |
| The list feels bad | The data provider | The timing job. The rows are fine, the reason to write today is missing. | Ask what happened to these accounts in the last month. |
| The tool is broken | The tool | The join. Both tools work when you open them, the handoff dropped rows. | Compare the count that left with the count that arrived. |
The rule in one line: when a number goes wrong, walk one layer upstream before you buy anything.
The four purchases those symptoms talk you into
Each misread has a shop attached to it. Here is what the money actually buys when the fault is one layer up.
Bought to escape the spam folder. If the addresses are the problem, you have bought more places to burn, and each one needs weeks of warming up before it can send.
Bought because nobody replies. Copy is the last thing you touched and the cheapest thing to change, which is exactly why it gets blamed and rarely deserves it.
Bought because the list feels bad. When the missing job is timing, the second source hands you the same companies again, at a second monthly price.
Bought because something is broken. The migration takes a fortnight, and the connection that was actually dropping records still has to be built on the other side.
A tool bought at the wrong layer hides the fault
The price is not the damage here. You have done something about the problem and there is a monthly charge to prove it, so nobody looks again until renewal, and three months of sending go past before the real cause is touched.
The joins are a layer, and nobody sells them
The six jobs get all the attention. The lines between them get none, and the lines are where the work quietly goes missing.
Every stack diagram on the internet draws six or seven boxes and no lines between them. The lines are the seventh layer. They cost money, they have a meter, they fail without telling you, and on a two-person team they usually belong to nobody.
It fails without an error
A tool that breaks shows you a red banner. A connection that breaks produces an absence, and nothing anywhere counts the records that never arrived. You find out when somebody asks why last week was quiet.
One allowance, not one per automation
Apollo's own documentation says its limits are per team rather than per key or per user, counted in three windows at once, per minute, per hour and per day. HubSpot publishes the same shape for the private integrations you build yourself.
Find out the unit before you wire anything
Some charge per step, some per whole run, and the newer AI products get their own rate. Zapier's help center says an ordinary step is one task and a routed lead is five. The two meters are compared in n8n versus Make.
Make's Core plan is $9 a month for 10,000 operations, n8n's cloud Starter about $24, as of June 2026.
The automation that breaks is almost never the newest one. It is whichever one runs at the busiest hour, which is why the fault looks random and why people end up blaming their data provider. Read a vendor's limits page before its pricing page.
A worked diagnosis: one bad week, walked backwards
An illustrative walkthrough of the method, not a specific client result. We report real numbers only when they are real.
-
1Monday · The symptom
Bounces climb, replies stop
The sending dashboard is the only screen with a number on it, so the week gets blamed there. The purchase it invites: more mailboxes, more domains, more weeks of warming up.
-
2One layer upstream
Open twenty addresses by hand
Several are dead. Bad addresses bounce, bounces cost reputation, and sending is only where the meter sits. The purchase this invites: a second data provider.
-
3One more layer upstream
Ask where those rows came from
Verification ran, and then the handoff into the sending tool dropped part of the batch. Nothing errored. Nobody counted the rows that never arrived.
-
4The actual fix
A check, not a subscription
One line added to the weekly routine: compare the count that left the data tool with the count that arrived. Three purchases avoided, and nothing new to cancel later.
Want to know which layer is actually costing you meetings?
Book a Fit CheckThe buying order, and what triggers each purchase
Not "you need five tools". Each purchase has a trigger you can observe in a given week, and none of them is a funding stage.
-
1
Buy sending the week you would otherwise use your own domain
This is the only purchase with a deadline in front of it. Cold mail from the address your investors reply to puts everything on one reputation.
-
2
Buy data the second time you rebuild a list by hand
Once is research. Twice is a job. The trigger is the repeat, not the size of the list, and the hours you spent are the number to compare against the price.
-
3
Buy verification the first time bounces move at all
It is the cheapest line in the stack and it protects the two purchases above it. Buying it late is the most common order mistake we see.
-
4
Buy timing only after you have watched it free
The trigger is a week you cannot explain: something worked and you cannot say why. Watch the free sources first, and buy when triage genuinely overflows.
-
5
Buy wiring the week you are the integration
You are copying rows between two tools every Monday morning. Write down what the connection must produce before you build it, because that sentence is the part only you can write.
Which tool does each job best is a different question, and our GTM tools hub answers it. What one working version costs is priced in the seed-stage outbound playbook, at the vendors' own August 2026 list prices. Those are mostly US dollar prices, so read the shape of the bill rather than the total.
You pay for a tool three times
One of the three is on the website. The other two arrive later, and the last one grows the longer you stay.
The number you argue about. A seed stack is about $70 a month for domains, mailboxes and a sender, $120 to $170 with a data seat, at August 2026 list prices. On anything you keep past a quarter it is usually the smallest of the three.
Not the onboarding call. The afternoon your data tool and your CRM disagree about whether two rows are one company, and every hour after it.
Nobody prices this one, and it is the only one that grows while you sit still. It is also the reason the first year is the cheap year.
The economics of switching costs describes exactly this: firms compete hardest up front because leaving is expensive later, which is what an introductory offer is buying. Add the setup and the leaving to every quote you read.
The exit test, run before you buy
A trial is too short for a cold campaign to tell you anything, so the decision gets made on the demo. Spend twenty minutes of it leaving instead.
Judge the trial on the demo
Looks great, let's start Monday. We can sort the data side out later.
- ✕You tested the parts they built for demos
- ✕Nothing you learned is about leaving
- ✕The trial ends before your first honest campaign
Export four things on day one
Settings, Global Block List, three dots, Export List, CSV
- ✓Your records come out of nearly everything
- ✓Your rules and your history usually do not
- ✓The do-not-contact list is the one with legal weight
The file that matters most is the list of people who told you to stop, because leaving it behind is the one exit that has a legal edge to it. Smartlead documents a CSV export for its block list. Find out whether your sender does, before you need it.
One tool or several, honestly
Both halves of this are true at once, and every page you will read about it is published by somebody selling one of the two answers.
- ✓One record, one login, one bill
- ✓One fewer place for two things to disagree
- ✓A join you were never going to own disappears
- ✓Nothing to reconcile at the end of the month
- !The weakest part sets the ceiling for every job it covers
- !You cannot replace one part without replacing all of it
- !One exit now covers six jobs at once
- !Every page recommending it is selling one
Buy the bundle when the connections are what keeps breaking. Not because somebody told you that separate tools are expensive.
Who owns the stack when there are two of you
Two people split a stack by comfort, not by design. One of you likes the data side, the other likes the writing, and the split shows up in the tools.
Split the tools however you like. Then name one person who owns the handoffs between them, in writing, and give that person permission to change somebody else's setup to fix one. The piece neither of you considers yours is the piece nobody ever checks.
A 1968 paper about how organizations build things noticed the general version: what a group builds ends up shaped like the way that group talks. At two people that is one conversation, which makes this easy to fix and easy to skip.
There are two of us, so we divide the jobs and then put a name against every handoff between them. The rule we hold to is that a handoff nobody owns is not a small gap, it is the place work disappears without anybody deciding to drop it.
When to stop adding
Tools add up. The places two things have to agree multiply, which is the whole argument against a bigger stack.
is how many places two tools can have to agree once you own five of them.
Three tools make three possible pairs. Four make six. Five make ten. You will never wire them all, and every pair you skip is still a decision somebody makes and re-makes.
Stop when the next connection costs more attention than the job it saves. The cost of a stack is not the sum of the subscriptions, it is the number of places two things have to agree, and that number grows faster than the invoice does.
Where the common advice is wrong
Almost every stack article is published by a company that sells one of the layers in it, and the advice bends in the same direction every time.
"Consolidate everything, buy these seven tools, and audit your stack every quarter."
- ✕Fragmentation is the problem
- ✕You need one tool per category
- ✕Review the stack on a calendar
- ✕The stack is what you own
"Consolidate where the join keeps breaking, cover six jobs, and audit the week a number moves."
- ✓The unowned job is the problem, not the tool count
- ✓A category is a shop's shelf, not a job
- ✓A quarterly review finds last quarter's fault
- ✓The stack is what runs, including the parts nobody bought
What is true right now, and what is just current
Four things changed in the last two years. The last column is how long each one is worth trusting, as of August 2026.
| What changed | What it means for you | How long to trust it |
|---|---|---|
| The bill moved from the seat to the meter | Size a plan from the campaign you are about to run, never from how many people will log in. | Durable. The units will keep changing. |
| The agent layer is priced as its own product | The same job can cost more or less depending on which door it comes through, on one platform. | Current practice, not a law. Check it yourself. |
| Platforms publish their limits | Read a vendor's limits page before its pricing page. It tells you more about month three. | Durable, while they keep publishing them. |
| Getting your records out became normal | The lock-in moved to the rules, the history, the sending reputation and the block list. | Durable. The exit test is how you check. |
A job nobody is doing still sends no invoice. A connection that fails still fails quietly. And the layer showing you a bad number is still, most of the time, not the layer that produced it.
What to realistically expect
Fewer tools than the lists tell you, more wiring than you planned for, and a stack that is never entirely healthy.
Two people can cover all six jobs with a small stack and a written routine. The gap between a working motion and a stalled one is almost never the number of subscriptions, and adding one has never yet fixed a job nobody was doing.
Building a connection got cheap, and noticing that one stopped did not. Budget the attention, not the build: somebody has to see the afternoon it quietly stops, and no tool has made that part cheaper.
A stack running a bit broken is the normal state, not a failure. The safety engineering literature has said for decades that complex systems always run partly degraded. What separates a working stack from a failing one is whether you find out.
Key takeaways
If you keep one page of notes from this guide, keep these five lines and the six jobs they refer to.
- 1 A stack is six jobs plus the wiring, not a set of subscriptions.
- 2 Every job is done by a tool, by you, or by nobody. Audit the names.
- 3 Walk one layer upstream before you spend anything.
- 4 You pay three times: the price, the setup, and the leaving.
- 5 Stop when the next connection costs more attention than it saves.
Questions founders ask
What is an outbound sales tech stack?
Which tool should I buy first?
My emails are going to spam. Do I need a better sending tool?
Nobody replies. Is my copy bad?
Should I buy one all-in-one tool or several separate ones?
Why did my automation stop working when I added another one?
What should I test before I commit to a tool?
Co-founder of Real Good GTM. He has been the first business hire and Chief of Staff at seed-stage B2B startups, building outbound pipeline before any playbook existed. This guide is the diagnosis he runs before touching anybody's tools: which of the six jobs nobody is doing, and which layer is lying about where the fault sits.
Connect on LinkedInFrom diagnosis to build
You know which layer to look at. These three cover building the motion as a system, what one working stack costs, and which tool does each job.
GTM engineering
Building the motion as a system rather than a set of tools: what the discipline is, and whether to build it, hire it, or rent it.
Read the guideThe seed-stage playbook
One working stack, priced line by line at the vendors' own list prices, inside a first ninety days you can actually run.
Read the postBest GTM tools
Sixteen guides grouped by the job they do, for once you know which layer you are actually buying for.
Compare the toolsWant the broken layer found for you?
Book a fit check. We'll look at which of the six jobs are being done, which are being done by nobody, and whether the fault is where you think it is. If outbound is not the right motion for your stage, we'll tell you that too.
Book a Fit CheckNo hard sell. No fake numbers. Real good work speaks for itself.