Every marketing tech stack I inherit tends to look the same: forty logos on a slide, half of them paid for and a third logged into once a quarter. Tool sprawl isn't sophistication. It's usually a sign that nobody owns the question of what each tool is actually for, and over the years I've come to read it as the first thing worth fixing.
So this is the marketing tech stack for growth teams I actually run, the one I've refined while scaling growth at high-growth companies. I tend to organize it by job-to-be-done rather than vendor category, because that's closer to how the work really connects. Each layer gets what it's for, the tools I use, and one opinion you're welcome to argue with.
The principle: tools are a tax until they're not
Every tool you add is a recurring tax, paid in license, integration, training, and the quieter attention cost of one more dashboard someone has to remember exists. A tool earns its spot only when it does a load-bearing job that nothing already in the stack does well. Most teams buy tools to feel like they have a capability, whereas I've learned to buy only when a workflow is already breaking without one.
It helps to remember just how much there is to say no to. The martech landscape has ballooned into thousands of products over the past decade, a sprawl Scott Brinker has tracked year after year, and the modern best practice has shifted from collecting capabilities to consolidating them. The stack is a means, not an end, and the discipline that matters now is subtraction.
Experimentation and CRO: where the leverage actually is
This is the layer that compounds. A strong conversion lift doesn't fade; it sits under every campaign you run afterward. It's also the one layer where I tolerate some redundancy, because "why did this win?" is a question worth answering three different ways.
- Amplitude — behavioral analytics and the source of truth for what users actually do
- Optimizely — server-side and client-side experimentation at scale
- Mutiny — fast personalization and landing-page variants without an eng queue
- FullStory and CrazyEgg — session replay and heatmaps for the qualitative "why"
If I had to pick one opinion to defend here, it's that most teams over-buy testing tools and under-invest in watching. They'll pay for an enterprise experimentation platform and then guess at why the loser lost. FullStory and CrazyEgg are cheap relative to the engineering hours wasted arguing about a result you could have just watched a user fail at.
Analytics and attribution: trust beats precision
The job at this layer isn't a perfect number, it's a number the room trusts enough to act on. Attribution that's 90% right and universally believed will beat a model that's 98% right and litigated in every meeting, and I'd make that trade every time.
- Looker — the shared semantic layer and the dashboards leadership actually opens
- Google Analytics — web behavior and acquisition baselines
- Full Circle Insights — closed-loop attribution tying marketing activity to pipeline and revenue
My strong view is that multi-touch attribution is where a lot of growth teams quietly go to die. Most lack the data hygiene or deal volume to make a sophisticated model trustworthy, so they build a beautiful black box nobody believes. Closed-loop attribution that ties spend to revenue, the kind that surfaces real influenced pipeline you can point to, tends to beat any fractional-credit fantasy.
Marketing automation: orchestration, not blast radius
Automation's job is to put the right message in front of the right person at the right moment and then get out of the way. The failure mode I see most often is using it as a megaphone, more sends and more nudges and more noise, instead of as a conductor.
- Marketo — lead nurture, scoring, and the campaign engine
- Braze — lifecycle and cross-channel messaging for engaged users
- Outreach — sales sequencing where marketing and sales actually meet
- Pendo — in-product guidance and onboarding nudges
In my experience two automation platforms is usually one too many, and most teams somehow run three. If Marketo and Braze both email the same person this week and neither knows about the other, that isn't orchestration, it's two megaphones at one ear. The fix is to pick the system of record for each lifecycle stage before you buy the next tool.
CRM and ABM: aim before you spend
For any motion that touches sales, this layer decides where the money goes. Get the targeting right and everything downstream gets cheaper; get it wrong and you've simply automated waste at scale.
- Salesforce — the CRM and the single source of truth for accounts and pipeline
- Demandbase and 6Sense — intent signals and account scoring to focus ABM on accounts already in-market
What I keep coming back to is that ABM tooling is a force multiplier on a strategy you already have, not a substitute for one. Run it against a defined target list with real intent data and the leverage is real; ours ran roughly 7× the efficiency of unfocused demand gen. Buy it before you've agreed on who you're selling to, and all you've bought is a faster way to chase the wrong accounts.
Web and content: the surface everything lands on
This is the layer learners and prospects actually touch. Every other tool exists to drive someone here, so friction on this surface ends up taxing the whole system upstream, which is why I treat it with more care than its line-item cost suggests.
- Contentful — headless CMS so content isn't hostage to a page template
- Drift — conversational capture on high-intent pages
- Smartling — localization at scale for a global learner base
- Ceros and On24 — interactive content and webinars for the considered, higher-touch sale
The opinion worth stating plainly is that a headless CMS is overkill for a team publishing a few pages a month; you're buying flexibility you'll pay for in complexity. It earns its spot the moment content volume, localization, and multiple front-ends collide, which for a global product happens faster than most people expect. The discipline is matching the CMS to your real publishing velocity, not the velocity on the roadmap slide.
AI and productivity: the new connective tissue
This is the layer that changed most over the last two years, and it's where I now see the most expensive mistakes. AI's job in my stack is to compress the distance between a question and an answer, and between an idea and a shipped artifact, and I judge every tool here against that.
- Claude — drafting, analysis, and the reasoning engine behind the agents I build
- Codex — code and automation for the deterministic glue between systems
- MintMCP — connecting agents to the tools and data they need
- Reforge — frameworks and benchmarks to pressure-test thinking
- Airtable and JIRA — the operational backbone where work actually gets tracked
My advice here is to resist buying a dozen single-purpose AI point-tools. Most are a thin wrapper around a model you already have access to, and each one is another login, another bill, another thing to govern. A general reasoning model plus the plumbing to connect it to your real data will, in practice, beat a drawer full of AI gadgets that each do one trick.
This stack is deliberately smaller than what most growth teams of this size run, and you can see the live version of this stack and how it's grouped. The discipline was never in the logos; it's in refusing to add the next one until a real workflow demands it.
For the layer underneath, how these tools get wired into repeatable systems, read how I build systems with these tools and my AI experimentation framework. Broader context lives on the Sebastian Behar homepage. I keep this article maintained, so the stack here is the one I'm actually running.
