The Tech Stack For A Small Business, And Why It Is Small

Why your stack got big without anyone deciding

Nobody chose to have twenty-six subscriptions. It happened the way every stack happens: something annoyed you, a tool fixed it, you signed up. Each decision was locally correct. The sum of thirty locally correct decisions is an operation nobody can describe, where the same client exists in nine systems under three spellings and the truth about what you said to them is distributed across four inboxes and a Slack DM.

The cost you are paying is not the subscriptions, which is why the annual audit where you cancel four of them changes nothing. The cost is seams. Every boundary between two tools is a place where your business can be wrong without telling you: the booking tool that does not know a deal closed, so the nurture keeps nurturing a client who signed on Tuesday. The form that writes to a field nothing reads. The spreadsheet that exists purely so an automation can remember something. Every seam is either a sync you own forever or a fact that quietly diverges.

And there is a second cost that is worse and harder to see. Each tool holds a piece of your operating logic — a filter, a threshold, a mapping, a rule — described nowhere else. Rename a field in one and something silently stops being true in another, and nothing in your business can tell you which. You do not have a stack. You have a body of undocumented law distributed across twenty-six vendors.

So the goal is not the cheapest stack or the most powerful one. It is the smallest one with the fewest seams, because seams are where founder-led businesses actually break — not at the tools.

The seven slots

Think in slots, not tools. A slot is a job your business genuinely has. There are about seven, and every tool you own either fills one or is a tab.

The record. Where your contacts, opportunities, pipeline, conversations and calendar live, as one object. This is the spine and it is the most consequential slot. We use GoHighLevel for founder-led service businesses, because the workflow engine lives inside the data — a workflow triggers on an appointment, evaluates a field a form wrote, moves a stage, and messages the thread the human was already using, with no integration and no sync. If you have a RevOps owner or an enterprise sale, this slot is HubSpot's and you should hire someone to run it.

The knowledge. Your SOPs, stage definitions, qualification criteria, ICP, the reason each automation exists, your voice reference. Notion, because it is prose a system can query, and because a definition that lives only in a tool's configuration cannot be argued with, improved, or taught to your next hire.

The automation layer. Where events become actions. Zapier for lines, Make for graphs, n8n or custom code for anything with state, iteration, real retries, or a silent failure mode. Most businesses need two of those three and think they need one.

The intelligence. Claude, ChatGPT, the OpenAI API — called from the automation layer on a trigger, not opened in a browser tab. Classification, drafting, summarizing.

The interface. Slack, where your systems hand control back to a human at the one moment a human is required, with one name tagged, one link, and one action.

The view. Where numbers get read — a Retool view, a Notion view, a Slack digest. Small, and downstream of everything else.

And money. Payments, invoicing, books. Deliberately last, deliberately boring, and the one slot where 'whatever your accountant likes' is a genuinely correct answer.

The rule for adding anything, and the tools that are not slots

Name the slot, name what it evicts. If a new tool does not replace an incumbent, it is not filling a slot — it is creating a new place where a piece of your state can live, and you have just added a seam. This rule sounds bureaucratic and it takes four seconds, and it is the only thing standing between you and twenty-six subscriptions.

The tools that are not slots, and are worth naming because they are the ones that sneak in. A second place tasks live. A second place clients are recorded. A third inbox. A tool that does one job a slot-filler already does adequately — 'adequately' being a word founders refuse to accept and should learn to love, because adequate-and-integrated beats excellent-and-adjacent every single time in a business with no system owner.

The strongest form of the rule: prefer coherence over capability. For a founder-led business the scarce resource is not features, it is attention, and every tool costs attention permanently. A best-in-class tool for each of nine jobs produces a business that is worse at all nine, because the state is fragmented and you are the reconciliation layer. That is the trade, and it is not close at this size. It reverses at some scale — but that scale has a RevOps person in it, and you do not have one.

The exception worth honoring: if a vertical system runs your delivery — practice management, dispatch, a case system — it is a slot, it is non-negotiable, and it should probably dictate the record slot too. Go where a maintained integration already exists. An integration you own yourself is a liability with a birthday, and you will meet it on a Sunday.

Where the seams go, and how to make them survivable

You cannot get to zero seams. What you can do is put them where they are cheap and make them legible. Cheap means at a system boundary — the record hands to the automation layer, the automation layer hands to Slack — rather than in the middle of a thought. The specific catastrophe is a process that starts in a Zap, hands to a Make scenario, and comes back to a different Zap: nobody can trace it, no error surfaces coherently, and the person debugging it at 11pm does not know three tools are involved. One process, one tool, one owner.

Legible means every seam is written down. An inventory in Notion: every automation, what it does in one sentence, what it triggers on, what it reads, what it writes, what it assumes, who owns it. Ten minutes each. That inventory is the artifact that turns 'an unknown number of things might have broken' into a number you can look up in ninety seconds, and it is the entire difference between a stack you can change and one you are afraid of.

And the definitions come out of the tools. When the same threshold appears in four automations, it is not a threshold — it is four thresholds that currently agree. Pull it into Notion and let the automations read it from one place. That single move is what makes a stack maintainable by a founder rather than by whoever built it.

Then the order of operations, which is what people actually want and rarely get: record first, and only the pipeline part. One intake surface. One no-touch timer escalating to one named human in Slack. Live there two weeks — the timer will tell you which stage genuinely stalls, and it is almost never the one you guessed, which means every automation you would have built first was aimed at the wrong place. Then knowledge: the five things people ask you most, written as SOPs with owners. Then the automation layer, for the seams that are now obvious. Then intelligence, embedded on triggers that already exist. The view comes last, because a dashboard built before the plumbing is a human pasting numbers on Monday, and it goes stale in five weeks.

The thing the stack cannot do

None of this is a system. This is the part that costs founders the most and it is why stack posts are a genre: the stack is where a system lives, and buying the stack does not produce the system. You can own every tool on this page, correctly configured, and still have deals go quiet for six weeks, because the thing that catches a quiet deal is not software. It is a stage with an exit event a stranger could verify, one named owner, and a threshold that interrupts a human on the day.

That is why the tool questions feel so urgent and matter so little. In 17+ years of building operations and 1,400+ founder interviews on The Wantrepreneur to Entrepreneur Podcast, I have not once seen a business fail because of its stack. I have seen a great many stall because nobody could say what makes a deal leave the Proposal stage, or who owns a lead that arrives at 4:40pm, or what done looks like for onboarding. Doing more doesn't create growth. Designing better does.

And the sequencing that follows from that: do it manually first. When you do something manually first, you feel its texture — the friction points, the moments that matter, the little places where care hides. That nuance is the data that makes your automation great. A founder who buys the stack and then designs the process builds something that works and is subtly, permanently wrong.

So: seven slots, one tool each, a new one has to evict an incumbent, seams at boundaries and written down, definitions out of the tools and into Notion, and the whole thing built in the order above rather than all at once. Then spend your actual energy on the stage definitions, the owners, and the triggers, because that is where the business is. Build the systems as you do the work, not instead of the work. If you want the map before you buy anything — what you actually run, what should be deleted rather than automated, and what to build first — that is what an OPERATE Report ($1,997) is. If the map is clear, a Build Day ($5K/day) makes a slot real, and a Custom Build, quoted per engagement, does the whole spine.