We get asked a version of this constantly: “if you were starting over, what would you buy?” So here it is, concretely: the exact stack we’d assemble for a five-person services agency starting tomorrow, with real prices, the order to deploy it in, and the things we’d deliberately not buy. This is the distillation of what we’ve deployed for clients — the same choices we’d spend our own money on.

The stack, with costs

Layer Tool Plan Cost/month Why this one
Automation n8n Cloud Starter→Pro $24-60 (or $10 self-hosted) Execution pricing survives volume; we’ve run the comparison
CRM HubSpot (free) → Pipedrive Free→Essential $0-15/seat Start free, upgrade when pipeline needs it
Scheduling Cal.com Free/Team $0-15 Open-source, embeds anywhere, no per-booking cost
Email sending Resend Free→Pro $0-20 Transactional deliverability; the thing free email tools do worst
Forms Tally Pro $0-29 Best free tier in forms, one-person-friendly
Documents/contracts PandaDoc or DocuSign — $0-30 E-signature is table stakes by year one
Accounting Wave → QuickBooks Free→QB $0-19 Wave free tier covers the first year cleanly
Project/tasks Notion or Linear Free tier $0 5 seats fit free tiers in both
Communication Slack Free $0 Free tier is fine at 5 people
Total ~$60-190

The point of the table isn’t the specific numbers (they drift; we re-verify quarterly) — it’s the shape: a five-person agency’s entire operational software stack costs less than one person’s daily lunch budget. The expensive line items in agency software (enterprise CRM seats, “client portal” platforms, all-in-one agency SaaS at $300+/month) are almost always paying for scale you don’t have or features you won’t use.

Why n8n as the spine

The automation layer is the one strategic choice in the stack, because everything else plugs into it. Our reasoning, compressed from the full comparison:

  1. The volume curve. Agencies multiply automations: client reporting, lead intake, invoicing, onboarding, content. Per-task and per-operation pricing (Zapier, Make) scales linearly with that growth; n8n’s execution model — or a $10 self-hosted VPS with unlimited executions — stays flat. At five-person-agency volume, the difference is hundreds of dollars a month within the first year.

  2. The complexity curve. Client work means messy data transformation — exactly what n8n’s code nodes and error transparency handle best. When a client’s lead feed changes schema at midnight, you want the tool whose failures are loud.

  3. Self-hosting as a client-sellable option. The ability to say “your data stays on our server” (or your client’s server) wins deals in regulated or privacy-conscious industries. That’s a n8n capability, not a Zapier one.

Make is the right answer for teams that are less technical, and we still recommend it for that profile — the comparison covers the trade-offs honestly. But for the “we’d assemble this tomorrow” stack: n8n, self-hosted from month three.

The deployment order (this is the part everyone skips)

The tools don’t matter as much as the sequence, because the sequence is what determines whether the stack actually gets built:

Week 1: The spine. CRM (free HubSpot), Cal.com, Tally, and the first n8n flows — the lead routing spine: form → CRM → Slack notification with acknowledgement email. One flow, running in parallel with manual process for a week. Total cost: $0-24.

Week 2-3: The money. Invoice reminder automation on Wave, plus the document-collection flow (onboarding lite). Cash flow and client experience improve before any fancy tooling exists. Still under $50/month.

Month 2: The delivery side. Project workspace, the onboarding sequence full build, weekly reporting automation (the single highest-ROI agency automation: the 3-hour weekly report → 2 hours saved and zero errors).

Month 3: Growth. Content cross-posting, the research agent for prospect and market monitoring, self-hosting migration once volume justifies it.

The principle behind the order: fix money-in and money-flow before you optimize anything else. Lead response, invoicing, onboarding — the three automations that touch revenue directly — are weeks one through three. The cool stuff (agents, content) waits its turn.

The 90-day plan, week by week

The deployment order above is the compressed version. Here’s the extended one we hand to a five-person agency starting from zero — same logic, with the checkpoints that keep a rollout from stalling at week three (where most do-it-yourself automation programs quietly die).

Days 1-7 — intake spine. Build the lead routing spine: form → CRM → Slack notification → acknowledgement email. Run it against a test sheet in parallel with the existing process. Checkpoint at day 7: the automation has fired correctly on every inquiry, and the team has stopped checking the form manually. If it’s not boringly reliable yet, fix this before building anything else — everything downstream plugs into it.

Days 8-14 — money flow. Invoice reminders go live on the accounting stack, with the skip-rules (custom schedules, disputed invoices, procurement clients) configured on day one. This is usually the first automation that pays for the entire stack — one avoided late-payment month typically covers a quarter of tooling.

Days 15-30 — client experience. The onboarding sequence (welcome, scheduling, document collection, internal checklist) plus the weekly client-report automation. By day 30, a new client can sign and be fully onboarded without anyone remembering to send anything. Checkpoint at day 30: three automations running, zero manual reminders, and the first “how are you this organized?” from a client.

Days 31-60 — the internal half. Weekly pipeline roll-up, the task-assignment nag (“this deliverable is due in 48h and nothing’s in the folder”), and the first research digest for client industries. This is also when the volume decision arrives: if operations-per-month is pushing against the tool tier’s ceiling, migrate the automation layer to self-hosted n8n over a weekend. Checkpoint at day 60: six to eight automations; run cost flat at $10-30/month; the team has stopped asking “did anyone reply to that lead?” because the Slack notification answered it.

Days 61-90 — the growth layer, earned. Content cross-posting with a review queue, and the first small agent — usually the research digest upgraded to per-client briefs — running with full guardrails. By day 90 the typical five-person agency has 10-12 automations, and the practice — not the tool list — is the asset.

Two rules keep the 90 days honest. First, one build at a time until it’s run clean for two weeks — parallel builds mean parallel debugging sessions, and that’s how rollouts stall. Second, every automation gets a maintenance owner — a named person who gets the error notifications. Unowned automations decay; owned ones compound.

The objections this stack invites (and our answers)

“My team won’t adopt half of this.” They don’t have to adopt all of it — that’s the point of the spine-plus-arrows shape. The automations run whether or not anyone “uses” them: leads get acknowledged, invoices get chased, reports go out. The only adoption the stack actually requires is the weekly ten-minute review queue (drafts waiting for a human) and the maintenance owner answering error notifications. Everything else is invisible infrastructure, and invisible infrastructure is the kind that survives office politics.

“Won’t it be outdated in a year?” The tool list, partially — prices drift and features shuffle, which is why we re-verify quarterly. The shape ages much better: unbundled best-in-class layers, a workflow engine as the spine, and free tiers until volume forces an upgrade. We’ve watched this shape survive multiple pricing overhauls across the industry because the exit cost of any single layer stays near zero — swap the CRM, keep the workflows; swap the forms tool, the flows rewire in an afternoon. That swap-ability is the actual insurance policy, not any vendor’s roadmap.

“We’re not five people — we’re two.” The stack barely changes. Two founders get the same spine (form → CRM → notification → acknowledgement), the same money flows (invoice chasing matters more when cash reserves are thin), and the same delivery automations — what changes is the seat count on the free tiers, which is to say the price goes down. The deployment order in the 90-day plan is identical; only the internal-Slack-channel step collapses, and the maintenance owner is both of you. We’ve run this exact stack for solo operators at under $100/month total, and the freelancer playbook covers that profile in detail.

What we’d deliberately not buy

  • The “all-in-one agency platform.” $200-400/month for a bundle of features you’ll use 20% of, with automation capabilities that are 10% of what a dedicated tool does. The unbundled stack above is both cheaper and better at each layer.
  • Enterprise CRM tiers. Five people don’t need pipeline forecasting AI; they need a pipeline that’s actually current. That’s an automation job, not a subscription tier.
  • Meeting schedulers with per-booking pricing. It’s a calendar embed. Cal.com is free and better.
  • Any AI tool with a per-seat “AI assistant” charge. The useful AI at agency scale lives inside your workflows — summaries, drafting, research — costing cents per run via API, not $30/seat/month for a chat window on top of software.

The twelve-month view

Month 12, with this stack fully deployed and the ROI discipline applied to each build, a five-person agency typically has: 10-15 automations running, 15-25 hours/week reclaimed from admin, response-time and invoicing metrics that were previously unmeasurable, and a tool bill of $60-190/month. The stack paid for itself by month two; everything after that compounds.

And the meta-point, the one we’d tattoo on the invoice if we could: the stack is cheap; the habit is the asset. The agency that deploys this stack in the week-by-week order above isn’t buying software — it’s building the operating practice of “we don’t do repetitive things twice.” That practice is worth more than every line item in the table, and it’s the only part of the stack that doesn’t show up on a bill.