Stack overview

Layer	Tools	Role
Orchestration	Self-hosted n8n	Client workflows, content ingest, alerts
CI/CD	GitHub Actions	Site build, deploy, sitemap, IndexNow
Analytics	GA4 + GTM + Looker Studio	Reporting
Ads	Google Ads + Meta Ads Manager	Paid media
Content	Jekyll blog + custom import scripts	Publishing
AI assist	LLM APIs (task-specific)	Drafts, summaries — human edited
DNS/CDN	Cloudflare	Performance, SSL

Monthly cost envelope (agency, 2026)

Item	Approx USD/mo
VPS (n8n + misc)	12–30
LLM API usage	20–100 (varies by volume)
GitHub	0–4
Cloudflare	0–20
Email/CRM	0–50

Compare to $300–800+/mo in SaaS automation at scale on Zapier-only stacks.

Principles

1. Automate repetitive, verify important — client reports yes; publish without review no
2. One source of truth in Git — content and config in repo, not scattered SaaS
3. Measure cost per workflow — if automation saves <1 hr/month, skip it

Failure modes I’ve hit

• n8n upgrade breaking credentials → pin versions, staging instance
• AI draft pipelines publishing without human QA → now gated in _temp import runbook
• Over-automating SEO without quality bar → salvage/consolidate strategy instead of volume

Layer-by-layer detail

Orchestration (n8n)

Production workflows: lead routing, rank alerts, client report triggers, webhook glue between CRM and ad platforms. See n8n cost comparison.

CI/CD (GitHub Actions)

Build, sitemap, IndexNow, artifact pruning — CI/CD guide.

AI assist (boundaries)

LLMs used for: outline drafts, meta description variants, regex debugging. Not used for: publishing without review, fabricating client results, mass programmatic SEO (deprecated on this domain).

Content pipeline

Jekyll blog + import runbooks in docs/ (not public). Quality gate: human edit + FAQ schema + internal links to Nepal service pages.

Monthly cost scenarios

Agency size	Automation stack USD/mo
Solo consultant	15–40
5-person agency	40–120
15+ with heavy AI	120–300

End-to-end agency workflow map

How tools connect in a typical Gurkha Technology client engagement:

Lead (ads/SEO) → Landing page (static/WP) → GTM events → GA4
       ↓
Form/WhatsApp → n8n webhook → CRM row + Slack alert + email auto-reply
       ↓
Weekly: rank API → n8n → task if drop → human strategist
       ↓
Monthly: GA4 API → Looker Studio → account review call
       ↓
Content: outline (LLM assist) → human edit → Git/Jekyll → GitHub Actions deploy

Broken links in this chain — usually missing conversion tracking or no CRM dedupe — waste ad spend before AI ever enters the picture.

Tool selection by agency function

Function	Primary tool	Alternative	Monthly USD
Workflow automation	Self-hosted n8n	Make, Zapier	12–30
Static site CI	GitHub Actions	Cloudflare Pages CI	0–4
Analytics	GA4 + GTM	Matomo (self-host)	0
Reporting	Looker Studio	AgencyAnalytics	0–50
Ads	Google + Meta native	—	0
CRM	Sheets/HubSpot free tier	Pipedrive	0–30
AI drafting	LLM API (task-specific)	ChatGPT Team	20–100
CDN/DNS	Cloudflare	—	0–20
SEO monitoring	n8n + rank API	Ahrefs (single seat)	0–99

Deep dive: n8n vs Zapier TCO · SEO pipeline architecture

LLM usage policy (what we allow)

Use case	Allowed?	Guardrail
Blog outline from brief	Yes	Human rewrite before publish
Meta description variants	Yes	Character limit check in script
Client ad copy first draft	Yes	Strategist approval
Regex/debug in GTM	Yes	Test in Preview
Auto-publish to production	No	_temp import runbook gate
Fabricated case studies	No	Immediate termination
Mass programmatic SEO pages	No	Deprecated on this domain

Students ask about AI replacing juniors — in 2026 it replaces busywork, not accountability. Hire and train for GA4, client communication, and QA.

Workflow recipes (copy-ready)

1. New lead intake (15 min setup save per lead)

• Trigger: Website form webhook
• Steps: Validate email → dedupe CRM → assign owner by city → Slack #leads → SendGrid template to client → Log UTM to Sheet
• Stack: n8n + Sheets + Slack
• ROI: 20 leads/week × 10 min saved = 3+ hrs/week

2. Client rank drop alert

• Trigger: Cron weekly Monday 6am NPT
• Steps: Pull rank API → filter >3 position drop → create Asana task → optional LLM summary email draft
• Human gate: Strategist sends email

3. Post-publish IndexNow ping

• Trigger: GitHub Actions on main push affecting blog/
• Steps: Build sitemap → IndexNow API → Slack success/fail
• Doc: GitHub Actions CI/CD

4. Monthly client report prep

• Trigger: Cron 1st of month
• Steps: GA4 Data API → populate Looker template → notify AM in Slack
• Human gate: AM adds commentary before client call

5. WooCommerce low-stock alert (Nepal retail clients)

• Trigger: Woo webhook on stock change
• Steps: If stock < threshold → WhatsApp to owner → log to Sheet
• Note: Respect Meta WhatsApp API policies for template messages

Stack tiers by agency maturity

Tier 0 — Manual (0–2 clients)

Spreadsheets, manual posting, no GTM container — fix tracking first (GTM recipes).

Tier 1 — Foundation (3–10 clients)

GA4/GTM standardized, Looker templates, n8n for lead routing only. Budget USD 25–50/mo.

Tier 2 — Scaled ops (10–25 clients)

Git-backed content, CI deploy, rank monitoring automation, LLM assist for drafts. Budget USD 50–150/mo.

Tier 3 — Productized agency

Custom dashboards, multi-tenant n8n, API integrations per vertical (law, trekking, e-commerce). Budget USD 150–300/mo still beats legacy SaaS sprawl.

Nepal-specific considerations

• Wallet checkout debugging — automate alerts on failed Khalti/eSewa callbacks (gateway guide)
• Festival campaign scaling — pre-build n8n branches to pause low-ROAS campaigns when inventory flags low (e-commerce calendar)
• Load shedding / connectivity — VPS automation keeps running when office internet drops; local desktop scripts do not
• Client education — report in NPR, explain CPA in plain Nepali/English mix

Hiring and training stack skills

When hiring juniors (salary guide), test:

1. Can they read a Looker Studio dashboard?
2. Can they trace a lead from ad click to CRM row?
3. Do they know when not to automate?

Cost control rules

1. One orchestrator — n8n self-host; avoid Zapier + Make + n8n overlap
2. Measure USD per workflow hour saved — if <1 hr/month, skip
3. Pin versions — n8n, Node, Ruby (Jekyll) — upgrade on schedule, not reactively
4. LLM token budget — cap monthly API spend; use smaller models for classification tasks
5. Audit SaaS quarterly — cancel seats for churned clients

What we deprecated in 2025–2026

• Fully automated blog publishing without editorial QA
• Mass AI translation to Nepali without native review
• Zapier for high-volume multi-step flows moved to n8n (cost math)

90-day implementation roadmap for agencies

Month	Milestone
1	Standard GTM container + 3 conversion events
2	n8n lead router + Slack + CRM Sheet
3	Looker template + monthly GA4 API pull
4+	SEO alerts, content CI, LLM draft assist

Pricing what to charge clients for setup: digital marketing pricing Nepal — analytics/automation lines often NPR 10,000–25,000 setup + retainer.

Security incident patterns (learn from our mistakes)

Incident	Cause	Fix
n8n admin exposed to internet	No IP allowlist	Cloudflare Access + strong auth
LLM prompt leaked client name	Pasted full brief in public ChatGPT	Client-only workspace + API
GitHub token in workflow log	Debug print	Secret scanning, rotate token
Auto-tweet misfire	Wrong webhook trigger	Human approval node

Run quarterly automation audit — same discipline as WordPress plugin audit (TCO guide).

Client onboarding automation pack

Deliverables to automate on new retainer:

1. Create GA4 property + GTM container from template
2. Slack channel + CRM row + folder structure
3. Looker Studio clone from master report
4. n8n lead test webhook verification email to AM

Saves 3–4 hours onboarding; scales when hiring juniors (salary guide).

Vertical-specific stacks

Vertical	Extra tooling	Automation priority
E-commerce	WooCommerce webhooks	Stock + wallet fail alerts
Law firms	Call tracking	Intake form → CRM SLA
Trekking	WhatsApp lead scoring	Seasonal campaign pauses
Home services	GBP review alerts	Call + form routing

Playbooks: law firms · trekking · home services

Build vs buy for internal tools

Before custom scripting:

• Does n8n core node exist?
• Is Make/Zapier one-time connector cheaper at this volume? (n8n math)
• Will client pay for maintenance?

Default: automate the boring, keep strategy human.

Measuring automation ROI (template)

Workflow	Hours saved/mo	USD cost/mo	Payback
Lead router	8	18	Immediate
Rank alerts	4	18	<1 month
Report prep	6	18	<1 month
AI draft assist	10	40	2–3 weeks

If payback exceeds 3 months, defer automation — focus client budget on ads and SEO.

Future stack (watch list, not hype)

• GA4 BigQuery export for large e-commerce
• Server-side tagging on Cloudflare Zaraz
• AI agents for internal ops only (not client-facing without review)

Re-evaluate annually — 2024’s “must-have” SaaS often dies by 2026.

Tool integration map (detailed)

Google Ads ──┐
Meta Ads ────┼──→ Landing (WP/Static) ──→ GTM ──→ GA4 ──→ Looker Studio
Organic SEO ─┘                              ↓
                                         n8n webhooks
                                              ↓
                                    CRM / Sheets / Slack
                                              ↓
                                    LLM draft (internal only)
                                              ↓
                                    Human edit → Git → CI deploy

Broken arrow anywhere — usually GTM or n8n credentials — wastes the rest of the stack.

Onboarding checklist for new agency clients

Week 1:

• GA4 + GTM from template
• Conversion events validated (recipes)
• Ad accounts in client Business Manager
• n8n lead router tested with fake submit
• Looker dashboard cloned

Week 2–4: content CI, rank monitoring, report automation per SEO pipeline.

SaaS sprawl audit (quarterly)

List every paid tool — cancel if:

• Used <2×/month
• Duplicate feature (Zapier + n8n + Make for same flow)
• Client churned but seat active

Target <USD 100/mo tools for 5-person agency excluding ad spend.

Training internal team

Role	Must know	Nice to have
AM	Read Looker, scope automation	n8n edit basic flows
PPC specialist	GA4 conversions, UTM discipline	GTM Preview
Content	Git PR for static clients	LLM prompt templates
Dev	CI, n8n host, webhook debug	GA4 Measurement Protocol

Align hiring with career salary guide.

E-commerce client automation pack

Beyond lead gen:

• Abandoned cart webhook → WhatsApp template (policy-compliant)
• Low stock → owner alert
• Failed wallet payment → Slack + retry instructions
• Weekly GMV → Sheet for client email

Requires WooCommerce webhooks + n8n — budget in website cost maintenance line.

Content + AI pipeline (our runbook summary)

1. Strategist brief (topic, keyword, internal links)
2. LLM outline → human outline approval
3. Draft → human edit (tone, facts, Nepal context)
4. FAQ schema + front matter
5. PR review → CI → deploy
6. IndexNow ping via GitHub Actions

No step skipped — quality gate prevents programmatic SEO mistakes.

Monthly stack review meeting (internal)

Agency ops should spend 60 minutes monthly:

1. Review automation failure log
2. LLM API spend vs budget cap
3. Cancel unused SaaS seats
4. Pin/schedule n8n and dependency upgrades
5. One workflow improvement proposal from team

Document decisions — junior participation builds career-ready ops skills.

White-label vs client-visible automation

Approach	Pros	Cons
White-label	Client sees outcomes only	You carry ops burden
Client n8n login	Transparency	Support tickets from edits
Hybrid	AM view-only	Best for retainers

Default white-label for SME; hybrid for tech-savvy e-commerce with in-house dev.

Data residency and Nepali client concerns

Banks and clinics ask where data lives:

• Self-host VPS: disclose region (Singapore, Frankfurt, etc.)
• LLM API: avoid pasting patient/legal PII into public models
• Use DPA templates when handling EU/Nepali personal data

Competitive moat for agencies

Tools are copyable; workflow design + QA discipline is not. Moat elements:

• Standard GTM container library
• Vertical playbooks (trekking, law, home services)
• Git-backed content with human editorial
• Honest reporting culture

Undercutting on price while running Zapier at scale erodes margin — n8n math protects agency EBITDA.

Stack evolution timeline (Gurkha Technology)

Period	Change
2022–2023	Zapier-heavy, manual reporting
2024	Migrated high-volume flows to self-host n8n
2025	Git CI for static clients, IndexNow auto
2026	LLM assist internal only; strict publish gate

Document your own timeline — prevents re-buying deprecated SaaS.

Client-facing automation SLAs

Define in retainer appendix:

• Lead webhook delivery: 99% within 60 seconds
• Report automation: ready by 9am NPT on 1st business day
• Incident response: 4 business hours for P1 workflow down

SLA without monitoring is marketing — pair with Uptime Kuma alerts.

Tool substitution guide (budget cuts)

If client cuts budget, degrade gracefully:

Cut target	Replace with	Trade-off
Ahrefs	Search Console + manual spot checks	Less convenient
Zapier	Self-host n8n	Ops time
ChatGPT Team	API with cap	Less UI polish
AgencyAnalytics	Looker + n8n GA4 pull	Setup labor
Canva Pro	Canva free + templates	Brand consistency

Never cut conversion tracking first — that blinds all channels.

Measuring stack maturity (score 0–20)

Question (yes = 1 pt)	Score
GA4 conversions validated?
GTM in client-owned account?
n8n or equivalent for lead routing?
CI deploy for at least one client site?
LLM use gated with human review?
Monthly automation failure review?
Looker template standardized?
IndexNow or sitemap automation?
Vertical playbook documented?
Client offboarding export procedure?

Below 10: prioritize foundation before buying more SaaS.

Handoff documentation template (client exit)

When client leaves retainer, deliver:

1. GTM container export JSON
2. GA4 property Admin access confirmation
3. n8n workflow exports + credential rotation note
4. Looker Studio editor access
5. 30-minute recorded walkthrough

Professional offboarding generates referrals — messy exits do not.

Related reading

• Self-hosted n8n costs
• Automated SEO pipeline
• PPC Marketing Nepal