You could rebuild Prewiro in n8n. Here is what you would actually be signing up for.

TL;DR

The fair version of the question

The most common question I get from technically sophisticated prospects is some variant of this: "Why pay you when I can wire this up in n8n in a weekend?"

It is a fair question. n8n, Make, and Zapier have made workflow automation effectively free. If a workflow can be expressed as "when this thing happens, do that thing," it can almost certainly be built in one of those tools, and built quickly.

So I am going to answer the question seriously rather than dodge it. The honest answer is that you can rebuild parts of Prewiro in n8n, and for some teams that is the right call. The parts you cannot rebuild — and the cost of trying to — are what this post is about.

If you have not read the playbook, the cadence Prewiro runs is six days of completed artefacts delivered before any pitch — six pieces of real work, picked per prospect from a library of skills (a tailored website, SEO/AEO content, a competitor brief, brand-voice social posts, an AI brand video, a 30-second voice note, an influencer shortlist, a free sample shipment, an ROI brief, a tailored case study). Two prospects never receive the same six.

For the n8n analysis below, we will walk through the typical service-business mix as a concrete example:

• Day 1: A live, deployed website for the prospect's business
• Day 2: SEO and AEO blog content published on it
• Day 3: A competitor analysis
• Day 4: Ready-to-publish social posts
• Day 5: An AI-generated brand video
• Day 6: A relevant local micro-influencer shortlist

Below is what each of those would look like as an n8n project. Bear in mind that this is one of the mixes the engine produces — a real n8n rebuild would also need a per-prospect decisioning layer that picks which six skills to deploy for each lead, which we cover at the end of the post.

Day 1: the parts list to build a website-generator workflow

What you would need:

• An LLM with strong site-structure reasoning (Claude, GPT-4-class). Prompted with the prospect's industry and offerings to generate page structure, copy, and CTA placement.
• Image sourcing or generation. Stock APIs (Unsplash, Pexels) for category-appropriate images, or a generative image model for hero shots and product visuals.
• A renderer. Either a static site generator (Astro, Eleventy, Next.js exporting static), or a headless CMS that takes the generated content and produces a deployable site.
• Hosting + DNS. Vercel/Netlify for hosting, plus a per-prospect subdomain assignment scheme. If you want the prospect to use their own domain you also need DNS automation.
• Visual QA. A screenshot-and-diff step that catches broken layouts before the site goes live to the prospect — because shipping a broken Day 1 destroys the entire cadence.
• A "is this good enough?" gate. This one is the killer. n8n cannot tell you whether the website it just generated is good enough to send to a real customer. Either you build a vision-model evaluator, or you have a human review every site, or you ship without quality gating and accept the risk.

Realistic engineering effort, end to end: 8 to 12 days for a working version, plus another week for the human-review queue if you choose that path.

Recurring cost: hosting roughly 50–150 USD/month at moderate volume, generative images roughly 100–400 USD/month, LLM calls 200–500 USD/month per several hundred prospects.

Day 2: SEO + AEO content is two pipelines, not one

In 2026, SEO content and AEO content are not the same thing.

SEO is what gets crawled by Google for the blue-link results. The structures that win are familiar — keyword-targeted titles, headers, internal links, schema markup, freshness signals.

AEO — answer engine optimisation — is what gets cited when a prospective customer asks ChatGPT, Claude, Perplexity, or Google AI Overviews a question. The structures that win are different: direct-answer paragraphs in the first 100 words, question-shaped headings, fact-shaped citable sentences, FAQ blocks, structured data emitting FAQPage and HowTo JSON-LD.

To do Day 2 well in n8n, you would need:

• Topic research. Either a paid keyword API (Ahrefs, Semrush) or scraping plus your own scoring.
• Two content generators in sequence. One tuned for SEO structure, one for AEO answer-extraction patterns. They are different prompts and different post-processing steps.
• On-page SEO QA. Title length, meta description, heading hierarchy, internal link insertion, image alt text.
• AEO QA. First-100-words direct-answer check, citable-sentence density, FAQ schema validation.
• Publishing pipeline back to the Day 1 site. This needs to integrate with whatever you chose as the renderer in Day 1.
• Indexing trigger. Submitting to Google Search Console; pinging IndexNow.

Engineering effort: 6 to 10 days. Recurring cost: keyword research APIs 100–300 USD/month, LLM calls 100–300 USD/month.

Day 3: competitor analysis means building a scraper

For competitor analysis you need:

• Competitor discovery. Given a business and a city, identify three to five competitors. This is itself non-trivial — Google Places API, scraping, plus an LLM to filter the noise.
• Public data extraction. Scraping competitor websites, their pricing pages, their social presence, their review profiles. Each platform has its own anti-scraping defences and its own rate limits.
• Structured comparison. An LLM step that takes the raw scrape and produces a comparison artefact — gaps, opportunities, specific weaknesses.
• Output formatting. A PDF or shareable doc the prospect can actually read.
• The "is this useful?" gate. Same problem as Day 1. n8n cannot tell you whether the analysis it produced is genuinely useful or just generic. Either you evaluate every output, or you accept that maybe 30 percent of Day 3 outputs will damage the prospect relationship.

Engineering effort: 8 to 14 days. The scraping reliability piece is the time sink — every platform breaks scrapers in slightly different ways, and maintaining the scrapers is a permanent ongoing cost.

Recurring cost: scraping infrastructure (proxies, headless browsers) 200–500 USD/month, LLM calls 100–300 USD/month.

Day 4: brand voice cannot be a single LLM call

Day 4 is where the workflow-tool model breaks hardest. Generating "social posts" is easy. Generating social posts in the prospect's brand voice — the voice you established on Day 2's blog post and that the prospect's existing public content reinforces — requires memory across the cadence.

n8n does not have memory across the cadence. Each workflow run is an isolated execution. To get brand-voice consistency, you would need:

• A per-prospect memory store. A vector DB or a structured profile that captures the brand voice from Days 1 and 2 plus any of the prospect's own public content.
• A retrieval step at the start of Day 4. Before generating, pull the prospect's voice profile from the store.
• Per-platform formatters. Instagram captions are not LinkedIn posts. Hashtag conventions vary. Image aspect ratios vary. You need a formatter per target platform.
• Image generation tuned to the brand. If the prospect has a colour palette, the generated images should respect it.
• A "this matches the brand" gate. Same theme. Either evaluate every output or accept a fail rate.

Engineering effort: 10 to 16 days, the bulk of which is the memory store and the per-platform formatters.

Recurring cost: vector DB 50–150 USD/month, image generation 200–500 USD/month.

Day 5: brand video is a render pipeline, not a workflow node

Generative video has gotten dramatically cheaper, but "cheaper than ₹50,000 per video" is not the same as "a node in n8n."

To produce a Day 5 video you need:

• A script generator. Tuned to the prospect's industry, the platform the video is going on, and the duration you are targeting.
• A voice generator. ElevenLabs, OpenAI TTS, or similar. With voice selection logic — a clinic does not get the same voice as a sneaker reseller.
• A visuals pipeline. Either generative video (Runway, Sora) or stock footage with an LLM-driven shot list. Generative video is the better artefact and the more expensive path.
• A render and assembly step. ffmpeg, or a cloud rendering service, with subtitle burn-in and brand colour treatments.
• Format outputs. 9:16 for vertical platforms, 16:9 for horizontal, 1:1 for feed. That is three render passes per video.
• A "this is not embarrassing" gate. Particularly on video. A bad Day 5 video does not just fail to convert — it actively damages the relationship.

Engineering effort: 10 to 18 days to a working pipeline. Generative video is the most volatile category in this entire list right now; whatever you build will need to be migrated to a new model within 3 to 6 months.

Recurring cost: generative video 400–1,000 USD/month at moderate volume, voice 50–150 USD/month, rendering infrastructure 50–150 USD/month.

Day 6: influencer matching needs a database you do not have

For influencer matching you need:

• A discovery database. Public influencer-discovery APIs exist (Modash, Heepsy, etc.) but they are expensive and have coverage gaps in non-US markets. Or you build your own scraper, in which case you are signing up for a much larger project.
• Niche and geo filtering. Matching the prospect's industry and city.
• Engagement quality scoring. Follower count is meaningless without engagement signal. You need to compute or fetch this per influencer.
• Brand fit scoring. An LLM step that evaluates whether each influencer's content fits the prospect's brand.
• Outreach template generation. A short, personalised intro the prospect can copy-paste.

Engineering effort: 6 to 10 days if you use a paid discovery API; 20+ days if you build your own.

Recurring cost: influencer discovery API 200–800 USD/month, LLM scoring 50–150 USD/month.

Now add it all up

If you map every day onto an n8n project:

Day	Effort (days)	Recurring (USD/mo)
1 — Website	8–12	350–1,050
2 — SEO + AEO content	6–10	200–600
3 — Competitor analysis	8–14	300–800
4 — Social posts	10–16	250–650
5 — Brand video	10–18	500–1,300
6 — Influencer match	6–10	250–950
Orchestration + observability	10–15	100–300
Total	58–95 engineering days	1,950–5,650 USD/mo

That is the build cost for the typical service-business mix above. The numbers vary with team skill, scope, and how much you care about quality gating. The honest middle estimate is roughly 70 engineering days and ~3,000 USD/month in recurring spend, before any human-review labour.

But this total assumes you only ever ship that one mix. The real engine selects six artefacts per prospect from a wider library — which means a real n8n rebuild also has to build pipelines for the artefacts not in the table above (a 30-second voice note, a sample-shipment trigger, an ROI brief, a tailored case study, retail product photography, neighbourhood market reports for real-estate prospects), plus a decisioning layer that picks which six to deploy for each lead. Realistically, that is another 15 to 25 engineering days plus a per-month run cost similar to the orchestration line — bringing the honest end-to-end build closer to 85 to 120 engineering days.

For most solopreneurs and small agencies, those numbers eliminate the build option immediately. For larger teams, the build option is real — but is "85+ engineering days plus a permanent maintenance burden" actually the best use of your engineering team?

The five things workflow tools structurally cannot do

The numbers above are the easy part of the argument. The harder argument is that even a well-funded build will not produce the same system, because workflow tools have five structural limits.

1. Judgement

Every quality gate in the cadence — is this website good enough to ship, is this competitor analysis genuinely useful, does this video meet the brand bar — is a judgement call. Workflow nodes do not make judgement calls. They execute deterministic logic against structured inputs.

You can simulate judgement with an LLM-as-judge step, and we do exactly that inside Prewiro. But the LLM-as-judge logic is a system in its own right — prompt engineering, evaluation criteria, calibration, escalation rules. It is not a node you add to a flow.

2. Memory across the cadence

Day 4's social posts depend on the brand voice learned on Day 2's blog. Day 6's influencer match depends on the positioning identified on Day 3. The whole cadence is stateful across six days.

n8n workflows are stateless by default. Each run is isolated. Adding cross-run memory means building a per-prospect data store, a retrieval layer, and a state machine — which is the bulk of what an agent framework provides out of the box and a workflow tool does not.

3. Escalation

Sometimes Day 1 fails. The prospect's industry is unusual, the website generator hallucinates a feature that does not exist, the screenshot QA flags a layout break that the LLM cannot fix on its own.

A good agent system pauses, asks a human, and resumes. n8n nodes throw an error and stop. You can wrap them in retry logic, but retry logic is not the same as escalation — escalation requires the system to know what kind of failure happened and who to escalate to.

4. Maintenance

This is the under-discussed killer. The stack you assemble on day one will not be the stack you have in six months.

• The image model you use today will be deprecated; the prompt format will change
• The scraper you wrote will break when LinkedIn changes its DOM
• The voice generator will release a v3 with different audio characteristics
• The Google AI Overviews format will change and your AEO QA will silently start failing

Every component of the stack ages independently. With a workflow tool, every age-out is a manual fix in your nodes. With a vertically integrated agent system, the maintenance is the operator's problem rather than yours.

The realistic ongoing maintenance burden for the n8n version is roughly 4 to 8 engineering days per month in steady state. Some months are quiet. Some months are an emergency.

5. Per-prospect artefact selection

This is the limit we promised at the top of the post. The day-by-day breakdown above describes one typical mix — the artefacts a service-business prospect tends to receive. But the engine that produces real results does not run that fixed mix for every lead. It picks the right six skills from a wider library for each prospect, based on their industry, their monetization tier, the lead's vertical, the channel they live on, and what proof would actually move them.

A Chennai gated-community RWA buying garden planters might need: tailored landing page → peer-reference proof → free 3-tower pilot offer → voice note from the founder → tower-by-tower yield numbers → polite breakup. A Bangalore HR head running Q4 corporate gifting needs: tailored landing page → boutique-hotel reference → sample resin coaster shipped → voice note from the brand owner → volume-tier pricing → polite breakup with a save-the-number ask. Same cadence. Different six. Different types of artefacts on Day 3 (a brief vs. a sample shipment), Day 5 (yield numbers vs. tier pricing), Day 6 (door-open close vs. quarter-saved close).

This per-prospect decisioning is itself a system: read the customer's offering and ICP, read the lead's vertical and buyer-side signals, score each artefact against the (customer × lead) pair, and orchestrate the six-skill mix that fits. n8n does not have nodes for any of this — you would build a complete decisioning agent on top of all the per-skill pipelines above.

When n8n is genuinely the right answer

I want to be honest about this because it builds credibility for the rest of the argument.

n8n, Make, and Zapier are excellent for:

• Internal ops glue — moving data between SaaS tools your team already uses
• Notification routing — Slack, email, SMS triggers off database events
• Simple ETL — pulling a CSV from one place, transforming it, pushing it to another
• Webhooks-and-actions — reacting to events in one system by triggering actions in another
• Recurring batch jobs — daily exports, weekly digests

If your problem fits one of those shapes, do not hire anyone. Build it in n8n in an afternoon.

The Prewiro cadence does not fit any of those shapes. It is a sequence of generative artefacts across a stateful 6-day window with quality gates and human-escalation paths. That is a different category of system. The right tool for it is an agent platform, not a workflow tool.

What this means for the build vs. buy decision

If you are reading this because you were sizing up the cost of building Prewiro yourself, the honest decision tree is something like this:

• If you have a strong AI engineering team already and outreach is a strategic differentiator for your business, building a version of this is reasonable. Plan for ~3 months and a permanent maintenance team.
• If you are a solopreneur, agency, or small business owner, the build economics never work. Use Prewiro or use traditional cold outreach — building it yourself is the worst of both worlds.
• If you have an existing n8n install, keep it. Prewiro can hand off completed artefacts to your n8n flows for the parts n8n is good at — CRM sync, Slack notifications, downstream automations.

Further reading

• The 6-Day Trust-First Outreach Playbook — what the cadence actually is and why each day exists
• Why we built Prewiro: a beta launch note — the founder note and the beta details

If you have read this far, you are exactly the kind of person who should be in the beta cohort. Apply here.