A plan, not a build. None of these bots exist yet apart from the parts of the prototype marked "built" in the listing (Matcher, Planner, Test referee). Every cost and hour here is an ESTIMATE with its assumptions shown. Tool prices come from the suppliers' own pages, accessed 23 October 2026. Currency: $1 = £0.75 and €1 = £0.87 (ASSUMPTIONS).
1. Summary
After Six runs on 11 bots plus you. They are organised in four crews under a Chief of Staff bot. The bots do the repeating work: writing kits, sending the Monday coach, triaging support, drafting content, reconciling Stripe and watching the site. You keep every decision that spends money, sends something new to customers, publishes, refunds, or touches the law.
- Extra running cost of the bot team: about £57 a month from month 3 (£40 in months 1 and 2). ESTIMATE. This comes on top of the one-page plan's £119.33 fixed costs and £1.05 per subscriber, which already pay for the product's own AI, hosting and email.
- All-in tools and bots: about £365 a month at 180 subscribers and £533 at 340 (the Base path at month 12). ESTIMATE.
- Your time: 10 to 14 hours a week before launch, then the listing's rule of 2 hours a week plus 1 hour per 70 subscribers: about 3, 5 or 8 hours a week in the Cautious, Base and Strong cases.
- This plan extends the one-page plan's seven agents. Matcher, Planner and Test referee stay as rules code inside the product, so they are tools the bots use rather than bots. Kit writer, Monday coach, Claims guard and Owner digest become bots. Seven new bots cover support, growth, finance, analytics, the library and reliability.
2. Org chart
3. The bots
1. Chief of Staff
Ops crew. Maps to: One-pager "Owner digest" (planned).
- Job: Runs the week. Collects every bot's output, keeps the approvals queue, writes the Monday owner digest, routes work and raises the alarm.
- Owns: The approvals queue, the digest, the run log, bot on/off switches (it can switch a bot off, never on).
- Inputs, tools and connectors: n8n (orchestration), Supabase (read), all bot logs, owner email
- Daily: 08:00 run check, chase stuck jobs, queue items for the owner.
- Weekly: Monday digest (sign-ups, reveals, paying, cancellations, flags, spend against cap).
- KPIs: Digest sent by 09:00 Monday; approvals answered within 48 h; zero silent failures.
2. Kit Writer
Product crew. Maps to: One-pager "Kit writer" (planned).
- Job: Writes each paying subscriber's name ideas, tagline, one-page site copy and first offer from the matched idea and their answers.
- Owns: Kit quality and the kit prompt.
- Inputs, tools and connectors: OpenAI gpt-5-mini, Supabase (subscriber answers, minimal fields), idea library
- Daily: write kits for new subscribers within 10 minutes of payment; regenerate on request.
- Weekly: send 5 random kits to the owner for spot-checks.
- KPIs: Kit delivered < 10 min; regeneration rate < 15%; Claims Guard block rate < 2%.
3. Monday Coach
Product crew. Maps to: One-pager "Monday coach" plus the built Planner and Test referee.
- Job: Sends each subscriber's weekly check-in, reads replies, moves the plan after a rubbish week, and points tax, law and employer questions to GOV.UK or Acas.
- Owns: The weekly coach email and reply handling. These are service messages to customers, not marketing.
- Inputs, tools and connectors: Postmark (transactional email), gpt-5-mini, Supabase (plan, ticks), Planner and Test Referee code
- Daily: answer replies (draft-then-send inside approved templates); flag anything about money worries, health, legal or tax.
- Weekly: Monday send to all active subscribers.
- KPIs: Reply within 4 working hours; flagged-reply rate; week-4 retention.
4. Library Keeper
Product crew. Maps to: Maintains the built Matcher.
- Job: Keeps the 40-idea library, the Matcher rules and the tax and rules page true. Checks each idea's GOV.UK links, and re-checks rates every April.
- Owns: The idea library, Matcher weights and the GOV.UK source list.
- Inputs, tools and connectors: GOV.UK Content API (public page JSON), the build generator in build/, GitHub (pull requests only)
- Daily: Weekly: check every GOV.UK link and "last updated" date for the library and tax page; open a pull request when something changed. Quarterly: suggest new ideas from search and support data.
- Weekly: see daily
- KPIs: Zero broken source links; April tax refresh merged before 6 April; library review each quarter.
5. Claims Guard
Trust crew. Maps to: One-pager "Claims guard" (planned).
- Job: Checks every customer-facing line before it is shown or sent: no earnings forecasts (CAP Code 20.4 and 20.7), no regulated work without its GOV.UK rule, no testimonials that do not exist.
- Owns: The block list, the claims rules and the audit trail of every block.
- Inputs, tools and connectors: gpt-5-mini plus a rules list, every other bot's output queue
- Daily: Continuous: screen kits, coach replies, emails, posts and ads.
- Weekly: report blocks and near-misses to the Chief of Staff.
- KPIs: 100% of outbound copy screened; zero earnings claims published; false-block rate < 5%.
6. Support Desk
Trust crew. Maps to: New.
- Job: Triages the support inbox, answers how-to questions from the help docs, drafts refund and complaint replies, and logs data-protection requests.
- Owns: The support inbox and the help docs.
- Inputs, tools and connectors: Google Workspace mailbox (the one-pager's domain and email line), Supabase (read), Stripe (read-only key)
- Daily: triage within 4 working hours; send how-to answers from approved macros; queue refunds, complaints and data requests for the owner.
- Weekly: top 5 questions to the Library Keeper and Analyst.
- KPIs: First reply < 4 working hours; macro accuracy (owner spot-check); data requests logged the same day.
7. Growth Writer
Growth crew. Maps to: New.
- Job: Drafts search articles, social posts and ad copy that sell the test, not the money. Everything goes through Claims Guard and then the owner.
- Owns: The content calendar and the draft queue.
- Inputs, tools and connectors: Buffer (drafts and scheduling), the site generator (articles as pull requests), Plausible (what is read)
- Daily: one social draft per channel.
- Weekly: one article draft and a batch of posts for the owner to approve in one sitting.
- KPIs: Drafts approved first time > 70%; organic visitors; reveal starts from content.
8. Lifecycle Mailer
Growth crew. Maps to: New.
- Job: Runs the marketing emails to people who ticked the consent box: welcome series after a free reveal, trial reminders, annual-plan offer.
- Owns: Consent records and the marketing email sequences.
- Inputs, tools and connectors: MailerLite (marketing email), Supabase (consent flag and timestamp)
- Daily: add newly consented contacts; remove unsubscribes everywhere within 24 h.
- Weekly: report opens, clicks and paying conversions per sequence.
- KPIs: Unsubscribe honoured < 24 h; zero sends without a consent record; reveal-to-paid conversion from email.
9. Analyst
Growth crew. Maps to: New.
- Job: Keeps the funnel numbers: visitors, reveals, paying, churn, CAC by channel, and the test-ad results. Proposes one experiment a fortnight.
- Owns: The KPI sheet and experiment log.
- Inputs, tools and connectors: Plausible, Stripe (read-only), Supabase (read), the 12-month model
- Daily: refresh the KPI sheet.
- Weekly: funnel by channel and scenario tracker (are we on the Cautious, Base or Strong path?).
- KPIs: KPI sheet fresh every morning; each experiment with a written result.
10. Finance Clerk
Ops crew. Maps to: New.
- Job: Reconciles Stripe payouts, tracks every tool bill against budget, chases failed payments with Stripe's own retry emails, and watches the VAT threshold.
- Owns: The monthly P&L, the budget caps and the VAT watch.
- Inputs, tools and connectors: Stripe (read-only restricted key, no refund permission), bank feed export (CSV), supplier invoices by email
- Daily: payouts and failed payments.
- Weekly: spend against caps. Monthly: P&L against this plan; rolling 12-month turnover against the £90,000 VAT threshold.
- KPIs: Books closed by day 5; no tool bill unexplained; VAT alert at £75,000 rolling.
11. Site Watch
Ops crew. Maps to: New.
- Job: Keeps the product up: uptime checks, broken links, page weight, error logs, failed jobs, dependency updates (as pull requests).
- Owns: Uptime, error budget, the deploy checklist.
- Inputs, tools and connectors: n8n HTTP checks, Supabase logs, GitHub (pull requests only), Playwright checks from the audit folder
- Daily: Every 5 minutes: uptime. error log summary, signup and checkout smoke test.
- Weekly: link and accessibility check of key pages.
- KPIs: Uptime ≥ 99.5%; checkout smoke test green daily; incidents written up.
4. What bots may do alone, and what needs your yes
| Action | Bot may do alone | Needs your yes |
|---|---|---|
| Spending money | Nothing. Bots run inside tools already paid for. | Any new tool, plan upgrade, ad spend or top-up. Ad budgets are set by you as a monthly cap; the Analyst may move money between ads inside the cap only if you have written that rule. |
| Sending to customers | Service messages from approved templates: kit delivery, receipts, Monday coach check-ins, how-to replies from approved macros. | Any new template, any marketing email campaign, any reply about money, health, law, tax, complaints or a refund. |
| Publishing | Nothing public. Drafts and pull requests only. | Every article, social post, ad and site change. From month 4 you may give standing approval to a named set of post templates. |
| Refunds and credits | Draft the refund and the reply, with the Stripe charge ID. | Every refund, credit or price exception (the Stripe key the bots hold cannot refund). |
| Legal and data | Log a data request; draft the reply; flag a suspected breach at once. | Any reply to a data request, any breach decision (the ICO must be told of a reportable breach within 72 hours), any contract, terms or privacy-notice change. |
| Product and prices | Propose; open pull requests; run A/B tests you approved. | Price changes, plan changes, new ideas going live, April tax-rate updates. |
| Bots themselves | Chief of Staff may pause any bot. | Switching a bot back on after an incident, changing a prompt, adding a connector or a permission. |
5. Handoffs between bots
| From | To | What passes | Trigger |
|---|---|---|---|
| Stripe (payment) | Kit Writer | New subscriber ID and answers | checkout.session.completed webhook |
| Kit Writer | Claims Guard | Draft kit | every kit |
| Claims Guard | Kit Writer / Owner | Pass, or block with the rule broken | every check; 2 blocks in a row go to the owner |
| Monday Coach | Support Desk | Replies about billing, refunds or cancellations | keyword or intent match |
| Support Desk | Owner (via Chief of Staff) | Refund drafts, complaints, data requests | same day |
| Support Desk | Library Keeper and Analyst | Top questions, confusing pages | weekly |
| Library Keeper | Owner | Pull request with the GOV.UK change and the link | when a source page changes; every April |
| Growth Writer | Claims Guard → Owner | Draft posts, articles, ads | daily drafts, weekly batch |
| Analyst | Growth Writer and Lifecycle Mailer | What worked, next experiment | fortnightly |
| Finance Clerk | Chief of Staff | Spend against caps, VAT watch, failed payments | weekly; immediately if a cap hits 80% |
| Site Watch | Chief of Staff → Owner | Incident | checkout or sign-in down for 10 minutes |
| Every bot | Chief of Staff | Run log and errors | every run |
6. Failure and escalation rules
| Level | Examples | What happens |
|---|---|---|
| Red: tell the owner now | Checkout or sign-in down > 10 min; suspected data breach; an earnings claim got out; a customer in distress; Stripe dispute | Chief of Staff emails and pushes a notification to you, pauses the bot involved, and writes what it knows. Nothing is retried that would send to customers. |
| Amber: today | Claims Guard blocks the same bot twice; a refund or complaint; a tool bill above budget; a GOV.UK rule changed | Goes in the approvals queue with a recommended answer. If unanswered in 48 h, it is repeated at the top of the next digest; customer replies get a holding reply from an approved macro. |
| Green: in the digest | A bot retried once and succeeded; drafts waiting; experiments finished | Monday digest only. |
- One retry, then stop. A failed job is retried once. If it fails again the bot pauses that job and the Chief of Staff tells you. Sends to customers are never retried automatically, so nobody gets a message twice.
- Kill switch per bot. Every bot is a separate n8n workflow with its own on/off switch. The Chief of Staff may switch a bot off but never on.
- Least privilege. Stripe keys are restricted and read-only. GitHub access is limited to pull requests. The marketing tool cannot see payment data. No bot holds an admin password.
- Spend caps in the tools. Hard monthly caps are set on the OpenAI project and in the ad accounts. A bot never raises a cap.
- Human fallback for every customer path. If the Kit Writer is down, the customer gets the written template kit the prototype already uses, plus an apology. If the Monday Coach is down, the written rule replies (as in the prototype) go out.
- Logs kept for 90 days. Every bot logs its input reference, output, model and cost. Customer content is referenced by ID, not copied into logs.
7. Your weekly hours and decisions
| Phase | Your hours a week (ESTIMATE) | What you do |
|---|---|---|
| Pre-launch (2 weeks) | 10 to 14 | The week-one playbook: accounts, Stripe, keys, prompts; read every test kit; set caps; write the approved templates. |
| Launch (month 1) | 6 | Read every kit for the first 20 subscribers; read every coach reply daily; approve the first content batch. |
| Months 2 to 3 | 3 to 5 | Daily 10-minute flag check; weekly one-hour approvals sitting; spot-check 5 kits. |
| Months 4 to 12 | 3 (Cautious), 5 (Base), 8 (Strong) | The listing's rule: 2 hours plus 1 hour per 70 subscribers. You read what the bots flag, approve batches, and make the monthly decisions. |
| When | Decision |
|---|---|
| Daily (10 min) | Red and amber flags; refunds; replies the Support Desk cannot send. |
| Weekly (about 1 h) | Approve the content batch and any campaign; read the digest; confirm ad cap for the week; 5-kit spot-check. |
| Monthly | Keep, move or cut channel budgets; which experiment next; any tool upgrade; P&L check against this plan's scenarios. |
| Quarterly | Idea library review; price review; whether to give the bots more standing approvals. |
| Every April | Sign off the tax-rate refresh before 6 April. |
8. Phased rollout
| Phase | Bots switched on | Mode | Gate to next phase |
|---|---|---|---|
| Pre-launch (weeks −2 to 0) | Chief of Staff, Site Watch, Finance Clerk (Stripe test mode), Claims Guard, Kit Writer (shadow) | Shadow: bots write, nothing reaches customers. | 20 test kits read by you with zero earnings claims; checkout smoke test green 7 days. |
| Launch (month 1) | + Kit Writer live, Monday Coach live, Support Desk (drafts only) | Every kit and every coach reply read by you for the first 20 subscribers. | Claims Guard pass rate ≥ 98%; no red incidents for 14 days. |
| Months 1 to 3 | + Growth Writer, Lifecycle Mailer (from consent box only), Analyst, Library Keeper | Drafts and batch approvals; spot-checks fall to 5 kits a week; Support Desk sends approved macros. | KPIs tracked for 8 weeks; the month-3 keep, re-price or pivot decision made (one-pager). |
| Months 4 to 12 (scaling) | All 11 | Standing approval for named post templates if you choose; fortnightly experiments; the VAT watch matters in the Strong case. | Quarterly review of permissions, costs and incidents. |
9. Monthly running cost (ESTIMATE)
The one-page plan already budgets for the product's own running costs. This plan adds only what the bot team itself needs.
| Already in the one-page plan | Monthly |
|---|---|
| Database, hosting and auth | £22.00 |
| Domain and email | £8.00 |
| AI for free reveals (about 1,500 a month at about 4p) | £60.00 |
| Transactional email | £15.00 |
| Small software | £10.00 |
| ICO fee spread monthly (£52 a year) | £4.33 |
| Fixed costs in the one-page plan | £119.33 |
| Per subscriber: AI for coach and kit, site hosting, coach emails | £1.05 each |
Extra for the bot team:
| Line | Price page says | Monthly, ESTIMATE | Source (accessed 23 October 2026) | Note |
|---|---|---|---|---|
| n8n Cloud Starter (orchestration for all bots) | €20 a month billed annually | £17.40 | n8n pricing | Self-hosting the Community Edition is free if you would rather run it. |
| gpt-5-mini tokens for the ops bots (digest, triage, drafts, guard) | $0.25 / $2.00 per million tokens in / out | £8.00 | OpenAI pricing | ASSUMPTION: about 3m in and 0.6m out a month is roughly $2; we budget about 5× that. |
| Plausible analytics (cookie-free) | Starter $9 a month at 10k pageviews | £15.00 | Plausible pricing | Allowance for the next traffic tier, not quoted. |
| MailerLite (marketing email, from month 3) | Free to 250 subscribers; Comfort from $12 a month | £9.00 | MailerLite pricing | Free plan has no API sending, so the bot needs Comfort once the list grows. |
| Buffer (social scheduling, from month 3) | Essentials $5 a month per channel | £7.50 | Buffer pricing | Two channels. |
| Bot team extras, from month 3 | £56.90 | £40.40 in months 1 and 2. |
All-in at different sizes (one-page fixed costs + bot extras + £1.05 per subscriber; card fees excluded):
| Paying subscribers | All-in tools and bots a month (ESTIMATE) |
|---|---|
| 60 | £239 |
| 180 | £365 |
| 340 | £533 |
| 400 | £596 |
| 690 | £901 |
What we left out on purpose: a separate CRM (the Supabase customer table plus MailerLite does the job at this size), a paid helpdesk (the existing mailbox is enough below about 500 subscribers, ASSUMPTION), and a paid uptime service (n8n HTTP checks are enough to start).
10. Risks and guardrails
| Area | The rule | Guardrail in the bot team |
|---|---|---|
| UK GDPR | Lawful basis, privacy notice, data minimisation, processor contracts, transfers outside the UK; data subject requests within one month; reportable breaches to the ICO within 72 hours (ICO: report a breach, accessed 23 October 2026). | Kit Writer gets only the answers it needs, not names or emails. A processor agreement (DPA) with OpenAI, Supabase, Postmark, MailerLite and n8n is checked before launch. Support Desk logs every request the same day. Logs keep IDs, not content. |
| PECR (marketing email) | Marketing email to individuals needs consent or the soft opt-in, and sole traders count as individuals. Every message needs a simple way to opt out. (ICO: electronic mail marketing, accessed 23 October 2026. The ICO says this guidance is under review after the Data (Use and Access) Act.) | Lifecycle Mailer only emails contacts with a consent flag and timestamp from an unticked box on the reveal. We do not rely on the soft opt-in for free reveals: it is unclear whether a free reveal counts as "negotiations for a sale". Unsubscribes are honoured everywhere within 24 hours. Monday coach emails are service messages to paying subscribers. They carry no promotions. |
| CAP Code | Work-from-home and business-opportunity ads must not forecast earnings, rules 20.4 and 20.7 (CAP Code section 20, accessed 22 October 2026). | Claims Guard screens every kit, reply, email, post and ad. Banned: income figures, "replace your salary", "make £X", invented testimonials or results. The product sells the 14-day test, not the money. |
| Subscription rules | From 1 January 2027 new UK subscription contracts must give clear up-front information and reminders, an easy online exit, and a 14-day cooling-off period after a trial or a 12-month-plus renewal (GOV.UK: Subscription contracts guidance, published 19 October 2026, accessed 23 October 2026). | Finance Clerk sends a renewal reminder before every annual renewal. Support Desk handles cooling-off refunds, and you approve them. Cancelling stays one click in the account page. |
| Regulated ideas | Food, childcare, pet care and driving ideas carry rules (one-page plan). | Library Keeper keeps each idea's GOV.UK link; Claims Guard blocks a kit for a regulated idea that lacks it. |
| Tax and law answers | The coach must not give tax or legal advice. | Monday Coach points to GOV.UK and Acas and flags the question to you. |
| AI quality and cost | A dull or wrong kit spoils the reveal; a runaway loop costs money. | Rules-based matcher, AI for wording only; hard OpenAI spend cap; written template fallback. |
| Bot sprawl | Eleven bots can hide failures. | One Chief of Staff, one run log, one digest; a quarterly review of permissions and costs. |