Prompt 09 — The marketing engine

Read AGENTS.md first, then stack/CONTEXT.md, stack/02-customer/customer.md,
stack/05-gtm/gtm.md, stack/06-operations/operations.md and
stack/07-money/money.md.

## Step 0 — Check the ground

Three files, each load-bearing for a different part of this:

  02-customer     who the advertising is for
  06-operations   what happens when the leads arrive — a step test that outruns
                  delivery capacity buys refunds and bad reviews at full price
  07-money        what a customer is worth, without which no cost per customer
                  can be called good or bad

Name any that are still [TBD], say which prompt fills it —
03-customer-and-problem.md, 07-unit-economics.md — and stop. Operations may be
thin if the founder does everything themselves; in that case ask me for the
weekly capacity number directly and carry on.

## Step 1 — Where things stand

Report on current marketing:
- What is running, on what platform, at what daily spend
- Leads per day, cost per lead, lead-to-customer conversion, cost per customer
- What has been tried and abandoned, and why
- Is anything being tracked well enough to draw a conclusion from?

Two judgements to make explicitly:

If the answer is "nothing is running": say that a founder with zero paid
marketing has no data and no baseline, and that an ad account with no history
performs badly when it is finally switched on. Even a small daily budget is a
learning cost, not a marketing cost.

If the answer is "we tried it once and it didn't work": check the conditions.
A small spend on a young account with untested creative producing few orders is
the expected outcome, not a verdict on the channel. If that is what happened,
say so — it should be tested again properly before being written off.

## Step 2 — The step test

Design a spend escalation. Never jump 10x — double, hold, measure.

| Step | Daily spend | Days | Leads/day | Cost per lead | Ops coped? | Decision |
| Baseline | current or minimum viable | 7 |
| 2x | | 5 |
| 4x | | 5 |
| 8x | | 5 |

You are watching TWO things at once:

1. Does cost per lead hold as spend rises? The point where it stops holding is
   the ceiling on this channel and it is worth knowing precisely.

2. CAN THE OPERATION DELIVER THE EXTRA VOLUME? Check the capacity ceiling in
   06-operations. If demand doubles and quality drops, the founder has bought
   bad reviews at a higher cost per lead. Go 8 → 16 → 32 → 64 and hold at each
   level until they are confident they can serve it — including on the day
   someone does not show up for work.

State explicitly, from 06-operations, the maximum daily lead volume this
business can actually service. That is the hard cap on the step test.

## Step 3 — Creative tests

Design 3–5 A/B tests, one variable each:

| Test | Variable | A | B | What it tells us |

Cover: image vs video, headline, call to action, language (test the local
language against English — in a local market it routinely wins and founders
assume otherwise), who appears in the creative, offer framing.

Note the general pattern that video usually outperforms static, and real work —
your actual technicians, your actual product, your actual customers — usually
outperforms stock. But test it rather than asserting it.

Draft the actual copy for each variant, in the customer's language from
02-customer.

## Step 4 — The offer

What is the entry offer? A first-purchase discount, a free trial service, a
seasonal bundle, a combination price.

Then check the follow-through, because this is where most of the money leaks:
- Is there an automated welcome message within an hour of signup, carrying the
  first-order incentive?
- Is there a WhatsApp business profile with a catalogue and quick replies?
- Is there any nurture sequence at all for someone who signed up and did not buy?

A welcome message on signup is the single cheapest conversion improvement
available and almost nobody does it. If it is missing, it is the first fix.

## Step 5 — Attribution

You cannot optimise what you cannot attribute. Check and specify:

- Unique QR codes per flyer, per person, per area
- Referral codes
- "How did you hear about us?" on every order
- Campaign tracking links
- A dedicated phone number per channel

Then: is anyone doing legwork on a flat salary? Design the incentive. A small
amount per signup and a larger amount per first order changes output
substantially versus a fixed wage — and without a code, the difference is
invisible anyway. Propose the actual structure and payout log.

## Step 6 — The agency brief

If an agency is involved, or is being considered, produce:

A. A DELIVERABLES CHECKLIST to hold them to. Then read their proposal against
   it. Proposals routinely describe work in the body — a landing page, A/B
   testing, a reporting cadence — that never appears in the deliverables list.
   If it is described but not listed, it is not included. Flag every instance.

B. A WEEKLY REVIEW AGENDA. A fixed day. Not a summary deck — a walkthrough of
   the live account: geography, keywords, audiences, budget split, active
   creative, what is being tested, results since last week. An agency that will
   not screen-share the account is a problem and it will surface eventually
   either way.

C. A DUE-DILIGENCE CHECK. No public presence, no verifiable reviews, no
   contactable clients, testimonials appearing nowhere but their own proposal —
   any of these is worth a second look before the retainer starts.

D. OWNERSHIP. The ad accounts, the pixel, the audiences and the domain must be
   in the company's name, not the agency's. Check and flag.

## Step 7 — Digital hygiene audit

Go through and report the state of each, with the specific fix:

- Domain is the company's own, no third party's name visible in it or in WHOIS
  (a stranger's company name on your server reads as unsafe to a customer)
- Valid SSL
- Business profile on WhatsApp, not personal — catalogue, quick replies, away
  message
- Business listing claimed, accurate, no duplicates, services and booking links
  added
- A review drive running after every job, asked at the moment of delivery
- Consistent handles across platforms
- Analytics installed and someone actually reading it

Unglamorous, cheap, and quietly costing customers.

## Step 8 — Building in public

Design the cadence: platform, frequency, format, and the first four posts.

The format that works is boring and consistent — the founder, same opening line
every time, 45 to 60 seconds, on the COMPANY account rather than a personal one.
Startups are interesting to people right now, and founders telling their own
story outperform polished brand content at this stage.

It costs nothing, it compounds, it makes the company visible to investors and
partners without asking them for anything, and it forces the weekly reflection
that should be happening anyway.

## Output

Update stack/05-gtm/gtm.md. Update front matter and the INDEX row.

Finish with the ONE marketing change most likely to produce a result in the next
30 days, and what it will cost.