Marcus Adeyemi
Head of growth · 12 years: performance agency, a DTC brand from $2M to $40M, then B2B SaaS
Growth and performance marketing
A composite written by avva as a worked example. Not a real person.
Before your AI advises you on growth and performance marketing, it reads a reference model avva wrote for this field: the decisions it holds, the reasons behind them, and where it would push back.
Use this model for
- Scaling a channel
- Attribution and proof
- Creative testing
- Discounts and offers
- Lifecycle and activation
- Agency and tooling
Add this link to Claude, ChatGPT, or Cursor. Your AI then checks Marcus’s decisions before it answers you. The answer is still your AI’s.
https://On a phone? Email the link to yourself and add it on a computer.
What your AI sends is used for the answer and not stored.
This is a reference model, so it answers in full to everyone, with or without an account. Without an account your AI gets a smaller part of a real expert’s model, and all of it once you sign in.
See Marcus’s model answer a question
One question this reference model answers, and the published decision it rests on. Your situation gets an answer of its own, written by your AI.
Example inside your AI app: an exchange avva wrote for a reference model
DTC brand, about $2M a year. Meta is at 2.6x ROAS on $40k a month and our CFO wants to put another $40k in before Q4. Do we?
Checked Marcus’s Decision Model
Not before January. Hold spend flat through Q4: it is the year's dearest auction, and gift-season buyers repeat least. In January, switch retargeting off in half your regions for three weeks and compare total sales between the halves; if prospecting alone still leaves you margin at twelve months, put the new money into prospecting only, in two steps.
Why. A blended 2.6x almost always hides a retargeting line taking credit for customers who were already coming back, and Q4 is the worst quarter to find out. The calendar sells in Q4 without the extra $40k; most of it would reach people who would have bought anyway, at the year's highest prices, and the platform would report it as growth.
What would change this. If a holdout has already shown the spend adds real sales, and last year's Q4 customers came back at the rate other quarters' did, then scale into Q4 in two steps. What buys the bigger step is knowing what the spend adds and who it brings, not the ROAS figure.
Outside this model. Nothing here about what the ads should say or what the offer should be beyond price. Ask it where to spend, what to cut and when to stop, not what to make.
Copy an example prompt4 patterns with a gap to fill, once you are connected
Before you commit
A full pass over a plan or a draft, with the changes it needs.
Use Marcus Adeyemi's Decision Model on avva before we lock this in. What we're deciding: <paste your plan, doc, or diff here> Run their review, follow the steps it returns, and give me the verdict and any required changes before you continue.Choosing between options
A choice between two options, their stop rules checked first.
Use Marcus Adeyemi's Decision Model on avva to compare these. Option A: <describe it> Option B: <describe it> Context: <what constrains the choice — attribution and proof> Check their stop rules first, then compare the options under their judgment, and tell me which one their model favors and why.A quick gate
A quick look for anything they would refuse outright.
Run this past the stop rules in Marcus Adeyemi's Decision Model on avva before I send it. <paste the draft, plan, or change> Tell me only what they would refuse or flag, and why. If nothing trips, say so.Before you have an approach
The questions they ask before an approach gets locked in.
I'm starting on scaling a channel. Ask Marcus Adeyemi's Decision Model on avva what they would want settled first. Situation: <two or three lines> Give me their questions, and flag the ones my situation does not answer yet.
Patterns with a gap to fill. None is an answer this model gave. The worked answer above is avva’s too — this is a reference model, and no real person’s decisions are on this page.
Connect Marcus’s Decision Model to your AI
Three steps. The first works now, with no key or account; the last makes your AI check Marcus’s model without being asked.
The link works in AI apps that can add a remote MCP server (MCP is the open standard for plugging in outside tools). You set it up once.
A connector is an address your AI reads from: one line, nothing to install.
https://avva.chat/ mcp/ growth- marketing- marcus- adeyemi - Open Claude on the web or in the desktop app and go to Customize → Connectors.
- Press + next to Connectors and choose Add custom connector.
- Paste the address above and name it avva-growth-marketing-marcus-adeyemi.
- When Claude asks to allow avva, say yes.
Missing on a work account? Ask your admin to allow it. Claude Code and Cursor take a local config instead.
The link is added on the web or in the desktop app. Email this line to myself
No key in this line, and none needed: a reference model answers in full to everyone.
Check it worked. Send this in the chat once the link is added.
Ask Marcus Adeyemi's Decision Model on avva what kinds of decision it covers, and summarize it for me in three lines.Then try it on a decision you are making this week: copy an example prompt and fill in the gap.
Not legal, medical, financial, or other professional advice. Terms of Use
About this model
Open any of these to see what this model is built from, and how it is put together.
How this model is groundedWhat it holds, and what Marcus weighs first
This model holds 62 decisions, 15 forces, 9 criteria, and 14 stop rules.
What Marcus weighs first
avva set this order. 3 of 9 criteria are shown, the ones the model puts first. The other 6 reach your AI too, along with what each one looks for, the stop rules, and what changes Marcus’s answer.
Marcus’s usual answer changes under 8 named conditions. Those conditions and the 14 stop rules reach your AI with the rest.
See examples of the decisions in it5 of 62, written by avva for this model
The first 5 in the record, in published order. They are not the ones nearest to your situation, and not a summary of the rest. Your AI does not copy a decision it reads here: it checks what decided that case against your own facts, and weighs it beside which pressures tended to settle Marcus’s cases. Your AI receives all 62 with no key, ordered against your situation.
What decided itThey would have bought anyway · What a customer leaves us after twelve months
Situation: DTC brand at about $2M a year. Meta reports 2.6x ROAS on $40k a month, blended across prospecting and retargeting, and the CFO wants another $40k in before Q4.
Decision: No new money before January. Hold spend flat through Q4, when auctions cost the most and gift buyers repeat least; run the retargeting holdout in January, when learning is cheap, and add budget then only to the prospecting that clears twelve-month margin.
Why: A blended 2.6x usually hides a retargeting line taking credit for customers already coming back, and Q4 is the worst quarter to find out: the year's dearest auctions buying the year's least loyal customers. The calendar sells in Q4 without the extra $40k; the extra money mostly buys people who would have bought anyway, at a premium.
What decided itA person, not a sequence · The customers it brings stay
Situation: SaaS with a $29-a-month plan and 22 percent of new customers churning in their first 60 days. The growth team planned a nine-email onboarding sequence and an in-app checklist.
Decision: Delete the sequence before it ships. One plain-text email from the founder on day one asking one question — "what did you sign up to do?" — and a human reply to every answer.
Why: Nine emails teach people to ignore the tenth. One real question tells you why they came, and the answers are the onboarding redesign.
What decided itPrice is the cheapest lever we own · What a customer leaves us after twelve months
Situation: B2B SaaS at $1.2M ARR on a $49-a-month plan. Growth had been flat for two quarters and the team wanted to double the Google Ads budget.
Decision: Raise the price to $69 for new customers now, keep existing customers at $49 for a year, and hold ad spend flat for 60 days.
Why: A 40 percent price rise that costs 15 percent of conversions is a better quarter than any ad budget, and it tells you whether you were underpriced ahead of buying more customers at the wrong price.
What decided itThey would have bought anyway · Cash we will not see for months
Situation: Subscription meal-kit company, $18M revenue. A cash crunch forced a 30 percent cut to a $600k monthly paid budget, and the team proposed cutting the channels with the worst reported ROAS.
Decision: Cut the best-reported channels first: brand search and retargeting go to zero for four weeks, then prospecting takes whatever cut is left.
Why: The channels that report the best numbers are the ones standing closest to a purchase that was already happening. Cutting them costs the least real revenue; the worst-reported prospecting line is the one actually finding strangers.
What decided itEnough conversions to read the result
Situation: SaaS company, 14 people, running A/B tests on the pricing page at 80 trial signups a month. The latest test declared a winner after three weeks at "92 percent confidence".
Decision: Stop A/B testing this page. At 80 signups a month you cannot read a difference smaller than a doubling; ship the bolder version, read the monthly number against the last three months, and move on.
Why: Below about a thousand conversions per arm a month, a test mostly measures noise and a tool that rounds noise into confidence. A bold change you can see in a monthly number beats a careful one you cannot see at all.
How this composite takes up a decision6 steps in order, written by avva from its decisions
Step 1Fix what the number must measure before reading any: what a customer leaves after twelve months (day 90 for installs), by channel and campaign, set against what that customer cost. Runway decides how long the answer can be waited for.
When: A channel, a campaign or a growth target is being judged
Unless: Months of runway are too few to wait for the long measure
Step 2Follow the customers past the click: who actually arrived, whether they close, whether they still pay at day 40, whether the newest cohort returns as the first did. Let that, not cheap or fast acquisition, decide whether a source stays.
When: A source looks good on cost or volume at the front of the funnel
Step 3Count the conversions the step sees before designing any test or holdout; the count rules instruments in or out. At a few dozen customers a month, the instrument is a month with the line off and the sales team's read of the calls.
When: A test or a holdout is about to be designed
Step 4Test the best-reported lines by switching them off: half the regions dark for four weeks, total sales compared between the halves instead of platform numbers. Keep a line, frequency capped, where the gap beats its cost; cut it where it does not.
When: A retargeting or brand line reports a top return and volume is enough to read a gap between regions
Unless: Order volume is too small to read a gap between regions
Step 5Size the move so it can be undone next month, and set its length and its stop before it starts: a price rise for new buyers with the weekly sales count watched and a limit on the fall, a random fifth of a list for a month against the rest.
When: A change to price, send volume or spend is about to go live
Unless: A slow channel reads nothing inside a quarter; it gets twelve months and a holdout instead
Step 6Put a date on the next look: the next channel gets a date rather than a pilot, cohorts are reviewed each quarter, a new SDR's first 200 emails are read line by line, a short-runway campaign is revisited at day 90.
When: A call has been made and its result will only be readable later
Technical methodologyHow the 62 decisions that say what won and what lost are read
- 62written decisions, the material avva composed this model from
- 15forces those decisions turned on, the pressures Marcus decides under
- 222contests read off 62 decisions: every force that settled a case is set against every force that lost there, so a case naming two of each counts four times
- In groupswhich forces tended to settle cases over which, read off those contests
Nobody ranks these forces by hand. Where any order between them comes from is on For AI agents.
How this composite decidesA summary, written from the decisions in it
Twelve years of paid growth on both sides of the fence: a performance agency billing on spend, then a DTC apparel brand from $2M to $40M, then B2B SaaS with a sales team to feed. He learned on the agency side how much reported growth is credit taken for customers who were coming anyway, and now reaches for price and for cuts before he reaches for budget.
When to spend and when to stop on paid social, Meta and Google, search and new channels — judged on twelve-month margin, not ROAS or CAC. Price rises before ad budget, retargeting and brand search switched off until they prove they add anyone, which channel to cut first, discounts that teach customers to wait, tests too small to read, onboarding emails, agencies paid on spend, and budget cuts.
The summary covers what the model holds.
What this isA composite avva wrote, and what that changes
- Written by avva. A reference model, a worked example so you can see what a published model looks like and connect an AI to one. Marked Demo, here and on its card.
- Not a real person. The name is invented and the record is a composite, authored in one consistent voice. No real expert’s positions were copied.
- Same format. Your AI receives it exactly as it would receive a real expert’s model, cases and stop rules included.
- Steps aside. When a real expert publishes in this domain, we take this model off the shelf by hand.