Capstone: a full-funnel AI-driven campaign plan with test design

0:00 Capstone

This is where everything comes together. Your capstone is a complete, decision-ready campaign plan that a founder, a CMO or a client could approve today and a team could execute next week. In this lecture, I'll walk you through the eight-part template, show an abbreviated example for a B2B software company, and give you a rubric to check your work before you submit.

0:27 Why it matters

Why does this capstone matter? Because a plan is where knowledge turns into money. Anyone can list tactics. A plan that ties economics, signals, creative, guardrails and tests together is rare, and it's what clients and employers actually pay for. Here's an analogy. An architect's drawing isn't valuable because it's pretty. It's valuable because a builder can follow it and the building stands up. Your plan should be the same: specific enough that a team can execute it next week, and rigorous enough that you'll know within ninety days whether it's working.

1:07 Choose a scenario

Choose a scenario, or use your own business. Option A, a beauty brand selling in the UAE and Saudi Arabia, launching in the UK next quarter. Option B, a software company in Pakistan selling to small businesses in the Gulf, with a sixty-day sales cycle. Option C, a group of clinics in the UK that take online and phone bookings. Each one stresses different skills: cross-border creative, long-cycle signals, or offline conversions.

1:38 Sections 1–2: foundation

Sections one and two are the foundation. First, the business goal and the economics: your objective, contribution margin, break-even ROAS or acquisition cost, and payback target. Second, measurement: which conversion events you'll send, how their values are calculated, your server-side tracking plan with deduplication and consent, and your backend scoreboard. If these two sections are weak, nothing else in the plan can be trusted.

2:06 Sections 3–5: the engine

Sections three to five are the engine. Three, channel and campaign architecture: for each platform, which automated campaign types, how many and why, which settings are hard controls and which are soft signals, plus a budget split based on marginal returns. Four, the creative system: your angle matrix, the first ten to twelve concepts, your AI production workflow with disclosure rules, and a refresh cadence. Five, guardrails: negatives, exclusions, suitability tiers, regulated categories and alerts.

2:39 Simple example: one test spec (illustrative)

Let's do a simple worked example of one test from a plan, so you can see the level of detail expected. Scenario A, the beauty brand. Hypothesis: TikTok Smart plus prospecting increases total new customers in the UAE. Treatment: TikTok on in Dubai and Sharjah. Control: TikTok off in Abu Dhabi and the northern emirates, matched on past sales. Metric: new customers from Shopify by emirate. Duration: five weeks plus one week cool-down. Minimum detectable effect: checked with a power analysis. Decision rule: keep TikTok if incremental cost per new customer is below target. If inconclusive, extend two weeks, then run a platform lift study.

3:25 Section 6: test design

Section six is the heart of the capstone: test design. You need at least one creative A/B test and one incrementality test, fully specified. Hypothesis. Treatment and control. The unit of randomization: users, regions or time. The primary metric and where it comes from. The minimum detectable effect and a power check. Duration and cool-down. A decision rule. And what you'll do if the result is inconclusive, because sometimes it will be.

3:56 Sections 7–8

Sections seven and eight close the loop. Seven, reporting and governance: your weekly memo structure, drafted by AI and reviewed by a person, your anomaly rules, and who approves what. Eight, a ninety-day roadmap. Weeks one and two, tracking and consent. Weeks three to six, launch and creative exploration. Weeks six to ten, the first incrementality test. Weeks ten to thirteen, reallocation and getting data ready for a mix model.

4:26 Example: Scenario B (illustrative)

Here's the abbreviated example for option B. The goal: qualified pipeline at under a quarter of first-year contract value, illustratively. Stage values from historical close rates, so a marketing-qualified lead, a sales-qualified lead and an opportunity each carry a rising value. LinkedIn Accelerate with predictive audiences from closed-won accounts, Google Search with AI Max and brand excluded, and a Meta leads test. All three receive CRM stages through conversion APIs. The incrementality test: a six-week LinkedIn holdout in two Gulf countries against three matched countries, measured by CRM opportunities.

5:05 AI as reviewer

Use AI as your toughest reviewer, not your author. The lesson includes a prompt that asks a model to rank the five biggest risks in your plan, check each test for a single variable, a backend metric, power and a decision rule, flag policy or legal issues in your markets, and suggest the single change that would most improve the outcome. Then you decide what to change. The thinking has to be yours.

5:37 Self-check rubric

Before you submit, score yourself with the rubric. Economics derived from margins, not guessed. Signals server-side, deduplicated and consented, with values that reflect profit or pipeline. A consolidated architecture with hard and soft controls explicit. At least ten distinct creative concepts with a compliant AI workflow. Fully specified, powered tests with decision rules. And governance with a ninety-day roadmap.

6:03 Mistakes + try this now

Common capstone mistakes. Plans full of platform features but no economics. Test sections that say we'll A/B test creative, with no hypothesis, metric or decision rule. Measurement sections that rely only on platform ROAS. Creative plans with ten variations of one idea. And no mention of consent, disclosure or regulated categories. Try this now: before writing anything else, write the one sentence that defines success for your plan, including a number and a source, like new customers from Shopify at a blended cost below a target. Every other section should serve that sentence.

6:43 Watch me do it: plan economics (illustrative)

Watch me do it. I'll draft the economics section of scenario B in real time, because it anchors everything else. Step one, average first-year contract value, from the CRM: say six thousand dollars, illustratively. Step two, gross margin on software: roughly eighty percent, so first-year contribution about four thousand eight hundred. Step three, the acceptable acquisition cost: leadership accepts payback within the first year, so up to about four thousand eight hundred per new customer, and we target lower, around three thousand, for safety. Step four, the funnel: historically ten percent of SQLs become customers, and thirty percent of demo requests become SQLs. So the target cost per SQL is about three hundred, and per demo about ninety. Step five, stage values for the platforms: demo worth about ninety, SQL about three hundred, opportunity about one thousand two hundred based on its close rate. Now every later section has numbers to serve: bids, budgets, tests and the success sentence. Ten minutes of arithmetic turned a vague plan into one a finance director can approve.

7:59 Your next step

That's the course. You now know how platform AI works, how to feed it signals and value, how to make creative your targeting, how to allocate budget on marginal returns, and how to prove what's working with experiments and mix models. Your next step is the capstone. Pick a scenario, complete the template, run the AI critique, revise, and you'll have a plan you'd be proud to put in front of any client.