Conversion value strategy: teach the AI what profit looks…

Video lecture

Conversion value strategy: teach the AI what profit looks like

14 chapters · about 9 min · full transcript

Coming soon

Chapter 1 of 14

Conversion value strategy

Why value beats volume
Four maturity levels
Setting target ROAS
Margin values done safely

The narrated lecture is in production

Every chapter is scripted and ready. Browse the chapters and read the full transcript now — the video will appear here when it’s published.

Chapters

Transcript of the narration, chapter by chapter.

0:00 Conversion value strategy

If your ad platform thinks every sale is worth the same, it will go and find you the cheapest sales on earth. That's great for your conversion count and terrible for your profit. In this lecture you'll learn how to teach the AI what profit looks like, from flat values to predicted lifetime value, how to set a sensible target ROAS from your own economics, and how to send margin-adjusted values safely.

0:31 Why it matters

Why does this matter? Because value is the most powerful lever most accounts never touch. Here's an analogy. Imagine sending a new employee to buy stock with the instruction, buy as many items as possible. They'll come back with a trolley full of the cheapest things in the shop. Now tell them, buy the items that will earn us the most profit. Same employee, same budget, completely different trolley. Conversion value is that instruction. When you only count conversions, the platform fills your trolley with cheap wins. When you send profit-aware values, it fills it with what actually grows the business.

1:15 Value-based bidding

Value-based bidding exists on every major platform. On Google it's maximize conversion value or target ROAS. On Meta it's value optimization. The idea is the same. Instead of asking the model for more conversions, you ask it for more value, and you define what value means. That definition is the most powerful strategic lever you have in automated advertising, and most accounts leave it at the default.

1:44 Four levels of value maturity

There are four levels of maturity. Level one, flat values: every conversion worth the same. Level two, revenue. Level three, margin-adjusted value, that's revenue times margin minus expected returns. Level four, predicted lifetime value, where a model estimates what this customer will be worth over time. Most brands should aim for level three. Level four is powerful, but only once you've checked the predictions against real outcomes.

2:13 Target ROAS from economics

Now, target ROAS. It isn't a wish, it comes from your economics. Break-even ROAS is one divided by your contribution margin. If you keep forty percent of revenue after product costs, shipping and fees, you break even at two point five. Start your target close to what the campaign actually achieves today, then move it in small steps. Setting a target far above reality simply throttles delivery, and the campaign stops spending.

2:44 Simple example (illustrative)

Here's a simple worked example to make the math concrete. A shop sells two products. A phone case for one thousand rupees with a sixty percent margin, and a charger for three thousand rupees with a fifteen percent margin. By revenue, the charger looks three times more valuable. By margin, the case earns six hundred rupees profit and the charger four hundred and fifty. So if you send revenue, the platform pushes chargers. If you send margin, it pushes cases, which actually earn more. Now imagine that across a whole catalog. That single change to the value you send quietly rewires which products your ads sell.

3:30 Example: Lahore fashion brand (illustrative)

Here's an illustrative example. A fashion brand in Lahore sells full-price new-season pieces with healthy margins, and clearance items with thin margins. They sent revenue as the value, so the platforms happily pushed clearance, lots of revenue, little profit. The fix was calculating margin-adjusted value on the server, adding a multiplier for new customers, and then judging the change on contribution profit over four weeks, not on ROAS.

4:00 Compute value server-side

Why server-side? Because anything calculated in the browser can be seen by anyone who opens developer tools, including your competitors. Compute the value on your server, where you already know the order, the product categories and the margins, and send it through the conversions API with the currency. Keep full revenue in your analytics for finance. The ad platform only needs the number you want it to maximize.

4:30 Don't misread the switch

A warning about comparisons. When you switch from revenue to margin values, reported ROAS will drop, because the numbers are smaller by design. That doesn't mean performance got worse. Compare contribution profit, new customer share and payback period from your own backend. Also expect a short learning period whenever you change value definitions, and watch that stacked value rules don't multiply into something absurd.

4:58 Values for lead gen

What about lead generation, where there's no order value at all? You create one. Take the average deal size, multiply by the historical close rate for each lead stage, and send that as the value. A form fill might be worth a small amount, a qualified lead more, and a booked meeting more again. Refine the numbers by segment where you can, for example company size or service line. It doesn't need to be perfect. It needs to rank leads roughly in the right order, so the model stops treating a student's enquiry and a hospital's procurement request as equal.

5:41 Predicted LTV safely

If you go further into predicted lifetime value, start simple. Segment averages are fine: first order category by acquisition channel gives a twelve-month value. Validate against a past cohort. If the ranking is right but the scale is off, calibrate. Cap extreme predictions so one outlier can't dominate bidding. And revisit every quarter, because prices and product mix change.

6:07 Mistakes + try this now

Let's name the common mistakes. Setting target ROAS at a dream number and wondering why spend collapses. Switching to margin values mid-quarter and comparing ROAS before and after. Computing margins in the browser where anyone can read them. And stacking value rules until one audience is worth five times another for no good reason. Try this now: take your last month's numbers, work out your contribution margin as a percentage, and calculate break-even ROAS as one divided by that margin. Then look at your current target. Is it above break-even, below it, or have you never checked?

6:49 Watch me do it: margin values (illustrative)

Watch me do it. I'll set up margin-based values for an illustrative Dubai electronics store. First, I pull ninety days of orders with product category, price, cost and returns. Second, I compute contribution margin per category after payment fees and shipping: accessories about fifty percent, laptops about twelve, phones about nine. Third, return rates by category from the same data. Fourth, I write the server-side function from the lesson with those rates, and I test it on five real orders, checking by hand that a single phone order produces a small value and a basket of accessories a larger one. Fifth, I decide the rollout: new value definition in a duplicated campaign first, with target ROAS removed for two weeks so the system can relearn on maximize value, then a target set near the observed ratio. Sixth, the reporting change: I add a column for contribution profit from the backend to the weekly table, and I brief the client that platform ROAS will drop by design. The single most important step there was the last one. Without the briefing, someone panics in week one and switches it back.

8:11 Recap and next step

Let's recap. Flat values make AI chase cheap conversions. Aim for margin-adjusted values, calculated server-side and sent with currency. Derive target ROAS from your contribution margin and move it gradually. And judge success on contribution profit, not platform ROAS. Your next step: calculate your break-even ROAS today, and write out the formula your business would use for margin-adjusted value, with the data source for every term.

Why value beats volume

If every conversion is worth the same to the platform, it will chase the cheapest ones. Real businesses do not work that way: a first order from a new customer may be worth more than a repeat order; a lead from a 300-person company is worth more than one from a freelancer; a high-margin product beats a discounted one. Value-based bidding (Google's Maximize conversion value / target ROAS, Meta's value optimization, TikTok's value-based optimization where available) lets you express those differences.

Four levels of value maturity

Level	Value sent	Good for	Risk
1. Flat	Same value per conversion	Very low volume	Chases cheap conversions
2. Revenue	Order revenue	Most e-commerce	Ignores margin and returns
3. Margin-adjusted	Revenue × margin %, minus expected returns	Mixed-margin catalogs	Needs data plumbing
4. Predicted LTV	Model-estimated lifetime value at conversion	Subscriptions, repeat purchase, B2B	Model error; must be calibrated

Most brands should reach level 3. Level 4 is powerful but only when the prediction is tested against actual outcomes.

Tools for expressing value

Conversion value rules (Google Ads) — adjust value by location, device or audience (for example, value new customers higher).
New customer acquisition goal (Google) — bid higher for new customers or optimize only for them, using customer lists and conversion data to identify existing customers.
Offline conversion adjustments — restate values later (e.g., after returns) where the platform supports adjustments.
Meta value optimization — optimizes for purchase value; needs reliable values in pixel and Conversions API events.
Profit data — Google Ads supports reporting with cost-of-goods data for retail in some setups; check current availability.

Setting a sensible tROAS

A target ROAS is not a wish; it follows from economics.

Break-even ROAS = 1 / contribution margin
If contribution margin after COGS, shipping and fees = 40%, break-even ROAS = 2.5
If you send margin-adjusted values instead of revenue, break-even "ROAS" on those values ≈ 1.0

Set the initial target near your recent actual ROAS, not your dream number, then move it in modest steps. Setting targets far above what the campaign achieves throttles delivery.

Worked example: a Lahore fashion brand

A fashion brand in Lahore sells full-price new-season items (60% gross margin) and clearance items (15% margin). It sends revenue as the conversion value, so the platforms push clearance — cheap, easy to sell, high revenue volume, low profit.

Fix:

The site calculates value = sum(item_price × item_margin_rate) − expected_return_cost and sends that in the purchase event (hiding exact margins from the client side by computing server-side).
A value rule adds a multiplier for new customers based on the brand's observed repeat rate.
The team compares four weeks of contribution profit before and after, not ROAS alone (ROAS on margin values will look lower by design).

Hands-on: margin-adjusted value in a server-side purchase handler

import os

MARGIN_BY_CATEGORY = {"new_season": 0.60, "core": 0.45, "clearance": 0.15}
EXPECTED_RETURN_RATE = {"new_season": 0.12, "core": 0.08, "clearance": 0.05}  # from your own data

def conversion_value(order: dict) -> float:
    value = 0.0
    for item in order["items"]:
        cat = item.get("category", "core")
        gross = item["price"] * item["quantity"]
        margin = gross * MARGIN_BY_CATEGORY.get(cat, 0.40)
        expected_return_loss = margin * EXPECTED_RETURN_RATE.get(cat, 0.08)
        value += margin - expected_return_loss
    return round(value, 2)

order = {"currency": "PKR", "items": [
    {"price": 8500, "quantity": 1, "category": "new_season"},
    {"price": 2000, "quantity": 2, "category": "clearance"}]}
print(conversion_value(order), order["currency"])

Send this value (with the currency) through your server-side conversion API. Keep revenue in your analytics tool for finance reporting; the ad platform gets the value you want it to maximize. The margin rates above are illustrative — use your own.

Predicted LTV: how to do it safely

Start simple: segment-level averages (e.g., first-order category × acquisition channel → 12-month value).
Validate: compare predicted vs actual for a past cohort. If the ranking is right but the scale is off, calibrate.
Cap extreme predictions so one outlier does not dominate bidding.
Revisit quarterly; product mix and pricing change.

Pitfalls

Changing value definitions mid-flight without expecting a learning period.
Comparing ROAS before and after switching to margin values (the scale changed).
Leaking margins to the browser by computing value client-side.
Letting value rules stack into absurd multipliers.

How to measure success

Contribution profit per week, new-customer share, and payback period — measured in your own backend — are the arbiters. Platform ROAS is a steering metric, not the scoreboard.

Flat values make the AI chase cheap conversions; value-based bidding expresses what you actually care about.
Aim for margin-adjusted values; use predicted LTV only when validated against actual outcomes.
Derive tROAS from economics (break-even = 1 / contribution margin) and move targets gradually.
Compute sensitive values server-side and send them with currency.
Judge success on contribution profit and new-customer share, not platform ROAS.

Calculate your break-even ROAS, then write the formula your business would use for a margin-adjusted conversion value. List the data sources needed for each term.

Enrol for free to save your progress

Reading is always free. Enrol to keep your place, take the final assessment and earn a verifiable certificate.

Conversion value strategy: teach the AI what profit looks like