IntermediateLesson 16 of 305 min read

Paid Social Advertising: Budgets, Creative Tests and Learning

Follow a social advert budget from showings to orders, compare creative versions fairly, and see why a small budget should optimise for an earlier action, with a worked example.

In this lesson you will

  • Connect CPM, CTR, CPC and CPA
  • Compare creatives by cost per order
  • Understand the learning phase
  • Plan a small budget sensibly

About 25 minutes

Noor puts 200 pounds into a social advert

Paid social means adverts shown inside social networks to people chosen by the platform’s targeting. Noor sets aside 200 pounds for her first test. Unlike search ads, where people are already looking, social adverts interrupt people who are scrolling, so the picture and the first line matter much more. This lesson walks through how the money turns into customers, how to test pictures and videos fairly, and why a small budget changes what you should ask the platform to optimise for.

How the numbers connect

  • CPM. Cost per thousand times the advert is shown.
  • CTR. The share of showings that become clicks.
  • CPC. What a click costs, which follows from CPM and CTR.
  • Conversion rate. The share of clicks that buy.
  • CPA. Cost per order, which is budget divided by orders.

These multiply together, as in lesson 7. Noor’s made-up numbers are an 8 pound CPM, a 1.2 per cent click rate and a 3 per cent conversion rate. The program below follows 200 pounds through those steps, compares three versions of the advert, and asks a question many beginners skip: is there enough money to give the platform what it needs?

# MADE-UP paid social test, pounds
budget, cpm, ctr, conv = 200.0, 8.0, 0.012, 0.03
impressions = budget / cpm * 1000
clicks = impressions * ctr
orders = clicks * conv
print("impressions:", round(impressions))
print("clicks:", round(clicks))
print("cost per click:", round(budget / clicks, 2))
print("orders:", round(orders, 1))
print("cost per order:", round(budget / orders, 2))

# three creatives given the same budget share
ads = {"photo": (66.0, 4, 22), "video": (67.0, 9, 31), "quote": (67.0, 3, 15)}
for name, (spend, orders_, clicks_) in ads.items():
    print(f"{name}: spend {spend:.0f}, orders {orders_}, clicks {clicks_}")
    print(f"   per order {spend / orders_:.2f}")
print("lowest cost per order:", min(ads, key=lambda a: ads[a][0] / ads[a][1]))

# can the budget buy enough events to learn from?
need = 50
cpa = budget / orders
print("orders wanted in a week:", need)
print("budget that needs:", round(need * cpa))
print("enough with", int(budget), ":", budget >= need * cpa)
print("cost per basket add if 1 in 5:", round(budget / (orders * 5), 2))
impressions: 25000
clicks: 300
cost per click: 0.67
orders: 9.0
cost per order: 22.22
photo: spend 66, orders 4, clicks 22
   per order 16.50
video: spend 67, orders 9, clicks 31
   per order 7.44
quote: spend 67, orders 3, clicks 15
   per order 22.33
lowest cost per order: video
orders wanted in a week: 50
budget that needs: 1111
enough with 200 : False
cost per basket add if 1 in 5: 4.44

The 200 pound budget buys about 25000 showings, which gives 300 clicks at 0.67 pounds each and 9.0 orders, so each order cost 22.22 pounds. For the creative test, the video cost 7.44 pounds per order, the photo 16.50 and the quote graphic 22.33. With only a handful of orders each the test is weak, but the video is clearly ahead and worth running again with a bit more money. The last block matters most. Platforms that optimise delivery use the results to find similar people. Meta’s help pages describe a learning phase and give a rough guide of about 50 optimisation events in a week for each ad set. To buy 50 orders at 22.22 pounds each, Noor would need about 1111 pounds a week, and she has 200. So she is unlikely to get out of the learning phase if she optimises for purchases.

What to do with a small budget

  1. Optimise for an earlier action. Basket additions or signups happen more often than purchases. At 1 in 5 baskets turning into orders, each basket add cost about 4.44 pounds, which makes the arithmetic of 50 events much more achievable.
  2. Keep it simple. One audience, two or three creatives, one goal.
  3. Do not keep editing. Meta’s pages say big edits can restart learning, so give it a few days.
  4. Test one thing at a time. Change the picture or the headline, not both.
  5. Judge on cost per order against margin as in lesson 7, and stop when the sums do not work.
Common mistake: liking the likes

Judging on reach and likes. A cheap advert that gets many likes and no orders is a loss. Check orders and cost per order first, and use the other numbers only to explain why.

Check each platform’s current rules

Targeting options, rules about what you can advertise, and the exact learning rules change often and differ between platforms. The 50-event figure is a rough guide from the platform’s own help pages at the time of writing, so check the current page. Never target in ways the rules or the law forbid.

Work out your own CPM, click rate and conversion rate, even from a guess. Compute your cost per order, and how many dollars or pounds it would take to buy 50 of them in a week.

Practice for this lesson

Check what you learned

Five quick questions. Pick an answer and you will see straight away whether it is right and why.

What does CPA mean?

  1. Clicks per advert
  2. Cost per page
  3. Cost per post
  4. Cost per order or acquisition

It is budget divided by orders.

Which ad was cheapest per order in the test?

  1. The video
  2. The photo
  3. The quote
  4. They tied

The video cost the least per order.

Why can a small budget struggle to leave the learning phase?

  1. Small budgets are banned
  2. It is too fast
  3. It cannot buy enough events
  4. It has no ads

About 50 events in a week needs more spend than 200 pounds.

What can you optimise for instead of purchases?

  1. Likes only
  2. Nothing
  3. Page colour
  4. An earlier action like a basket add

Earlier actions happen more often, so you reach enough events.

Why avoid frequent edits during learning?

  1. Edits are free
  2. Big edits can restart learning
  3. Edits are slow
  4. The platform dislikes ads

Platform help pages say significant edits can reset learning.

Summary

  • Showings, clicks and orders multiply, so each step matters.
  • Judge adverts by cost per order against margin.
  • Platforms need enough events to learn; small budgets should aim for earlier actions.
  • Test one thing at a time and avoid constant edits.

Next lesson

Lesson 17 covers analytics and KPIs.

Questions

Is 50 events a fixed rule?

No. It is a rough guide that the platform's help pages mention for its learning phase. It may change, so check the current page.

Are the numbers real?

No. They are made-up sample numbers in pounds to show the arithmetic.

Should I use broad or narrow targeting?

It depends on the platform and goal. Very narrow audiences can limit delivery. Check the platform's guidance and test.

Progress is saved in this browser.