Paid Search
Date: 2026-08-16
Buying placement against expressed intent. It looks efficient in every attribution model precisely because it captures demand that already exists — which is also why a large share of it isn’t incremental.
What it is
Paid search places ads against search queries, charged per click, allocated by auction on bid and quality.
The defining property: the customer arrived with a stated intention. That’s why it converts well and why measuring its contribution is hard.
Brand and non-brand are different channels
The single most important split, and blending them hides everything:
| Branded | Non-brand | |
|---|---|---|
| Query | Your name | Category or product terms |
| Conversion rate | High | Lower |
| Cost per click | Low | Higher |
| Apparent ROAS | Excellent | Modest |
| Incrementality | Often low | Higher |
Someone searching your brand name was already coming. Paying for that click frequently buys a visit you’d have received organically — the classic incrementality question, and the classic finding is that a substantial share of branded paid clicks are not incremental.
Report them as separate channels, always — Channel Taxonomy.
The counter-argument for bidding on brand: competitors bidding on your name, and controlling the message. Both are real. The point isn’t never to bid, it’s that branded ROAS is not evidence the spend works — Incrementality Testing.
Testing brand incrementality
The most valuable single test available in paid search, and it’s straightforward:
pause branded paid search in a set of regions
compare total branded traffic and orders
against matched control regions
→ how much traffic simply moved to the organic listing?
The usual finding is that most of it does, at zero cost. A geo holdout answers this in a few weeks and often reallocates a meaningful budget.
The economics
On the running model — £22.50 CAC, £15 first-order contribution:
cost per click £0.90
conversion rate 4.0%
cost per order £22.50
first-order contribution £15.00
────────
first order −£7.50
Paid search rarely pays back on the first order, so it depends entirely on repeat purchase — which makes Repeat Purchase Rate the number that decides whether the channel works, not the return-on-ad-spend figure in the platform.
Where the platform reporting misleads
- Its conversions include view-through and modelled figures — Modelled Conversions, Walled Garden Reporting
- It attributes on its own model and window — Attribution Windows
- Automated bidding optimises to the platform’s conversion definition, which may include returns, cancellations and test orders unless you feed it clean data
- Smart bidding needs good conversion data. Feeding it revenue rather than contribution optimises for the wrong thing, and feeding it inflated conversions optimises for noise — Bid Strategies and Budget Allocation
Feeding contribution rather than revenue as the conversion value is the highest-return change available in most paid search accounts, and it requires only that you know your margin by product — Contribution Margin.
Managing it
- Split brand and non-brand budgets and reporting
- Judge on marginal CAC, not average — the next £10,000 costs more than the last — Marginal Analysis
- Negative keywords are the cheapest optimisation there is
- Watch new versus returning. Paying to reacquire existing customers inflates apparent volume — New vs Returning Customer Acquisition
- Landing page match matters more than bid at the margin — Landing Page Strategy