Search advertising does not buy intent. It buys the right to enter an auction when a query may express that intent. Targeting, messaging, the landing page, measurement and margin sit between the two.
1. Definition: search advertising responds to demand; it does not qualify it on its own
Search advertising is the purchase of advertising placements triggered by queries on a search engine or comparable environment. The advertiser chooses objectives, targeting, messages, pages and data; the platform uses bids, quality and signals to select and rank adverts.
This channel works particularly well when demand can articulate its need: a product, category, problem, brand, location or comparison. It is less effective when the audience is unaware of the solution, volume is too low or the difference requires extensive education. In those cases, content, social media, video, partners or direct sales often need to create the context upstream.
A click is never the final outcome. A profitable campaign turns media cost into incremental margin, within a timeframe compatible with cash flow and at an acceptable level of customer quality.
2. Key figures: a vast, concentrated and increasingly automated market
| Figure | Source and scope | Useful interpretation | Limitation |
|---|---|---|---|
| €131bn | European digital advertising market in 2025, 30 countries, IAB Europe. | Digital media remains a major acquisition infrastructure. | Sell-side spend, not advertiser profitability. |
| +10.5% | European growth in 2025, IAB Europe. | Auctions are evolving in a market that is still expanding. | Growth varies across countries and formats. |
| +8.8% | Growth in European paid search in 2025, IAB Europe. | Search is growing more slowly than social media and video, but retains structural importance. | The report groups paid search under harmonised definitions. |
| €12.4bn | French digital advertising market in 2025, up 11%, SRI/UDECAM. | The French market is following a strong growth trajectory. | A sell-side view combining declarations, interviews and estimates. |
| €4.93bn | French search advertising in 2025, representing 40% of the market and up 10%, SRI. | Search remains France’s largest channel by revenue. | Includes various forms of search and different participants under the methodology. |
| 76% | Share captured by eight global participants; 83% of growth, SRI. | Dependence on platforms makes data control and allocation scenarios essential. | A macro-level concentration figure, not the structure of every account. |
| 200 conversions / 2,000 interactions | Volume recommended by Google over 30 days to improve data-driven attribution, Google Ads. | Smaller accounts should avoid fragmenting their signals excessively. | DDA also works below this volume; Google presents it as a recommendation. |
| 4 to 6 weeks | Minimum duration recommended by Google when an experiment remains inconclusive, Google Ads Experiments. | Weekly decisions can interrupt learning too early. | The required time depends on volume, effect size and the conversion cycle. |
| 2 to 4 arms | Number of groups available in certain Google Ads video tests, Google. | An experiment should limit variants to preserve statistical power. | This reference point does not apply to every type of experiment. |
| €1 → €2 | Google example of an iROAS of 2: two euros of incremental value for every euro of cost, Conversion Lift. | Incrementality measures what would not have happened without media investment. | An illustrative example, not a performance benchmark. |
The market shows where money is flowing. It does not tell an SME whether to increase its budget. That decision comes from unit economics, operational capacity and measurement capable of distinguishing captured demand from created demand.
3. The profit-and-loss account of a search query
Calculate the thresholds before designing the campaign structure. For ecommerce: net price, product cost, fulfilment, payment, returns, support and contribution margin. For a lead: qualification rate, meetings, proposals, sales, margin and time to cash.
Maximum customer CPA = expected contribution margin × acceptable share allocated to acquisition.
Maximum lead CPA = maximum customer CPA × lead-to-customer conversion rate.
A purely illustrative Logiks example: a sale generates €1,200 in contribution margin and the company accepts allocating 35% of it to acquisition. The target customer CPA is €420. If 20% of leads become customers, the maximum lead CPA is €84. With a conversion rate of only 8%, it falls to €33.60.
The same Google Ads account can therefore appear profitable or destructive depending on sales quality. Importing actual sales and their value changes decisions more than gaining one percentage point of click-through rate.
Include the time dimension. A customer who becomes profitable after eighteen months does not help the cash flow of a business paying for clicks today. Segment first purchase, margin, repeat purchase, cancellation and bad debt. LTV should become a bidding input only when it is observed, consistent by cohort and compatible with financial capacity.
4. Seven management levers, in the right order
4.1. Define the conversion that deserves a bid
A page view, button click or form start can support diagnosis. It is not automatically a primary conversion. Automated strategies optimise what you give them; if the objective mixes qualified demand, spam and micro-actions, they learn to produce the easiest signal.
Classify actions: primary for bidding, secondary for observation and diagnostic for the funnel. Deduplicate tags and imports. For leads, send qualification, opportunity and sale status back from the CRM using a click identifier or compliant mechanism.
Deliverable: conversion dictionary, owner, source, window, value, deduplication rules and end-to-end test.
4.2. Map queries by unit economics
Keywords are only a matching mechanism. The actual query reveals more: brand, competitor, category, problem, urgency, local intent, information or support.
Create economic groupings. Brand campaigns often capture demand that already exists and should not conceal the profitability of non-brand search. Highly specific queries have little volume but strong intent. Informational terms require content and delayed value. Competitor terms carry cost, trade-mark considerations and particular expectations.
Review search terms regularly, but do not turn the negative-keyword list into a museum. Every exclusion should protect an intent or an economic case. Broader match types work better when conversions and values are reliable; they become dangerous when the objective is noisy.
Deliverable: governed query-intent-page-value taxonomy and negative-keyword list.
4.3. Build a structure that concentrates signal
Traditional granularity — one campaign for every detail and one ad group per keyword — can starve automation of data and multiply inconsistencies. Conversely, a fully consolidated account conceals genuinely different geographies, budgets, margins or constraints.
Separate where a decision needs to remain separate: objective, budget, country, language, brand/non-brand, margin type, network or regulation. Consolidate what shares a conversion, value, audience and allocation decision.
Naming conventions should make the account intelligible without tribal knowledge. Document settings, exclusions, audiences, schedules, URLs, owners and change history.
Deliverable: account map linked to every budget decision.
4.4. Write the advert as a relevance contract
The advert connects the query, offer and page. It must qualify as much as it attracts. A price, timeframe, location, condition or audience can reduce CTR while improving economic quality.
Build angles around outcome, evidence, difference, risk reduction and action. Assets provide more surfaces, but every line must remain accurate when combined. Avoid pinning every position without good reason; equally, do not entrust a regulated claim to an unreviewed combination.
Test one hypothesis, not twenty wordings. For example: displaying the minimum price will reduce leads without sufficient budget and increase the qualification rate. Measure the outcome through to the CRM.
Deliverable: intent-evidence-promise matrix with compliance rules.
4.5. Make the landing page an extension of the query
Within seconds, the page should confirm: the right offer, the right audience and the next step. Echo the language of the query without mechanically copying the keyword. Show evidence, conditions, objections and alternatives.
Technical performance matters, but so does continuity. An advert promising a “quote within 24 hours” that leads to a generic page and receives a response five days later destroys the promise. The form should request only the data required for processing and explain what happens next.
Segmenting a page by intent is useful when content and economics differ. Multiplying near-identical variants for every query creates debt, inconsistency and SEO risk.
Deliverable: page with tested messaging, evidence, form, tracking and operational response time.
4.6. Choose a bidding strategy that matches data maturity
Automated bidding can use many signals, but it depends on the objective and volume. Begin by verifying the conversion, delay and value. Do not change the structure, target, budget and page at the same time: you will lose the ability to explain the outcome.
Target CPA suits relatively homogeneous value; target ROAS becomes more appropriate when values differ and reflect margin or outcome. An overly constrained budget, unrealistic target or daily changes restrict delivery and disrupt learning.
Plan for exceptional events. Promotions and outages must not become the model’s new normal. Document seasonality, adjustments and the return to standard settings.
Deliverable: bidding rule, economic threshold, volume, observation period and conditions for change.
4.7. Allocate budget on marginal returns, not average ROAS
An average ROAS of 5 does not say what the next euro will produce. Scaling often raises marginal cost through new, less profitable queries, positions, times, audiences or geographies.
Build a stepped response curve. Increase spend in a controlled way, allow the cycle to complete, and observe marginal CPA, value, quality and fulfilment capacity. Retain a control group or geography when volume permits incrementality testing.
Google defines iROAS as incremental value divided by incremental cost. This measure answers “what would have happened without the campaign?” better than attribution, which allocates credit among observed interactions.
Deliverable: base/high/low budget scenario, stop threshold and marginal analysis.
5. The management dashboard that connects the platform with finance
| Level | Measures | Decision |
|---|---|---|
| Auction | eligible impressions, CPC, queries, quality and coverage. | Where does the campaign enter, and at what cost? |
| Journey | click, meaningful engagement, form, errors and speed. | Does the promise hold after the click? |
| CRM / sales | qualification, meeting, sale, delay and cancellation. | Are the conversions actually customers? |
| Economics | margin, customer CPA, margin ROAS, cash payback and observed LTV. | Does the media investment create an acceptable contribution? |
| Causality | lift, iROAS and confidence interval. | What share would not have existed without the investment? |
Reporting should specify conversion date and click date, windows, attribution model, taxes, refunds and missing data. Without a shared dictionary, two teams can report two accurate yet incompatible ROAS figures.
6. Three value, volume and margin decisions to make before changing bids
A search campaign becomes genuinely manageable when the team can respond to three very different situations: demand that exists but converts poorly, demand that is profitable but too narrow, and attributed revenue that produces insufficient contribution after variable costs. The dashboards can look similar. The decisions are very different.
6.1. Scenario A — Demand is present, but conversion is breaking down
Consider a B2B software company buying 3,000 clicks a month at €4, generating 90 demo requests and converting 18 customers: cost per lead is €133 and media acquisition cost is €667. Raising bids will fix neither an overly demanding form nor a sales process that calls prospects back five days later. The leak must be located.
Break the journey into useful micro-steps: qualified arrival, evidence viewed, form started, form submitted, qualification, meeting attended, proposal and signature. This analysis is not intended to multiply optimisation objectives; it identifies the failing link before an instruction is given to the algorithm.
One example is enough. If 8% of clicks become enquiries but only 20% of those enquiries pass qualification, the priority is the promise, exclusions and form questions. If 60% qualify but few meetings take place, the issue shifts to response time and sales preparation. The same cost per lead; a different problem.
6.2. Scenario B — The economics work, but the pool is narrow
When a set of queries produces satisfactory contribution and already covers most available impressions, increasing the budget creates little growth: the campaign reaches a demand ceiling, while marginal cost rises because the remaining auctions involve less favourable positions, times or intents. Expansion must be methodical.
The team can explore adjacent phrasings, problems higher in the decision journey, new geographies or entry offers; however, these extensions must be isolated so that their initial performance does not dilute the established core. Every expansion needs a hypothesis, an acceptable loss ceiling and a dated decision rule.
Search does not create its market on its own. When non-brand coverage plateaus despite good quality, content, referrals, public relations, video or social media can nurture future demand. This interaction calls for a combined view of brand queries, category searches and awareness signals. Search then captures part of work that began elsewhere.
6.3. Scenario C — ROAS looks good, but contribution is deteriorating
Consider two campaigns, each reporting 500% ROAS: the first sells a high-retention subscription with low service costs; the second sells a discounted product whose fulfilment, returns and commissions absorb most of the revenue. Their platform metric is identical. Their ability to fund the business is not.
Management should therefore import, at a minimum, product category, new or existing customer status, discounts, cancellations and, where available, an estimate of contribution or customer value. Without these nuances, automation quite reasonably favours what it has been told is valuable, even when that declaration is economically incomplete. The data sent to the platform is an allocation policy.
One simple rule remains. Whenever an increase in spend is considered, compare the outcome from the additional euros with the financial threshold, not the average inherited from previous weeks. A historically profitable portfolio can destroy contribution on its latest investment tranche while retaining a reassuring average ROAS.
7. The evidence file to retain for every important decision
Serious optimisation leaves an audit trail. For a change in bidding, targeting, page or budget, the minimum record contains the hypothesis, primary metric, guardrail, observation period, known external factors and decision taken. This discipline prevents history being reinterpreted after the event when results become ambiguous.
Document anomalies as well. A promotion, stockout, form outage, price change or delayed CRM import can make two periods incomparable, even when the interface confidently calculates a percentage change. Operational memory protects decisions better than isolated screenshots.
It is concrete. And verifiable.
8. An eight-week launch method
8.1. Weeks 1 and 2 — Economics and measurement
- calculate maximum customer/lead CPA and value per outcome;
- audit conversions, consent, CRM and deduplication;
- establish a baseline for demand, brand and seasonality;
- assess sales capacity and stock;
- select the intents to test.
8.2. Weeks 3 and 4 — Account and messaging
- structure brand, non-brand and economic constraints;
- write adverts and assets;
- build pages and forms;
- test sales imports;
- set the budget, initial bidding strategy and guardrails.
8.3. Weeks 5 to 8 — Controlled learning
- analyse search terms and lead quality;
- correct noise, the page and operational process;
- avoid simultaneous changes;
- launch a documented experiment;
- decide whether to increase, maintain, revise or stop.
9. Logiks recommendations: signs of an account controlled by the platform
- Every recommendation is applied without a business case.
- ROAS includes revenue but ignores margin and returns.
- Brand and non-brand share the same verdict.
- Micro-conversions drive bidding.
- The CRM sends no quality signal back.
- The target changes after three poor days.
- Budget increases on the basis of average ROAS without marginal analysis.
- Nobody knows how to restore the previous configuration.
Our rule: no automation is granted more autonomy than data quality and the ability to exercise control can support.
10. FAQ
10.1. How much should you invest in search advertising?
Start with demand volume, maximum CPA, conversion rate and operational capacity. A test budget must generate enough conversions to learn without exceeding the acceptable loss. There is no universal minimum.
10.2. What is a good ROAS?
One that covers margin, variable costs, acquisition, returns and financial objectives. A revenue ROAS of 4 can be excellent or loss-making depending on margin. Calculate the threshold, then observe the marginal increment.
10.3. Should you bid on your own brand?
Brand bidding can protect space, control the message or support partners, but it often captures existing demand. Measure competition, SEO, cannibalisation and incrementality; always report it separately from non-brand activity.
10.4. Is broad match dangerous?
Broad match increases discovery and relies on automation. It becomes risky with noisy objectives, weak query control or generic pages. Test it within a defined scope, with qualified conversions and governed negatives.
10.5. How long should you wait before optimising?
Monitor errors and out-of-scope spend immediately. To judge performance, respect the conversion cycle and volume. Google sometimes recommends four to six weeks for inconclusive experiments; that does not prevent you from correcting an obvious fault.
10.6. Can Google Ads measure offline sales?
Yes, through suitable imports and connections, subject to lawful collection, identifiers, CRM quality and Google’s rules. Test matching, delay, deduplication and value before using these data to drive bidding.
11. Conclusion
Search advertising becomes a growth channel when it connects query, promise, journey, sale and margin. Without that chain, automation optimises a platform metric and the business pays to learn the wrong lesson.
Control does not mean setting everything manually. It means defining the economics, providing a reliable signal, authorising the algorithm within that framework and retaining the evidence needed to change course.
12. Main sources
- IAB Europe — AdEx Benchmark 2025, 7 July 2026.
- SRI/UDECAM/Oliver Wyman — 35th Online Advertising Observatory, 10 February 2026.
- Google Ads — Data-driven attribution, accessed 13 July 2026.
- Google Ads — Attribution models, accessed 13 July 2026.
- Google Ads — Experiments, accessed 13 July 2026.
- Google Ads — Geo-based Conversion Lift metrics, accessed 13 July 2026.
- Google Ads — Customer data policies, accessed 13 July 2026.
- CNIL — Consolidated cookie recommendation, January 2026.