On this page 10 sections
A/B TESTING · CONVERSION MEASUREMENT · GOOGLE ADS INTEGRATION
Buying traffic is straightforward. Turning it into measurable growth takes a system.
Advertising brings people to your website. What happens next determines whether that investment creates value. Marketing Pro connects acquisition, website experiments and business outcomes to help you understand which campaigns, messages and experiences generate qualified enquiries—and which generate customers.
Review my measurementSpending more than €1,000 a month on advertising? Start by checking whether your measurement reflects your actual business results.
The short version
Specific searches, tested pages and connected outcomes all feed one question: which part of your marketing produces qualified enquiries and sales. This guide covers each piece, and the two Google APIs that can carry data between your CRM and your reports.
- 01Ad interaction
- 02Enquiry
- 03Qualified lead
- 04Sale
Measurement feeds back into the start of the journey: outcomes reported at each step inform what you buy next. This is a feedback loop, not a complete attribution model.
Search intent is becoming more explicit. Your website needs to respond.
Compare a search for “Catering London” with this one: “Who provides office catering for 30 people in Central London, with vegetarian options and delivery before noon?” The second states the headcount, the area, the dietary needs and the delivery deadline. The visitor has already told you what a good answer looks like.
Conversational tools such as ChatGPT encourage this style, including follow-up questions that refine the requirement. Short searches have not disappeared, though, and they remain relevant.
Long-tail means specific, typically lower individual-volume queries. It does not simply mean long phrases. A relevant page answers the questions a specific buyer has:
Questions a specific buyer has
- Delivery area
- Dietary options
- Minimum quantities
- Lead times
- Quote requirements
Your website should operate as a measurable hypothesis.
A/B testing is the random assignment of eligible visitors to a control or a variation, measured against an outcome defined in advance. Headlines, information order and forms all embody assumptions about what visitors need. A test turns those assumptions into something you can check.
Changing many things at once tests the overall package, not the causal contribution of each individual change. That can be a valid choice, as long as you read the result that way.
| Control | Variation | Hypothesis |
|---|---|---|
| Office Catering in London | Office Catering Planned Around Your Team’s Dietary Requirements | Relevance improves qualified enquiries. |
| Open-message form | Structured form: event date, headcount, postcode | Structure improves qualification. |
| Request a quote | Explanation of what the quote includes | Clarity improves enquiry quality. |
Define success before launching the experiment.
Metrics to define
- Conversion rate: the predefined conversions divided by eligible visitors, expressed as a percentage.
- Qualified-lead rate: qualified leads divided by visitors for this experiment. If you instead divide by enquiries, label that different denominator explicitly.
- Cost per qualified lead: the stated advertising or acquisition spend divided by qualified leads in a comparable period.
- Acquisition cost: the stated cost divided by new customers. Say whether the scope includes media only, or also creative, agency and sales costs.
- Revenue per visitor: attributable revenue divided by eligible visitors. Contribution margin per visitor instead uses revenue less the stated variable costs, with the same observation window.
Guardrails
Metrics that must not get worse while the primary metric improves:
- Speed
- Errors
- Lead quality
- Cancellations
- Sales workload
| Measure | Variant A | Variant B |
|---|---|---|
| Visitors | 1,000 | 1,000 |
| Enquiries | 60 | 45 |
| Qualified leads | 12 | 18 |
| Enquiry conversion | 6% | 4.5% |
| Qualified leads per visitor | 1.2% | 1.8% |
These figures are illustrative, not real results or sufficient evidence of statistical significance. Variant A produces more enquiries, yet Variant B produces more qualified leads. More enquiries can mean worse qualification; assess uncertainty using the predefined analysis rather than declaring a winner from this table.
A reliable test requires more than two page designs.
Design decisions
- Assignment unit, and a consistent returning experience where permitted
- Minimum detectable effect
- Baseline, power and uncertainty, which together determine the sample needed
- Fixed-horizon versus sequential analysis
- Delayed conversion windows
Choose the unit before assignment: for example, an eligible visitor rather than each page view. Where permitted, keep returning visitors in the same experience so exposure is not mixed.
The minimum detectable effect is the smallest change the test is designed to detect. Baseline conversion, that effect, desired power and the uncertainty you can tolerate determine sample needs—not a universal visitor count.
A fixed-horizon test waits for its planned sample and observation window. Sequential analysis permits interim decisions only with a method and stopping boundaries designed for that purpose.
Integrity checks
- Predefined stopping rules
- False-positive risk from repeated peeking at results
- Event validation
- Sample ratio mismatch checks
Repeatedly checking an ordinary fixed-horizon test and stopping when it looks positive increases false positives. Decide stopping rules in advance and allow delayed enquiries and sales time to mature.
Validate events before launch. Check sample ratio mismatch: an unexpected imbalance between the planned and observed assignment proportions can reveal a broken experiment, not a winning variation.
Positive, negative and inconclusive results are all valid outcomes.
Measure the journey beyond the initial click.
- 01Ad interaction
- 02Visit
- 03Enquiry
- 04Qualified lead
- 05Opportunity
- 06Sale
Fields worth capturing
Define qualification criteria before the test. For a London catering enquiry, these might include a serviceable postcode, a feasible event date and headcount, dietary requirements the team can meet, and a genuine contactable buyer. Apply the same rules to both variants.
- Unique lead or transaction ID
- Event and timestamp
- Experiment and variant
- Permitted ad identifiers
- Qualification status
- Meaningful value and currency
- Consent status
Be precise about what a click means
Phone-link clicks are not connected calls.
WhatsApp clicks are not conversations.
Track the outcome you care about, not the nearest proxy for it.
Google Data Manager API: connect first-party outcomes to advertising measurement.
The Data Manager API supports first-party audience and event ingestion into Google products, including qualified-lead and transaction workflows, subject to destination and matching requirements.
What a sound workflow involves
- Capturing permitted identifiers
- Consistent CRM records
- Supported status-change events
- Supported deduplication and safe retries
- Delivery diagnostics
Limits to keep in mind
API acceptance does not guarantee matching or reporting.
Server-side delivery does not remove consent requirements or data gaps.
Ingestion: sends first-party events and audiences into Google products.
Reporting and operations: reads results and manages campaigns programmatically.
Google Ads API: build reporting and operational control around your business.
The Google Ads API supports programmatic reporting and campaign management. Typical uses include:
- Dashboards and spending checks
- Conversion actions and values
- CRM and experiment reporting
- Controlled campaign updates
Reporting uses GAQL, the Google Ads Query Language, to select resources, metrics and segments. Joined reporting needs compatible identifiers, periods and attribution definitions. Recent conversion data can change, and conversions can arrive with latency, so recent periods should be read with care.
For example, a campaign is a resource, cost and conversions are metrics, and date or device are segments that break the results down. Not every field combination is compatible. A dashboard joining CRM or experiment outcomes must preserve the agreed identifiers and measurement definitions.
Attribution and experimentation answer different questions.
| Approach | Question it answers | Limit |
|---|---|---|
| Attribution | How should credit be assigned across touchpoints? | Assigns credit; does not by itself prove causation. |
| Experiments | What is the causal effect of this change? | Only answers the question the design was built for. |
A website A/B test does not establish advertising incrementality. Measuring whether advertising itself adds outcomes requires appropriate holdout designs.
Spending more than €1,000 a month? Audit your feedback loop.
At that level, you are spending over €12,000 annually. The questions below show whether decisions rest on solid evidence:
- Are events accurate and unduplicated?
- Are enquiries qualified?
- Are customers linked back to their source?
- Are decisions based on outcomes?
- Is the sample adequate?
- Are blind spots known?
The €1,000 figure is a prompt for review. It is not an API requirement and not a guarantee of sufficient sample size. Lower volumes may need measurement and usability fixes first.
How Marketing Pro approaches the work.
- Stage 01
Measurement audit
We review what is tracked today, how events fire and where the gaps sit.
- Stage 02
Outcome definitions
We agree what counts as a qualified lead, an opportunity and a sale, with explicit denominators.
- Stage 03
Simplest suitable integrations
Where existing connectors are enough, we use them. APIs are for cases that need them.
- Stage 04
Experiment design
Hypothesis, assignment unit, primary metric, guardrails and stopping rules, fixed before launch.
- Stage 05
Validation
Event checks, deduplication checks and sample ratio checks before any result is trusted.
- Stage 06
Analysis and iteration
We report results with their uncertainty and commercial relevance, including inconclusive ones.
Questions we are often asked.
Do I need a dynamic website to run experiments?
No. Static sites can support experiments. Personalisation is a separate matter and has its own requirements.
Are longer search queries always better?
No. Intent, relevance and the offer matter more than query length. Short searches remain relevant.
Do I need both the Google Data Manager API and the Google Ads API?
No. They have different roles, and your existing tools may be enough for your situation.
What if my traffic is limited?
We assess feasibility first. Few conversions mean slower conclusions, and sometimes measurement or usability fixes come before testing.
Can server-side measurement recover every conversion?
No. Consent, matching and data quality all set limits on what can be recovered.
Can you guarantee sales growth?
No. Disciplined testing can also reject changes that do not work, which protects you from rolling them out.
Make your next marketing decision with better evidence.
Bring us your website, your campaigns and the business outcome you want to improve. We will identify what you can measure today, where the gaps are and which changes are worth testing.