Measurement
Part of What social media agencies measurement means for buying and reporting
Choosing and auditing social media attribution methods
Learn how social media attribution methods assign credit, compare named models and platform reports, and choose and audit a method against client records.
What to take away
- You have to pick one attribution method to report on. That choice shapes every number a client sees.
- Start with the method the client can auditplatform-reported last click reconciled against CRM records. Add multi-touch only when the client can supply data to support it.
- Ask which model the agency reports on, which sources sit behind it, and where client-side tracking runs out.
- Platform figures and client records will differ. Fix a tolerance before the first invoice.
- Score methods on the evidence they produce, not on how current they sound.
Where credit goes in each attribution family
Attribution methods for social media agencies cluster into three groups. Last-touch models credit the final click before a purchase or sign-up. First-touch models credit the earliest recorded interaction. Multi-touch and data-driven models share credit across several touchpoints, using fixed rules or modelled probabilities.
Named variants make the differences clearer. Last-click is a last-touch variant; last non-direct click credits the final eligible visit before conversion while excluding direct traffic. Linear attribution shares credit across recorded touchpoints, time decay gives more weight to nearer interactions, and position-based attribution gives greater weight to selected first and last touchpoints. View-through credits an eligible ad impression without a click, while engaged-view credits a qualifying video view without a click, according to the platform's definition.
Attribution families compared
Last-touch
- Credit goes to
- Final click
- Answers
- Who closed
- Evidence base
- Observed click
- Audit strength
- Defensible
- Best use
- Closing report
First-touch
- Credit goes to
- Earliest interaction
- Answers
- Who opened
- Evidence base
- Observed click
- Audit strength
- Defensible
- Best use
- Opening report
Multi-touch / data-driven
- Credit goes to
- Several touchpoints
- Answers
- How credit spreads
- Evidence base
- Rules or probabilities
- Audit strength
- Weak for audit
- Best use
- Direction only
Each group answers a different question. Last-touch shows who closed the sale. First-touch shows who opened the relationship. Neither says much about which piece of creative persuaded anyone.
Modelled methods fill gaps with probabilities rather than observations. That makes them useful for direction and weak for audit. Label modelled or data-driven figures in the report with the model name, the inputs and the date the model last changed.
Where a client needs numbers they can defend to a finance team, platform-reported figures reconciled to client records are easier to stand behind.
The underlying record is patchy. Browser restrictions, app tracking prompts and consent choices mean the trail from impression to purchase is often incomplete. The ICO's guide to PECR explains the consent rules for cookies and electronic marketing that decide how much of that trail you may hold. The UK GDPR and the Data Protection Act 2018 also apply to the processing of personal data. Where consent is absent, agencies fall back on modelled or aggregated data.
Reporting discipline matters as much as the model. Social Partner's measurement and reporting guide sets out how monthly reporting, quarterly reviews and contract schedules fit together.
A scoring rubric for comparing attribution methods
Score each method on five criteria before you choose. The weights below are illustrative, not industry standards. Change them to match what the client actually has to prove. A retailer with a clean CRM might weight reconciliation highly. A brand with no direct sales may weight provenance instead.
Take a homeware retailer that runs Google Ads and Meta campaigns and takes orders through its own checkout. Its CRM holds confirmed sales, so reconciliation could carry a high weight. A wholesaler that sells through stockists has no direct order record, so provenance and transparency matter more.
| Criterion | What good looks like | Weight |
|---|---|---|
| Provenance | Named sources, documented joins, known gaps | 25 |
| Lawful basis | Consent or another basis recorded per channel | 20 |
| Reconciliation | A signed tolerance band and escalation route | 25 |
| Transparency | Rules or modelled assumptions written down | 15 |
| Review cadence | Fixed dates for re-scoring and renegotiation | 15 |
How to run the score
Ask each shortlisted agency for a one-page method note. It should name the model and variant, the reporting platform and its attribution setting, the conversion source, the data joins and the known blind spots. Score the submissions independently, then compare notes as a panel.
A method that scores well on provenance and reconciliation is usually worth more than a fashionable one. Set a pass mark before you read anything.
When building a UK shortlist, you could ask agencies such as We Are Social, Socially Powerful and The Goat Agency for the same method note. Their names alone are not evidence that their attribution approaches are comparable.
Audit the tracking before comparing results
Use this checklist with the agency and client team:
- Test that the Meta Pixel and Conversions API record the agreed events, and check that matching browser and server events use the agreed event identifiers and are deduplicated.
- Check that campaign links follow an agreed UTM taxonomy, with consistent source, medium and campaign values.
- Confirm that the platform and CRM use the same conversion definition and reporting period; document the deduplication rule for records appearing in more than one source.
- Check the consent banner and preference records, and verify that Google Consent Mode signals reflect the recorded choices where it is used.
- For a stated join key, calculate match rate as CRM-confirmed conversions matched to a platform record divided by eligible CRM-confirmed conversions. Record the numerator, denominator and period so the figure can be reproduced.
- Reconcile the result against CRM or booking records, then record gaps and any changes to settings in the method note.
For platform coverage, name the reporting source and capture its selected setting and conversion window. In Google Ads, record the attribution model selected for each conversion action, such as data-driven attribution or last click. In Google Analytics 4 (GA4), record whether reports use data-driven attribution or paid-and-organic last click. In Meta Ads Manager, record the selected attribution setting and its click-through and view-through windows. For TikTok Ads Manager, LinkedIn Campaign Manager, X Ads Manager, Pinterest Ads Manager and Snapchat Ads Manager, capture the corresponding setting and window shown in the account. Check paid YouTube campaign attribution in Google Ads and keep it distinct from YouTube Analytics channel reporting.
When to re-score
Re-score when consent rates shift, when a platform changes its measurement settings, or at contract renewal. Put the date in the contract. If an agency cannot explain a fall in match rates, treat it as an open question rather than a blip.
When platform figures and client records disagree
Platform interfaces report what each platform can see, using its own model and settings. Google Ads attribution reports show conversion paths and attribution-model reporting; Google documents the conversion and reporting settings behind credit assignment in its Google Ads best practices pages. Record the model selected for each conversion action, such as data-driven attribution or last click. Google Analytics 4 (GA4) has attribution reports, including data-driven attribution and paid-and-organic last-click reporting; record the model selected for the report. Meta Ads Manager reports conversions using its selected attribution setting and click-through and view-through windows. TikTok Ads Manager, LinkedIn Campaign Manager, X Ads Manager, Pinterest Ads Manager and Snapchat Ads Manager also have platform-specific campaign reporting settings to record. Check paid YouTube campaign attribution in Google Ads. Document each source's model and settings before comparing its figures with client records.
Client records usually come from a CRM or booking system. Deduplication rules differ, and closing periods follow the client's own calendar.
Set the band before the first invoice. For example, a team paying £2,500 a month might accept a 15% variance between platform-reported conversions and CRM-confirmed sales, escalating anything wider within five working days. At £25,000 a month, a band of 8% may be more realistic, because larger volumes make outliers easier to spot.
Social Partner's key metrics in England guide shows which figures clients ask for most often, so it is useful when you shape the reporting pack.
Consent and content rules that decide what you can measure
Consent rules decide whether you can link a person to an interaction. Content rules decide whether the post you are measuring stays visible for the whole campaign.
Consent also shifts mid-campaign. A dip in match rates after a banner or preference-centre change is a compliance story before it is a performance story.
Platforms remove content under their own policies, and the Online Safety Bill supporting documents on GOV.UK sit alongside the Online Safety Act regime that now applies to services hosting user content.
A takedown mid-flight changes impressions, reach and engagement for the period, so the report needs a note explaining it.
Common questions
Which attribution method should a client ask for first?
Ask first for a named last-touch method, such as a platform-reported last-click view, reconciled against CRM-confirmed sales. It is easier to audit; compare multi-touch or data-driven results only when the client can supply data to support them.
Why do social and search numbers never match?
Each platform claims credit for the same conversion on its own terms. Deduplication across platforms is a modelling choice rather than a fact, so combined totals stay estimates.
How often should attribution be reviewed?
At contract renewal as a minimum, and sooner if consent rates or platform settings change. Put the date in the contract so the review is not optional.
