PERFORMANCE ENGINEERING
Performance Marketing
An ROI-focused performance program built on server-side conversion architecture, automated bid layers and full-funnel attribution across Google Ads, Meta, TikTok and LinkedIn.
Falling CPA is not a tactic; it is the outcome of a correctly engineered data architecture.
The post iOS 14 world rewrote the rules of performance marketing: third-party cookie erosion, consent-mode mandates, attribution darkness and platform algorithms becoming 'black boxes' made the classic agency reflex of 'just reallocate the budget' meaningless. The Roibase performance program builds the infrastructure first — server-side Conversion API, consent-aware GTM, feed engineering, incrementality testing — then launches the campaign. This way the right signal reaches the algorithm, the bid layer optimizes real lift, and the incremental contribution of every dollar spent is measured. We are not talking about writing better ads; we are talking about engineering a better system.
METHODOLOGY
Our operating framework
We run performance marketing as a six-layer engineering system — each layer's output feeds the next, and the loop repeats every sprint.
DISCOVER
Account & data audit
Google Ads, Meta, TikTok, LinkedIn, GA4, Merchant Center, CDP and CRM connections are scanned; signal losses, attribution gaps and feed pathologies are mapped.
ARCHITECT
Signal & data architecture
sGTM + Conversion API + consent mode + BigQuery pipeline + feed hierarchy + attribution stack are built as a single technical document. We clean the data going to the platform.
LAUNCH
Campaign & creative pipeline
A channel x funnel x audience x creative matrix is created; portfolio bidding, Advantage+ / Pmax architecture and the weekly creative fitness loop go live.
MEASURE
Attribution & incrementality
GA4 DDA + Markov + MMM + geo incrementality tests meet on one dashboard; channel lift, incremental conversions and new customer share are measured.
OPTIMIZE
Weekly sprint iteration
Creative rotation, bid/budget shifts, negative keyword hygiene, audience refresh and feed updates run on a weekly rhythm; hypothesis + result of each iteration is documented.
SCALE
Scale & new channel launch
On mature accounts, new channels (TikTok Spark, LinkedIn ABM, Pinterest, Reddit, retail media) or new markets are launched; scaling thresholds are raised in controlled steps.
— COMPARISON
The difference between performance-marketing approaches
Same media budget, very different outcomes. There are three typical models in the market; we recommend the engineering-based program.
| Criterion | In-house performance manager | Classic media agency | Roibase engineering-based |
|---|---|---|---|
| Server-side Conversion API | Planned, not implemented | On request, extra cost | Standard — full setup in the first 2 weeks |
| Consent Mode v2 | Usually missing | Handed off to developers | Built and QA'd by our own team |
| Attribution model | Platform panel | GA4 DDA is considered enough | GA4 DDA + Markov + MMM + incrementality |
| Creative testing | Ad-hoc | 2-4 variants per month | 8-20 weekly variants + fitness score |
| Feed engineering | Standard feed | Category-based segmentation | Hierarchical + supplemental + custom_label structure |
| Incrementality testing | Not done | 1-2 times per year | Quarterly + before major changes |
| Reporting | Platform screenshot | Monthly PDF | Alerted Looker dashboard + weekly sprint review |
| Optimization rhythm | Daily manual | Weekly small tweaks | Script automation + weekly hypothesis-driven sprint |
PROOF
Outcomes, measured
First 90 days, 2024 average across e-commerce accounts.
2024 portfolio average for accounts with AOV > €50.
Increase in attributable conversions after sGTM + CAPI.
Net new customer ratio after LCV bidding + incrementality testing.
12-month cohort analysis, average across 18 clients.
Audit pass rate across GDPR + KVKK + DMA.
WHAT WE DO
Engagement scope
Every offering is an outcome-based work package. Roibase blends strategy and execution inside a single team — no hand-offs.
Server-side Conversion API & sGTM
Server-side transmission for Google Ads Enhanced Conversions for Web + Leads, Meta CAPI, TikTok Events API, LinkedIn Conversion API and Pinterest Conversion API. We deploy sGTM via Stape or our own cloud infrastructure. We recover 30-50% of the conversions lost after iOS 14+.
Consent Mode v2 & privacy-first signal architecture
We preserve the full signal flow within a GDPR, KVKK and DMA-compliant consent framework: advanced consent mode, consent-aware sGTM and modeled conversions integration. 40-50% conversion reconstruction even for rejected cookies.
Bid automation & portfolio strategy
A custom script layer for Target ROAS, Maximize Conversion Value, Advantage+ Shopping, ABO/CBO strategies; thresholds, negative keyword hygiene, brand defense and day/hour/device modifiers are managed algorithmically. Automatic guardrails instead of manual intervention.
Creative testing pipeline
Weekly production of 8-20 variants + fitness scoring; separating hook, thumbnail, USP, promise, CTA and duration gives a clear answer to 'which creative element is producing performance'. The next wave is ready before creative fatigue hits.
Full-funnel attribution
A triangle of GA4 DDA + our own Markov chain model + MMM (Meridian / Robyn based). Instead of the last-click story, the incremental contribution of each channel; MMM for budget allocation, MTA for channel comparison, platform attribution for daily optimization.
Feed & Performance Max architecture
Merchant Center feed hierarchy, category/brand/price segmentation, supplemental feed, custom_label structure, Pmax asset groups and audience signals — we capture optimization in the feed, not in the campaign. Typical 20-30% ROAS lift after feed cleanup.
Incrementality testing
Geo-experiment (Google Ads Geo Test, Meta Conversion Lift), hold-out audience tests and matched-market studies. A clear answer to 'how much of Pmax is brand traffic versus real new demand' — budget waste surfaces.
Audience & first-party data strategy
We build lookalike and value-based bidding fed by first-party data through Customer Match, Advantage+ Audience, LiveRamp and CDP integrations. In a cookieless era, continuously piping high-value audiences to the algorithm is critical.
LCV (Lifetime Value) bidding
A bid strategy that optimizes not just the first purchase but 12-24 month customer value; works integrated with Shopify, WooCommerce, Magento, Salesforce and custom CRMs. We target LTV/CAC ratio instead of a standalone CAC.
Reporting, alerts & executive dashboard
A KPI tree with daily alerts on Looker Studio + BigQuery, weekly sprint review, monthly executive summary. At a glance for the C-level: spend, CPA, ROAS, new customer share and incremental lift.
— OUTCOMES
What the performance-marketing program delivers for you
Six concrete gains our typical client sees within 90-180 days. A shift from a short-term ROAS game to a long-term growth system.
Predictable CAC & ROAS
We commit to a CAC ceiling and ROAS floor on a quarterly basis; a performance model that speaks the same language as your finance team.
Closing the signal gap
Server-side infrastructure brings most lost conversions back; because the algorithm is fed correct data, bidding works more efficiently.
Real incrementality visibility
Which campaign is 'brand cannibalization' and which is generating real new demand becomes clear quarter by quarter; budget allocation changes fundamentally.
LTV-focused customer base
The target shifts from first purchase to 12-24 month customer value; low-quality customer flow stops and repeat-purchase rates rise.
Creative fatigue control
The weekly test rhythm keeps new variants ready before the fatigue wave arrives; campaign performance does not plateau.
C-level reporting discipline
Looker dashboard + weekly sprint review + monthly executive summary; CFO, CEO and CMO debate the same number, the 'black box' risk disappears.
DELIVERABLES
What we ship inside the service scope
Concrete artifacts rather than verbal reports. By the end of the first 30-60 days, you have these 12 deliverables in hand.
Account audit report
A 40-60 item audit output for Google/Meta/TikTok/LinkedIn, ranked by risk.
sGTM + Conversion API setup
Server-side transmission, event schemas, QA documentation, validation outputs.
Consent Mode v2 configuration
CMP integration, modeled conversions, GDPR/KVKK/DMA compliance report.
Feed hierarchy playbook
Merchant Center structure, supplemental feed logic, custom_label documentation.
Portfolio bidding strategy
Bid strategy map and thresholds across a channel x funnel x audience matrix.
Creative test calendar
A weekly 8-20 variant test plan, with a hypothesis x expected lift x sample-size table.
Attribution stack document
The integration architecture of GA4 DDA + Markov + MMM, specifying which model serves which decision.
Incrementality test plan
Geo-experiment or hold-out test design, sample-size power calculation, findings report.
Looker Studio & BigQuery dashboard
Spend, CPA, ROAS, new customer share, incremental lift on one screen + alert configuration.
Weekly sprint review document
Hypothesis, result, deviation, next step — documented in a Notion template.
Monthly executive summary
A one-pager for the C-level: revenue, spend, CAC, LTV/CAC, new hypotheses, quarter forecast.
Annual media plan & forecast
Scenario-based budget plan, channel-share model, month-by-month forecast — ready for the finance team.
— SCOPE
What's in the performance program, and what isn't
What we cover needs to be as clear as what we don't — expectation management is the most critical part of a performance-agency relationship.
What this service covers
- Campaign management across Google Ads, Meta, TikTok, LinkedIn, Pinterest
- sGTM + Conversion API + Events API setup and maintenance
- Consent Mode v2 + modeled conversions integration
- Merchant Center feed hierarchy + Pmax asset group architecture
- Portfolio bidding + custom script automation layer
- Weekly creative testing pipeline (brief + quality control)
- GA4 DDA + Markov + MMM attribution stack
- Quarterly incrementality tests
- LCV bidding + first-party data integration
- Alerted Looker dashboard + weekly sprint review + monthly executive
- New channel and new market launch playbook
- Response protocol for algorithm changes (Pmax, Advantage+)
What this service does not cover
- Creative production (brief + QC are included; production via in-house studio is a separate package)
- SEO + content marketing (separate service)
- Influencer / affiliate campaigns (separate scope)
- Email / CRM automation (separate service)
- Sales team training / CRM rollout
- Hosting / infrastructure management
- PR / press release distribution
- Guaranteed sales promises (no serious performance agency makes them)
HOW WE WORK
Three weeks of setup, then a weekly iteration rhythm.
Week 1 — Account & data audit
Google/Meta/TikTok/LinkedIn/GA4/Merchant Center connections scanned, a 40-60 item audit output ready; baseline metrics are frozen.
Week 2 — Signal architecture setup
sGTM + Conversion API + Consent Mode v2 goes live; event schemas pass QA, modeled conversions are enabled.
Week 3 — Feed & campaign architecture
Merchant Center feed hierarchy is built; Pmax and Advantage+ asset groups go live with audience signals.
Week 4-5 — Portfolio bidding & creative wave 1
Target ROAS / Maximize Conversion Value strategies are wrapped in the script layer; the first creative test wave (8-20 variants) starts.
Month 2 — Attribution stack & incrementality test
GA4 DDA + Markov + MMM integration is completed; the first geo-experiment or hold-out test runs, channel lift is measured.
Month 3 — LCV bidding & first-party audience
Customer Match, CDP integration, value-based bidding; LTV/CAC ratio is added to the measurement board and the new-customer focus is activated.
Month 4 — New channel & new market launch
On a mature account, TikTok Spark, LinkedIn ABM, Pinterest or a new country launch; controlled scale.
Month 5+ — Sustaining & scaling
Weekly sprint rhythm, monthly executive summary, quarterly incrementality test, annual media plan refresh — the program becomes a permanent capability.
— ECOSYSTEM
The platforms & tools we use
Best-in-class tools at every layer. We integrate with the client's existing stack; we do not force new licenses if they are not needed.
CAMPAIGN PLATFORMS
SIGNAL & DATA INFRASTRUCTURE
ATTRIBUTION & INCREMENTALITY
REPORTING & WORKFLOW
QUESTIONS
Frequently asked
— GLOSSARY
Performance marketing glossary
In performance marketing, language clarity drives decision quality. Short definitions for the 12 concepts we use most.
- CPA (Cost Per Acquisition)
- The media cost of winning one conversion. 'Blended CPA' covers all channels combined; 'paid CPA' only paid media. Mixing them leads to wrong decisions.
- ROAS (Return On Ad Spend)
- The ratio of revenue produced by media spend to that spend. ROAS alone is not a performance measure; unless margin and LTV are integrated, high ROAS can mean low profitability.
- Server-side tracking (sGTM)
- Event tracking performed on the server rather than the browser. It preserves signal integrity against iOS 14, ad-blockers and cookie erosion; the cornerstone of modern performance infrastructure.
- Conversion API (CAPI)
- A conversion API that platforms like Meta, TikTok and LinkedIn receive directly from the server. Preferred over browser-based pixels; more reliable, more complete signal.
- Consent Mode v2
- Google's consent model introduced for GDPR/DMA compliance. It preserves aggregated signal flow even with rejected cookies; correctly configured, modeled conversions fill 40-50% of the missing data.
- DDA (Data Driven Attribution)
- GA4's machine-learning-based attribution model. It assigns marginal contribution to each touchpoint; far more realistic than classic last-click.
- MMM (Marketing Mix Model)
- Produces month-by-month budget allocation via aggregated regression modeling. In a cookieless world, the most important attribution approach for the future; Meridian (Google) and Robyn (Meta) are open-source solutions.
- Incrementality
- The measure of whether a campaign actually creates new demand. Measured via geo-experiments or hold-out tests; critical for detecting 'brand cannibalization'.
- Performance Max (Pmax)
- Google Ads' product that combines all inventory into one campaign. Optimization points are feed + asset + audience signal + incrementality testing; 'black box' control runs on these 4 levers.
- LCV / LTV bidding
- A bid strategy that targets long-term customer value rather than first purchase. Customer-value signals from Shopify, CRM or CDP are piped to the algorithm; LTV/CAC is optimized instead of CAC.
- New Customer Share
- The share of total conversions that come from new customers. Critical during expansion; a low share indicates you are mostly retargeting existing customers.
- Creative fatigue
- The decay in CTR, CVR and engagement caused by repeatedly serving the same creative. Managed with frequency caps and creative rotation; without a performance model, frequency above 4-7 is the standard alarm threshold and should be reviewed weekly on Meta accounts.
- Demand Gen
- Google's visual-first performance format unifying YouTube, Discover and Gmail inventory in one campaign. Successor to Discovery Ads; uses social-style creatives to feed upper-mid funnel and measures attribution via GA4.
- App Campaigns
- Google App Campaigns (UAC) is a single campaign type that automates Search, Display, YouTube and Play inventory for app installs and in-app actions. ML mixes creative assets; runs against tCPA or tROAS targets.
- Lookalike Audience
- A prospecting audience that ML-derives behavioural twins of an existing customer seed list (CRM, conversion pixel, app events). Tunable on Meta within a 1-10% similarity band; quality scales with seed signal cleanliness.
- Custom Audience
- An audience built by hashing first-party lists (email, phone, mobile ad ID) uploaded to an ad platform. Meta Custom Audience, Google Customer Match — used for retargeting, exclusions and as a LAL seed.
- Broad Match
- The widest match type in Google Ads — captures any query that fits the keyword's intent. Paired with Smart Bidding it opens long-tail opportunity; run in isolation it scatters spend — negatives and audience signals are critical.
- Smart Bidding
- Google Ads umbrella for ML bidding strategies that bid per-auction using real-time signals (device, location, user signal, time): tCPA, tROAS, Maximize Conversions, Maximize Conversion Value. Quality scales with the conversion signal.
- Target CPA (tCPA)
- Smart Bidding strategy that drives conversions at or near a target average cost per acquisition. Requires sufficient historical conversion data and a realistic target; volatile on low-volume accounts.
- Target ROAS (tROAS)
- Smart Bidding strategy that bids to hit a target return-on-ad-spend. Requires conversion value tracking and accurate margin/value signal; performance improves when enriched with offline conversion + value upload.
- Value-based Bidding
- A bidding approach that assigns each conversion a real margin or LTV-proxy value rather than counting them equally. The prerequisite that makes tROAS meaningful — fed by value upload, dynamic value or CRM-bound value rules.
- Pixel
- A client-side JS snippet placed by an ad platform (Meta, TikTok, LinkedIn, etc.) to measure on-site interaction and conversion. Insufficient on its own due to ITP and cookie limits — supplemented with server-side events (CAPI, sGTM).
— DECISION TREE
Is the performance program right for you now?
Four short questions. In 30 seconds we tell you the right starting point — full program, signal-infrastructure-first sprint, a foundation before small scale, or strategic discovery.
01 / 04
Is your monthly media budget 150,000 TL (~4,500 USD) or more?
Below that, the agency cost consumes the net impact; we recommend an in-house or freelance setup first.
— LET'S BEGIN
How much of your ad budget is actually producing results?
In a 72-hour account scan we surface lost signal, misattributed conversions, feed pathologies and 'dark budget'. Free and non-binding.