The Importance of Data-Driven Strategies in Google Ads: Myths, Facts, and a Workable Decision Pipeline
Why most Google Ads accounts are optimising on distorted inputs - and how to rebuild the signal chain from conversion definition to bid decision.

Image via 123RF
In This Article
Signal Quality Beats Automation
Descriptive Is Not Data-Driven
Incrementality Is the Arbiter
Myth 1: 'More Conversion Data Is Always Better' - Fact: Miscategorised Conversion Actions Teach Smart Bidding to Chase the Wrong User
In a lot of Google Ads accounts, the conversion column is treated as a scoreboard. A form fill, a newsletter signup, a PDF download, a quote request, and a closed-won deal all land in the same table and are reviewed with the same reverence. They are not the same event, and Smart Bidding does not know the difference unless you tell it.
Every conversion action in Google Ads carries settings that determine whether it is used for bidding. If you mark a low-intent action such as 'brochure download' as a Primary conversion, you have just told Target CPA that this is a thing worth paying for. Bidding will optimise toward the people most likely to download a brochure - often researchers, students, or competitors - while ignoring the behaviours that correlate with pipeline.
Warning: Marking every tracked event as a Primary conversion is the single fastest way to degrade Smart Bidding. Secondary actions should stay Secondary until you have a deliberate reason to promote them.
The fix is not 'more data'. It is cleaner labels. In a B2B lead-gen account we worked with, roughly 40% of recorded conversions were content downloads that had never generated a sales conversation. Once those were demoted to Secondary and the primary signal was switched to qualified-lead imports, the account's reported conversion volume dropped sharply - but the cost per qualified lead became measurable for the first time, and Target CPA stopped chasing the discounted audience.
What 'Data-Driven' Actually Means Inside a Google Ads Account (Not in a Marketing Deck)
A data-driven Google Ads strategy is a repeatable loop: you define a business outcome, instrument the account so that outcome is visible, feed it to bidding, then test whether the decision produced the effect you claimed. Everything else - dashboards, reports, channel mix debates - sits downstream of that loop.
Three properties separate a genuinely data-driven account from one that merely reports a lot of numbers:
- Traceability: every bid, budget, and targeting decision can be traced back to a specific data point.
- Falsifiability: the account is capable of proving itself wrong - via a test, a holdout, or a controlled comparison.
- Decision latency: the gap between a signal appearing and a change being made is short enough that the signal is still relevant.
Most accounts score well on none of these, even when the agency reports weekly on CTR and impression share.
The Signal Chain: How a Raw Event Becomes a Bid Adjustment
Understanding why data quality matters is easier when you trace one event end to end. A user clicks an ad, fills a form, and later books a call with sales. That commercial fact has to survive five conversions of form before it can influence a bid:
- The website fires a tag (via Google Tag or Google Tag Manager).
- Google Ads records a conversion action with a counting method, attribution window, and value.
- Your CRM creates a qualified lead and its status is pushed back via offline conversion import or a Google Ads API integration.
- Smart Bidding receives the signal alongside device, location, time-of-day, audience, and query context.
- The bid strategy outputs an auction-time bid adjustment.
Any break in that chain - a missing value, a mismatched GCLID, a counting method set to 'one' when it should be 'every' - means the bid is being adjusted on a distorted picture. The bidding algorithm is not the failure point. The inputs are.
Myth 2: 'Smart Bidding Figures It Out on Its Own' - Fact: Target CPA and Target ROAS Are Downstream of Your Conversion Definitions
Google Ads Help is explicit that Smart Bidding uses your conversion data as the training signal: bid strategies such as Target CPA, Target ROAS, Maximise Conversions, and Maximise Conversion Value adjust bids in real time using auction-time signals plus the conversion history you supply. There is no second, secret dataset. Whatever you choose to call a conversion is the target.
This is why two accounts with identical budgets and landing pages can produce radically different outcomes: one is optimising to booked revenue, the other to form fills. Both are 'working'. Only one is aligned with the business.
The Four Data Sources That Feed Google Ads Decisions - and What Each One Gets Wrong
| Data source | What it is good at | Where it distorts |
|---|---|---|
| Google Ads conversion tracking | Fast, in-platform, bid-ready | Only sees what you tagged; misses qualified stages unless imported |
| GA4 + Google Tag / GTM | Cross-channel context, event-level detail | Consent-dependent; often undercounts versus Ads |
| CRM (HubSpot, Salesforce, Pipedrive) | Revenue truth, lead quality | Requires GCLID retention and a reliable import path |
| Offline conversion imports | Pushes qualified and closed deals back to bidding | Delayed; needs clean value rules and identity matching |
No single source is correct. A data-driven account reconciles the three and quantifies the gap, rather than declaring one 'the truth'.
How Conversion Action Settings Silently Change Bidding
Four settings inside a conversion action do most of the quiet damage:
- Primary vs Secondary: only Primary actions feed bid strategies.
- Count One vs Every: 'Every' suits eCommerce repeat purchases; 'One' suits lead gen where each lead is a single person.
- Attribution window: click-through windows (default 30 days) and view-through windows silently change which clicks get credited and therefore how bidding values them.
- Conversion value rules: when the value is set to '1' for all actions, Target ROAS has no meaningful revenue signal to optimise toward.
Consent Mode v2 is now part of the input chain. For declined-consent users, Google models conversions, so reported and modelled numbers diverge - a fact most US accounts are only beginning to see as state privacy laws expand.
Myth 3: 'Last-Click Is Dead, So Attribution Doesn't Matter' - Fact: Data-Driven Attribution Is Still a Model, Not a Measurement
Google removed first-click, linear, and time-decay attribution from Google Ads and GA4 in 2023, leaving data-driven attribution (DDA) as the default. That shift did not make attribution measurement - it made it a better-fitted model. DDA still distributes credit using observed paths; it does not tell you whether the click caused the conversion. The only way to answer that question is an experiment.
If you are auditing an agency or an in-house account, ask a simple question: what did your last incrementality test show? A strategic partner should have an answer. See how this fits an end-to-end performance setup before assuming platform metrics are the finished picture.
Myth 4: 'Incrementality Testing Is Only for Big Budgets' - Fact: A Single Geo Holdout Beats a Year of Attribution Debates
Incrementality testing has a reputation for being an enterprise-only activity. In practice, a geo holdout can be run on a modest budget with a spreadsheet and discipline. You split a campaign's target regions into test and control groups, pause or reduce spend in the control for a defined window (typically two to four weeks), then compare the change in new-customer or qualified-lead volume against the test group and a pre-period baseline.
What it does not require: Google's conversion lift product, a data science team, or a six-figure monthly spend. What it does require: enough baseline volume per geo to distinguish signal from noise, and the patience to let the test finish.
Tip: Pick one contested question per quarter - brand search incrementality, remarketing contribution, or Performance Max cannibalisation - and design a single test around it. One answered question compounds; five half-run tests do not.
The Four Data Strategies Mapped to Real Google Ads Decisions
'Data-driven' is a ladder, not a switch. Most accounts live on the first rung and describe it as maturity.
| Strategy | Question answered | Typical Google Ads artefact | Decision it feeds |
|---|---|---|---|
| Descriptive | What happened? | Spend, conversions, CPA, device splits, search terms report | Reporting and anomaly flags |
| Diagnostic | Why did it happen? | Search-term analysis, Auction Insights, Quality Score components, landing page diagnostics | Negative keyword lists, landing page fixes, ad relevance work |
| Predictive | What is likely next? | Budget pacing models, LTV-based value assignment, seasonality curves | Bid strategy selection, budget floors and ceilings |
| Prescriptive | What should we do? | Value rules, budget reallocation rules, test designs, creative rotation schedules | Executable change, then measurement |
The core argument of this article: most accounts stop at descriptive and call it data-driven. Every rung up requires cleaner inputs than the last, which is why the next section starts with conversion definitions and ends with experiments.
The Metrics That Actually Move Decisions - and the 3-3-3 Cadence Heuristic (With Its Limits)
The 3-3-3 rule is a marketing heuristic, not a Google-official framework. It suggests reviewing three metrics across three time horizons - three days, three weeks, three months - to discipline your attention:
- 3 days: pacing, anomalous spend, broken tracking, disapproved ads.
- 3 weeks: CPA trends, search-term quality, audience segment shifts, budget-vs-target variance.
- 3 months: qualified lead cost, incrementality results, value-rule accuracy, channel-level payback.
It breaks down in low-volume accounts (three days is noise), long B2B sales cycles (three weeks is not enough lag), and highly seasonal verticals (a three-month window spans unrelated demand regimes). Treat it as a scheduling default, not a measurement standard.
Common Mistakes That Make an Account Look Data-Driven While It Isn't
- Treating every conversion as equal - a form fill and a signed contract are not the same unit of work.
- Optimising to platform-reported conversions without reconciling against CRM or eCommerce order data.
- Setting Target ROAS on an account with too few conversions per month for the model to stabilise.
- Ignoring the gap between reported and modelled conversions after Consent Mode changes.
- Running 'A/B tests' on overlapping audiences, which contaminates both cells.
- Never running an incrementality test, so brand and remarketing spend are assumed rather than validated.
A Worked Example: B2B Lead Gen Rebuilt Around Offline Conversion Imports and Qualified-Lead Values
Consider a US B2B services company spending $25,000/month on Google Ads, generating 120 form fills at a reported cost per lead of about $208. Average contract value is roughly $14,000 with a 12% close rate, so realised revenue per lead is around $1,680 - but Smart Bidding has no visibility of that, because every conversion value is set to $1.
The rebuild: (1) demote brochure downloads and newsletter signups to Secondary; (2) import qualified-lead status from the CRM via Google Ads offline conversion imports, using GCLID from a hidden form field; (3) apply value rules so a marketing-qualified lead carries a modest value and a sales-qualified opportunity carries a much higher one; (4) switch the campaign to Maximise Conversion Value with a Target ROAS floor set from the realised revenue figure. Reported leads drop; pipeline-quality leads rise; the account optimises toward commercial outcomes rather than volume.
Offline import can be structured via the Google Ads API, scheduled Google Sheets, or a Click Conversions workflow. A simplified reconciliation formula for auditing the gap looks like this:
Google Ads conversions = Platform-tracked events
GA4 conversions = Consent-adjusted, cross-channel
CRM qualified leads = Ground truth (lagged)
Gap % = (Google Ads conversions - CRM qualified leads) / CRM qualified leads
An Eight-Step Data Pipeline Workflow You Can Run This Quarter
- Define the business outcome before touching any setting.
- Audit every conversion action: Primary/Secondary, count, window, value.
- Fix data supply - enhanced conversions, Consent Mode, offline imports, server-side where relevant.
- Reconcile Google Ads vs GA4 vs CRM and quantify the gap.
- Rebuild bidding inputs: value-based bidding where supported, correct tCPA/tROAS floors.
- Establish a search-term and audience review cadence that maps to the 3-3-3 horizons.
- Design one incrementality test per quarter.
- Report on business outcomes, not platform metrics.
For teams that want to compare notes against a partner's operating model, the Prebo Digital approach is a useful reference point, and the contact page outlines how audits are scoped.


