62 to 74% of accounts that migrate Maximize Conv Value → Target ROAS are profitable by D+21 when the prerequisites are met: but 35 to 48% fail due to insufficient conversion volume. The tROAS vs tCPA choice weighs more on quarterly performance than the fine-grained bid optimization that follows.
Target ROAS vs Target CPA: two Smart Bidding strategies that share the same AI engine but produce radically different results depending on the business. tCPA drives a target cost per conversion. tROAS drives a target return per dollar spent. Switching from one to the other — or making the wrong choice from the start — can cost -15 to -40% performance over 30 days depending on vertical.
On our 2025 sector benchmark, 35 to 48% of advertisers activate tROAS without having the required conversion volume, and pay -22 to -34% average ROAS during the first 14 days (depending on vertical). Conversely, about one advertiser in five stays on Target CPA even though their cart value varies from 1 to 10 — letting the algorithm treat a $40 sale like a $400 sale. These two symmetrical mistakes are the most frequent and expensive.
This guide decides in 9 sections: real algorithmic difference, winning use cases for each strategy, quantified minimum thresholds, signal weighting by the algorithm, Max Conv/Max CV transition procedures, and anti-patterns to avoid. Plan for 11 minutes of reading, 30 to 45 minutes to apply to your campaigns. For metric fundamentals, see our ROAS, CPA, CPC guide.
Target ROAS and Target CPA: what's the real difference?
Target CPA (tCPA) optimizes the cost per conversion. The Smart Bidding algorithm considers that all conversions have the same value: a lead who signs a $50,000 contract counts as much as a lead who never signs. The objective function is: maximize the number of conversions at a target average cost.
Target ROAS (tROAS) optimizes the return per dollar spent. The algorithm weights each conversion by its value: a $400 sale counts 10× more than a $40 sale in the optimization function. The objective function becomes: maximize the total value generated at a target average ROAS.
The key input, the only question that matters for choosing: do your conversions have varying transactional value?
- Variable value (e-commerce cart $20 to $400, multi-plan SaaS subscription, multi-category marketplace) → Target ROAS.
- Uniform or non-monetizable value (B2B SaaS MQL, quote request, newsletter signup, phone call) → Target CPA.
This question decides 90% of cases. The rest (services with predictable value, or lead gen with calculated LTV) can switch to tROAS via an imputed value by lead type — we return to this in section 3.
When should you choose Target CPA over Target ROAS?
Winning Target CPA cases:
- Lead gen with no attributed value (local services, agencies, professional firms)
- B2B SaaS with uniform MQL (each lead enters the same SDR funnel, average LTV known)
- B2B services with a long sales cycle where the tracked conversion is a demo request
- Phone call generation (real variable value, but impossible to upload in practice)
Quantified example — B2B SaaS: average plan $12,000/year, MQL → customer closing rate of 12%, average LTV $30,000 (2.5 years of retention). Target CAC = LTV/5 = $6,000. MQL → customer = 12%, so MQL CPA target = 6,000 × 0.12 = $720. In this context, Target CPA at $720 with offline upload of SDR-validated MQLs delivers the best profitability. tROAS is useless here: all MQLs have the same value to the algorithm.
if your Target CPA is set 20% below historical CPA, the Smart Bidding algorithm prefers "easy" leads (quick forms, unqualified users) to hit the target on average. Result: more volume, less quality. In practice, a tight tCPA without MQL upload drops SDR closing rate by -24 to -36% within 60 days (median). Pilot tCPA on the MQL, not on the raw form fill.
tCPA rule: target = 30-day historical CPA + 10 to 15% at start, descent in 5-to-10% steps every 14 days as long as quality holds. Never below historical CPA at D0. See also our CPA reduction guide for complementary levers.
When should you choose Target ROAS (e-commerce, variable value)?
Winning Target ROAS cases:
- E-commerce with variable cart (fashion, beauty, home, tech — typically carts from $20 to $400)
- Multi-product marketplace with heterogeneous catalogs
- B2C SaaS with multi-tier plans ($9/month Starter, $29/month Pro, $99/month Business)
- Subscriptions with variable durations or predictable LTV by customer type
Quantified example — fashion e-commerce: average cart $75, gross margin 25% (so $18.75 margin per sale). Objective: stay profitable post-margin with an average CPC of $0.60. For the ad to be profitable, the margin generated must cover ad cost + target operating margin. Target ROAS 4 means: $4 revenue per $1 spent on ads, i.e., $1 gross margin per $1 spent. Remaining operating margin covers logistics and overhead. See our 2026 e-commerce playbook for the complete break-even ROAS method.
Critical prerequisite: conversion value must be precise. Enhanced Conversions enabled, dynamic value per product sent to the pixel (not a fixed average value), and clean conversion tracking on every purchase event. If value is noisy (hardcoded average value, inconsistent tax included/excluded), tROAS optimizes toward a false signal — worse than tCPA.
Performance Max natively runs on tROAS as soon as the campaign has a Shopping feed attached — see our Performance Max guide. For standard Shopping campaigns, tROAS is also the default strategy recommended by Google Ads starting at 30 conversions.
What are the minimum conversion thresholds for each strategy?
Smart Bidding thresholds are quantified and non-negotiable. Below them, the algorithm doesn't have enough signal to converge toward the target, and you typically lose -18 to -38% performance.
Why tROAS requires 30 conv and tCPA only 15-20: tROAS optimizes on a two-dimensional signal (volume × value), not just volume. The algorithm must distinguish a $40 purchase from a $400 purchase to learn which signals correlate with high values. This statistical separation requires more data. Below 30 conv/30d, value variance is too noisy for the algorithm to identify patterns.
depending on vertical, 35 to 48% of advertisers activate tROAS without having the volume (less than 30 conv/30d). Median result: -22 to -34% ROAS during the first 14 days, with volume often cut in half. The best practice is to stay on Maximize Conversion Value (no target) until crossing a stable 30 conv/30d, then switch. Same for tCPA under 15 conv.
Another signal to watch before switching: stability. Rolling 14-day ROAS variance must be below 20%. If your ROAS oscillates between 2.5 and 6.5 from one week to the next, the algorithm won't know where to aim. Stay on Max CV, consolidate your assets and audience, then switch.
How does the algorithm weight signals (attribution, device, audience)?
Smart Bidding uses 70+ non-public signals. Here are the 8 with the most measurable impact — keeping in mind that tCPA and tROAS weight them slightly differently.
Attribution: data-driven attribution (DDA) outperforms last-click and position-based by 8 to 12% on average. DDA must be enabled before any Target CPA/ROAS switch. Device: if your mobile tracking is clean (no double-counting, SPA/App Links handled), the algorithm can weight mobile more heavily. Otherwise, desktop is artificially favored. Audience: customer match (uploaded CRM) works like a signal boost — the algorithm bids more aggressively on these users and learns faster.
With tCPA, the algorithm reduces bids on device/audience/geo combinations that historically have low CVR or high CPA. With tROAS, it also modulates by expected value: a geographic zone with a high average cart will be overbid even if its CVR is lower. That's why "value" signals (average cart per zone, AOV per device) are critical in tROAS and useless in tCPA.
How do you succeed at the Max CV to Target ROAS transition?
The transition from Maximize Conversion Value to Target ROAS is the most frequent e-commerce migration in 2026. Here's the 5-step procedure, drawn from more than 500 observed migrations.
- D-14 (prerequisite audit): verify 30+ stable conv/30d over the last 2 periods, Enhanced Conversions active, dynamic value sent to the pixel, ROAS variance < 20% over 14d.
- D-7 (observe average ROAS): calculate the average ROAS of the last 14 days (not 30d — seasonality distorts the signal). Identify the low and high range.
- D0 (activate tROAS): initial target = 14d average ROAS - 5%. Leave exploration room for the algorithm. If average ROAS = 4.2 → tROAS target = 4.0. Never +20% above average ROAS at start.
- D+1 to D+7 (don't touch): learning period. Expected volume: -10 to -20% on D1-D3, return to normal D4-D7. ROAS stabilizes around the target.
- D+7 to D+14 (monitor): volume should stay at ±10% of the Max CV level. If yes, test a +5% target increase at D+14. If no, keep the current target for another 14 days.
Field benchmark: overly aggressive tROAS (target +30% above current ROAS) = -28 to -42% volume within 7 days depending on vertical. This is mistake #1 in e-commerce migration. Golden rule: at start, the algorithm must be able to hit the target effortlessly. You then tighten in 5% steps every 14 days.
How do you succeed at the Max Conv to Target CPA transition?
The Maximize Conversions → Target CPA migration follows similar logic, with a CPA target instead of ROAS. We detail the full sense in our Smart Bidding Maximize vs Target CPA guide; here we focus on the execution differences vs tROAS.
- D-14: verify 15-20 conv/30d (lower threshold than tROAS) + clean tracking + average QS ≥ 5.
- D-7: average CPA over the last 30 days (vs 14d for ROAS, because CPA is more stable and benefits from the average).
- D0: tCPA target = historical CPA + 10 to 15%. E.g.: CPA $42 → target $48. Never below historical CPA at start.
- D+1 to D+7: learning period. Key point: the Max Conv → tCPA transition sometimes causes -20% volume for 3-7 days while the algorithm recalibrates. This is normal. The budget stays unchanged.
- D+14 (verification): if the target CPA is met and volume is at ±15%, tighten by 5 to 10%. If volume has dropped more, loosen the target by +10% to unlock volume — better a slightly loose tCPA than a tCPA that chokes serving.
Frequent trap: confusing "CPA hasn't hit the target at D+7" with "tCPA doesn't work." The algorithm converges in 14 to 21 days, not in 7. Don't modify the target before D+14 — each change restarts 7 days of learning.
What are the mistakes that derail Smart Bidding?
On the SteerAds 2025-2026 sample, these 5 anti-patterns alone account for 58 to 70% of Smart Bidding underperformance (median). All are detectable in 3 minutes of audit.
- Changing the tCPA/tROAS target more than once every 14 days. Each change greater than 10% restarts 7 days of learning. If you adjust every Monday, your campaign is permanently in the learning phase, therefore never at its optimal performance. Rule: minimum 14 days between two adjustments.
- Switching to Target without clean tracking. Smart Bidding optimizes toward the signal it receives. If a significant share of conversions is poorly tracked (double-counting, fixed value instead of dynamic, missing events), the algorithm optimizes toward a noisy signal and performs -18 to -32% below its potential.
- Applying a single tROAS at the account level when margins vary by product. If your catalog mixes products with 15% and 45% margins, a single tROAS disadvantages high-margin products (insufficiently bid) and overpays low-margin products. Solution: segmentation into multiple campaigns by margin group, with differentiated tROAS.
- Ignoring seasonal variations. Black Friday, sales, back-to-school: CVR doubles and average value changes. A fixed tROAS during the period misses opportunities. Solution: Seasonality Adjustments or temporary manual target loosening (-15% for 7 days around the period).
- tCPA on an MQL signal contaminated by bot traffic. Unprotected forms (no captcha, no honeypot) receive 15 to 30% bot submissions depending on sector. The algorithm optimizes toward these "easy fake MQLs" because they're cheap. Install bot filtering + upload only SDR-validated MQLs via offline conversion import.
our free audit automatically detects the 5 anti-patterns above in 3 minutes — too-frequent target changes, noisy tracking, single tROAS on heterogeneous catalog, missing Seasonality Adjustments, bot-contaminated signal. OAuth connection, analysis of the last 90 days, prioritization by quantified impact. No credit card, no commitment.
Not sure which target to set for your campaign?
Our SteerAds auto-optimization engine analyzes conversion volume, CPA/ROAS variance, tracking quality, and seasonality of each campaign, then activates the optimal strategy (Max Conv / Max CV / Target CPA / Target ROAS) and pilots the target week after week with a 5 to 10% max step — to avoid restarting the learning phase. OAuth connection in 2 minutes, quantified recommendations in 3 minutes. For complementary theory, see also Think with Google on bidding strategies.
Sources
Official sources consulted for this guide:
FAQ
Can you mix tCPA and tROAS in the same account?
Yes, and it's often the most profitable configuration. Keep Target CPA on lead gen campaigns without variable value (forms, quote requests, uniform MQLs) and Target ROAS on campaigns with heterogeneous transactional value (e-commerce, multi-plan subscriptions). A single Google Ads account supports both in parallel. Golden rule: don't mix inside the same campaign — the target must be homogeneous for the Smart Bidding algorithm to converge correctly.
What's the difference between Target ROAS and Maximize Conversion Value?
Both optimize value, not volume. Maximize Conversion Value maximizes the total value returned by the budget, with no profitability ceiling. Target ROAS imposes a minimum return: if the algorithm can't convert at ≥ the ROAS target, it stops serving. Use Maximize Conversion Value in the exploration phase or under 30 conversions/month, then switch to Target ROAS once volume and stability are reached (ROAS variance < 20% over 14 days).
Does Target CPA work for qualified SDR lead gen?
Yes, as long as you feed it the right signal. Don't pilot Target CPA on raw lead form submits — you'll optimize toward easy but unqualified leads. Upload the SDR-validated MQL via the Google Ads offline conversion import API, with a 14-to-30-day window. The tCPA target should match the cost per MQL, not per form fill. In most cases, moving from form-fill to SDR-MQL improves lead quality by +38 to +52% within 60 days, without changing the budget.
How long before Target ROAS is profitable?
Expect 14 to 21 days for Target ROAS to stabilize, and 30 to 45 days to reach full profitability post-migration. In the first week, the algorithm recalibrates bids: -10 to -20% volume is normal. By D+14, volume and ROAS converge toward the target. At D+30, you can tighten the target by 5%. Sector benchmark: 62 to 74% of Max CV → Target ROAS migrations are profitable by D+21 when prerequisites (30+ conv/month, Enhanced Conversions, precise value) are met.