Most Target ROAS campaigns optimise against first-purchase value. Someone spends £80 on their first order, you send £80 to Google as the conversion value, Smart Bidding optimises for more £80 customers.
But not all £80 customers are equal. One never buys again. Another spends £400 over the next year. Optimising against first-purchase value treats them identically. It shouldn't.
Predicted LTV (pLTV) bidding sends the expected lifetime value of each customer to Google instead of just the first purchase. It's the next step after profit-based bidding. The accounts that do this properly outspend their competitors profitably because Smart Bidding is finally optimising for long-term customer value, not transactional revenue.
Here's how to build a pLTV model that's accurate enough to bid against, without needing a data science team.
The TL;DR: First-purchase value tells Smart Bidding what someone spent today. pLTV tells it what they'll spend over the next 12-24 months. Different customers have radically different lifetime values, and Smart Bidding will bid much more aggressively (and profitably) for high-LTV segments if you give it the right signal. Build a cohort-based pLTV model, send it as the conversion value, and watch Smart Bidding redistribute spend toward customers worth keeping.
Why first-purchase value isn't enough
Imagine two e-commerce customers:
- Customer A buys £80 of running shoes. Never buys again. Total revenue: £80.
- Customer B buys £80 of running shoes. Comes back monthly for socks, gels, accessories. Total 12-month revenue: £480.
To Smart Bidding optimising on first-purchase value, both customers are worth £80. To your business, Customer B is six times more valuable.
If Customer B's first purchase came from a slightly more expensive click, Smart Bidding would have lost the auction in favour of acquiring Customer A. Smart Bidding is making bidding decisions on the wrong number.
pLTV fixes this by sending Google the expected long-term value of each customer at the moment of conversion.
The four levels of pLTV sophistication
Pick your starting point based on data maturity.
Level 1: Flat multiplier
Multiply first-purchase value by an account-wide LTV multiplier (e.g., 2.2x if average customer LTV is 220% of first-purchase value).
Crude. Better than nothing. Easy to implement. Use this if you're getting started.
Level 2: Segment-based pLTV
Calculate LTV multipliers by segment — by product category, by acquisition channel, by first-purchase value tier.
A customer whose first purchase was £200 in premium skincare has a different expected LTV than one whose first purchase was £15 in clearance lip balm. Segment-level multipliers capture this.
Level 3: Cohort-based pLTV model
Build a regression or machine learning model that predicts each customer's LTV from features observable at the moment of first conversion — products purchased, acquisition source, geography, device, time of day.
Requires a data analyst and a few weeks of work. Substantially more accurate than segment-based.
Level 4: Real-time scoring
A live ML pipeline that scores each conversion against the model and sends pLTV to Google Ads via API or dataLayer in real-time.
Requires data engineering. The gold standard for high-spend accounts.
For most accounts under £100k/month in ad spend, Level 2 or Level 3 is the sweet spot. The cost of Level 4 rarely justifies the marginal accuracy gain.
Building a cohort-based pLTV model
Here's the practical Level 2-3 approach that works for most accounts.
Step 1: Pull historical customer data
You need at least 12 months (ideally 24) of customer-level data:
- Customer ID.
- First purchase date.
- First purchase value.
- First purchase product category.
- Acquisition channel (Google Ads / Meta / organic / direct).
- Cumulative revenue over 12 months after first purchase.
Pull this from your CRM, e-commerce platform, or warehouse.
Step 2: Segment by features available at first purchase
Group customers by features you'd know at the moment of conversion:
- First-purchase value tier (e.g., £0-50, £50-150, £150-300, £300+).
- First-purchase product category.
- Acquisition channel.
- Device type at first purchase.
- Country / region.
For each segment, calculate:
- LTV multiplier = avg 12-month cumulative revenue / avg first-purchase value.
Step 3: Build a lookup table
The output is a table like:
| First-purchase tier | Channel | Product category | LTV multiplier | |---|---|---|---| | £0-50 | Google Ads | Skincare | 3.4x | | £0-50 | Google Ads | Haircare | 1.8x | | £150-300 | Google Ads | Skincare | 4.1x | | £150-300 | Direct | Skincare | 2.6x |
Each customer gets a pLTV at first purchase = first-purchase value × matching multiplier.
Step 4: Apply margin (if you're already on profit-based bidding)
If you've already switched to profit-based bidding (from the previous post in this series), apply margin too:
conversion_value = first_purchase_value × ltv_multiplier × margin_factor
This is now your "predicted lifetime profit" — the right number to send Google.
How to send pLTV to Google Ads
Three implementation paths.
Option A: At the moment of conversion (dataLayer)
If you can calculate pLTV in real-time at checkout (lookup table in your e-commerce platform):
``javascript dataLayer.push({ event: 'purchase', ecommerce: { transaction_id: '12345', value: 89.50, // first-purchase revenue predicted_ltv: 312.50, // calculated pLTV margin_adjusted_ltv: 110.30, // pLTV × margin factor currency: 'GBP' } }); ``
In GTM, your Google Ads conversion tag uses margin_adjusted_ltv as the conversion value.
Option B: Conversion value rules
For coarser pLTV adjustments (channel-level or segment-level multipliers without full per-customer scoring):
- Tools > Conversions > Value rules in Google Ads.
- Create rules that multiply conversion value by your LTV multiplier for each segment.
Less accurate than per-customer pLTV. Works without dev effort.
Option C: Offline conversion adjustments
If pLTV can only be calculated post-purchase (e.g., on day 30 or 60 after first order):
- Fire the original conversion with first-purchase value at checkout.
- Run a nightly job that calculates pLTV for each customer.
- Upload conversion adjustments via the Google Ads API to revise the conversion value.
Smart Bidding will retroactively use the revised value.
Re-baseline your tROAS target
When you switch from first-purchase to pLTV, conversion values get bigger. If your tROAS was 400% on first-purchase value, and your average pLTV multiplier is 2.5x, your equivalent tROAS on pLTV is 1,000%.
Re-baseline the target so Smart Bidding doesn't go crazy with bids:
- Old tROAS (first-purchase): 400%.
- New tROAS (pLTV): 400% × 2.5 = 1,000%.
Run for 4-6 weeks and let Smart Bidding learn the new target. Then iterate.
Use case: a subscription beauty box
A composite based on patterns I've seen.
A monthly subscription beauty box was running Google Ads with first-purchase value as the conversion. tROAS at 200% on the trial price of £15. Smart Bidding was capping CPC at uncompetitive levels because the perceived value per conversion was so low.
We built a Level 2 pLTV model. Key findings:
- Average customer retained for 6.4 months.
- Average lifetime revenue per customer: £96 (£15 trial + ~£81 in renewals).
- LTV multiplier: 6.4x first-purchase value.
- High-LTV customer segments (specific acquisition sources, specific demographics): 9-12x multiplier.
- Low-LTV segments (discount-driven, certain ad creative): 2-3x multiplier.
We applied segment-level multipliers via dataLayer, switched tROAS from 200% (first-purchase) to 1,200% (pLTV).
Results over 90 days:
- Smart Bidding raised bids substantially on high-LTV segments. CPC went up 65%. Volume from those segments rose 4x.
- Smart Bidding lowered bids on low-LTV segments. Spend in those segments dropped 40%.
- Total trial sign-ups dropped slightly (10%). Long-term retained customers rose 28%.
- Real 12-month cohort ROAS improved by 47%.
Same Google Ads account. Same budget. Different optimisation target. Materially different business outcome.
Common mistakes
- Building pLTV from too little data. Less than 12 months of cohort data isn't enough to estimate retention curves accurately. Use historical data or industry benchmarks if you're early-stage.
- Forgetting to re-baseline tROAS. Without the new target, Smart Bidding either bids too aggressively or doesn't bid at all.
- Sending pLTV for every conversion type. Use pLTV for first-purchase only. Subsequent purchases shouldn't get pLTV applied or you'll double-count.
- Not segmenting the model. Account-wide multipliers are 10x better than nothing, but segment-level captures most of the available accuracy.
- Confusing pLTV with predicted profit. Predicted revenue over lifetime is one thing. Predicted profit is another. Combine pLTV with margin factors for the cleanest signal.
- Updating the model too often. Build it. Run it. Re-validate every 6 months. Don't change multipliers monthly or Smart Bidding can't stabilise.
Bottom line
First-purchase value is the conversion signal most accounts give Google Ads. It's also incomplete. Customers have lifetime value, and the best customers have dramatically more than the average.
- Build a Level 2 (segment-based) or Level 3 (cohort-based) pLTV model from your historical customer data.
- Apply margin factors if you're already doing profit-based bidding.
- Send the predicted lifetime profit to Google Ads at the moment of first conversion.
- Re-baseline your tROAS target to match the new value scale.
- Let Smart Bidding learn for 4-6 weeks before judging.
pLTV bidding is one of the few performance marketing techniques that compound. The accounts that get this right pay more for the right customers and less for the wrong ones, every day, automatically. Their competitors, still optimising on first-purchase value, can't figure out why they keep getting outbid on the most valuable segments.
The data was always there. Sending it correctly is the unlock.
Sources and further reading: