LTV Data Arrives Too Late for Smart Bidding

The Timing Problem

Smart Bidding learns from conversion signals. When a click converts, the algorithm receives a signal: "this click was worth £X." It uses this feedback to adjust future bids. The tighter the feedback loop, the better the optimisation.

Most ecommerce conversion windows are 7-30 days. A click today needs to convert within that window for Smart Bidding to associate the conversion with the click's characteristics (device, location, time, audience signals). After 30 days, the signal is too late - the algorithm has already moved on.

LTV, by definition, unfolds over months or years. A health supplements customer might place their first order in January, reorder in March, and become a monthly subscriber by June. Their true LTV isn't knowable until Q4 at the earliest. By then, the January bidding decisions are irrelevant - you need to optimise March, April, May, and June bidding right now.

What Smart Bidding Actually Sees

When you send a first-order value of £35 to Smart Bidding, the algorithm knows exactly what that click was worth. It's accurate, immediate, and actionable. When you try to send a "predicted LTV" of £140, several things go wrong:

• • The prediction is uncertain: Your LTV model might be 40% accurate. Smart Bidding doesn't know this - it treats £140 as fact.
• • Variance is enormous: Some £35 first-order customers will indeed reach £140 LTV. Others will never return. The spread is typically 3-5x between percentiles.
• • The model degrades: LTV predictions based on 2024 cohort behaviour may not apply to 2026 customers acquired through different channels at different price points.
• • Seasonality warps everything: A customer acquired during Black Friday has fundamentally different repeat behaviour than one acquired in March.

The result: Smart Bidding optimises toward a noisy, uncertain signal. It bids aggressively because the projected values are high. When actual LTV falls short of projections (it almost always does), you've overpaid for customers whose real value doesn't justify the acquisition cost.

The LTV Projection Trap

LTV projections are seductive because they make the economics work on paper. If a customer is "worth" £140 over 12 months, you can justify paying £50 to acquire them - even if their first order only generates £12 in margin. The spreadsheet says you'll make it back.

The trap is that projections become assumptions, and assumptions become bidding inputs. Nobody goes back 12 months later to check whether the customers acquired at £50 CAC actually generated £140 in LTV. The feedback loop is broken.

We've audited dozens of accounts that bid on projected LTV. In every case, actual LTV was 30-60% lower than the projection used for bidding. The reasons are consistent:

• • Repeat rates from paid acquisition are lower than from organic/direct
• • Customers acquired through broad targeting have lower retention than high-intent buyers
• • Discount-driven first purchases attract price-sensitive customers who don't return at full price
• • Market conditions change - competitors launch, preferences shift, alternatives emerge

The gap between projected and actual LTV is a hidden subsidy to Google. You're paying today for value that may never materialise.

Cohort Lag Reality

Real LTV data looks like this for a typical ecommerce brand:

• • 30 days: You know first-order revenue and margin (after returns)
• • 90 days: You can see early repeat behaviour - maybe 15-25% have reordered
• • 180 days: Repeat patterns stabilise - you can estimate 12-month value with 60-70% confidence
• • 365 days: You have actual 12-month LTV data - but it's for last year's customers, not today's

This means the most accurate LTV data you have is always about customers you acquired 12 months ago. The world has changed since then. Your product range, pricing, competition, and marketing mix are all different. Applying last year's LTV to this month's bidding is like driving with a map from last year - the roads might have changed.

The faster-growing your brand, the less useful historical LTV data becomes. If you've doubled your customer base, half your customers are too new to have meaningful LTV data. You're making bidding decisions for the majority of your customers based on the behaviour of a smaller, older, potentially different cohort.

Practical Workarounds That Actually Work

Instead of chasing LTV-based bidding, focus on signals that are accurate and immediate:

• • First-order contribution margin: The most reliable signal. You know it within 30 days (after returns window). Feed this to Smart Bidding as your conversion value.
• • Validated cohort multipliers: If 12+ months of data prove that subscription customers have 3x LTV vs one-time buyers, apply a 1.5x multiplier to subscription first orders (conservative, not the full 3x).
• • New vs returning customer signals: Bid higher for genuinely new customers if your retention data validates higher LTV for paid acquisition. But use conservative multipliers.
• • Product category proxies: Some categories have proven higher repeat rates. Consumables repeat more than durables. Use category-level LTV data to adjust bids by product type.
• • 90-day validation cycles: Every quarter, compare actual cohort performance to the assumptions embedded in your bidding. Adjust multipliers based on reality, not projections.

The common thread: use what you know, not what you hope. First-order margin is knowable. LTV is projectable at best. Build your bidding on the foundation of certainty and layer in directional LTV adjustments only where the data justifies them.

When LTV Bidding Actually Works

There are narrow scenarios where LTV-informed bidding is defensible:

• • Mature subscription businesses: If you have 3+ years of cohort data showing consistent 70%+ retention rates, your LTV predictions are stable enough to use directionally.
• • Replenishment categories: Pet food, supplements, skincare - categories where repeat behaviour is nearly guaranteed. But even here, use conservative multipliers.
• • Known customer segments: If you can identify at the point of conversion that a customer matches a high-LTV profile (e.g., signed up for subscription, high AOV first order), you can adjust the conversion value for that specific transaction.

In all cases, the principle is the same: never bid on projected LTV at 100%. Use 40-60% of your LTV projection as the bidding input. This gives you upside if the projection holds while limiting downside if it doesn't. The rest of the LTV value is a bonus, not a bidding input.

Next Steps