A consumer fintech offering checking, savings, and an investing wrapper replaced application-start bidding with predicted 12-month funded deposit value using AdZeta's ValueBid™. The bidder began winning impressions for applicants likely to fund and retain meaningful balances, reducing CPA on funded accounts 36%, lifting average 12-month deposit value 28%, and improving funded-account ROI 22%.

A growth-stage consumer fintech offering a checking account, high-yield savings, and a managed investing wrapper, primarily acquiring through Google Search, Meta Advantage+, and YouTube. Customer economics depend on funded balance retention at month 6 and month 12, not application-start volume.
Roughly 47% of started applications never funded. Of the 53% that did fund, average 12-month deposit value spanned a roughly 14x range from $250 to $3,500. The bidder saw all application starts as identical conversions. Google and Meta were optimizing toward application volume, and the auction was systematically pulling in promo-driven sign-ups that funded $50, hit the welcome bonus threshold, and went dormant.
Campaigns ran on Google Search with Maximize Conversions and Meta Advantage+ Shopping with first-event optimization toward application start. The team had already tested funded-account-only conversion windows, but the longer feedback loop (4 to 8 weeks for funded confirmation) starved the bidder of training signal. Audience layering on credit-quality proxies produced marginal lift but did not solve the core signal problem.
AdZeta deployed ValueBid™ to predict each applicant's 12-month funded deposit value at the moment of application from non-PII first-party signals (application path depth, account-tier interest, employer/payroll-direct-deposit indicators where consented, product mix during application). Predictions flowed into Google via OCI and Meta via CAPI as the conversion value the bidder optimized against, in real time, without waiting for funded confirmation.
Integration ran from the brand's identity, application, and core banking systems into the AdZeta pipeline under SOC2 Type II, ISO 27001, GDPR, CPRA, and PIPEDA frameworks. Model training completed in Week 3 with AUC 0.84 on 12-month funded deposit value prediction. ValueBid™ went live in Week 4 with a paired control on 50% of media spend for eight weeks. Following the relearning period, ValueBid™ became the default conversion signal across both platforms.
At Week 22 post-relearn, CPA on funded accounts was down 36%, average 12-month deposit value was up 28%, funded-account ROI was up 22%, and the share of applications converting to funded accounts had improved 19%. The brand expanded paid search prospecting on high-pLTV intent terms that had previously failed first-event-optimization unit economics.
In consumer fintech, the cost of acquisition is paid in full at application start. The revenue is paid out over months and years of funded balance retention. The auction sees only the application. The brand sees a 14x value distribution between funded and dormant applicants. Without a forward-looking signal that captures funding probability and deposit magnitude, the bidder cannot tell a $50-funder from a $3,500-funder. Both look like the same conversion.
Optimizing for application starts maximizes the number of people who fill out a form, regardless of whether they will fund. In a category where 47% of starts never fund, application-start optimization is fundamentally pointed at a vanity metric.
The team had tried optimizing toward funded-account conversion as a workaround. The 4-to-8-week funded-confirmation lag starved the bidder of training data, and learning phase performance never stabilized. ValueBid™ solved this by predicting funded value at application time, giving the auction a real-time signal with the predictive power of a downstream event.
Welcome bonuses pull in promo-driven sign-ups who fund the minimum, claim the bonus, and go dormant. Discount-aware features in the ValueBid™ model down-weighted predicted value for promo-driven application paths, preventing the bidder from over-investing in cohorts that historically produced near-zero deposit value at month 6.
ValueBid™ predicts each applicant's 12-month funded deposit value at the moment of application, then delivers that prediction as the conversion value to Google OCI and Meta CAPI. The auction relearns on a forward-looking signal with the predictive power of a downstream event, and starts winning impressions for applicants likely to fund and retain meaningful balances rather than promo-hunters.
AdZeta's brand-specific model predicts 12-month funded deposit value using application path depth, account-tier interest, payroll-direct-deposit indicators where consented, and product mix during application. The model is trained on 24 months of funded and lapsed-applicant data and reaches AUC 0.84 on holdout.
The integration runs under SOC2 Type II, ISO 27001, GDPR, CPRA, and PIPEDA frameworks (trust.adzeta.io). All signals are non-PII at the AdZeta layer. Every prediction sent to Google or Meta is logged with feature contributions for the brand's compliance, audit, and finance teams.
Predictions flow as conversion values within seconds of application submission, eliminating the 4-to-8-week funded-confirmation lag that previously starved late-conversion bidders. Google Search and Meta Advantage+ campaigns kept their existing structure.
AdZeta's reporting layer breaks 12-month deposit value forecasts out by acquisition cohort, channel, and account tier. The brand's finance team uses the cohort forecast as a primary input to deposit growth planning and treasury modeling.
Google Search with Maximize Conversions optimized toward application start. Meta Advantage+ Shopping with first-event optimization. The default configuration the brand had been running for 30 months.
Same campaign structure, same audiences, same creative. ValueBid™ predicted 12-month funded deposit value replaced application-start as the conversion value sent to Google OCI and Meta CAPI.
The cleanest comparison was pre/post on the funded-account cohort, holding audiences and creative constant. Application volume actually dipped modestly during the rollout. Funded volume rose. Funded-account deposit value rose more. The auction stopped paying for promo-driven sign-ups that hit the bonus threshold and went dormant, and started paying for applicants likely to direct-deposit their paycheck. The compounding effect on quarterly deposit growth dwarfed the application-volume delta.