We began by resetting the performance model. Instead of optimizing to lead submissions, we shifted the program toward downstream business outcomes: completed property profiles, verified seller opportunities, buyer-accepted matches, scheduled consultations, and signed transactions weighted by expected gross profit. This aligned paid media to how the marketplace actually created value.

That required a substantial upgrade to the measurement stack. Working across growth, product, engineering, analytics, and marketplace operations, we rebuilt conversion tracking across GA4, Google Tag Manager, server-side tagging, enhanced conversions, Meta CAPI, Google offline conversion imports, and CRM event capture. We connected click and lead data back to outcomes in HubSpot, used CallRail and WhatConverts to improve call and source visibility, and centralized reporting in BigQuery and Looker Studio so leadership could evaluate performance by market, seller type, offer path, and unit economics.

To make the system more precise, we strengthened both audience activation and cross-channel measurement. Hightouch was used to sync first-party lifecycle data into Google, Meta, and TikTok, enabling cleaner suppression, retargeting, reactivation, lookalike seeding, and segmentation based on actual seller progression rather than platform proxies. Northbeam provided a broader measurement layer across search and paid social, helping the team evaluate blended CAC, assisted conversion impact, and the contribution of upper-funnel media more accurately as the channel mix expanded.

We then added a market economics layer that sat above the media account. Using 2025 FRED housing and rate data, paired with internal buyer density, offer acceptance rates, average gross profit per transaction, and days-to-close by market, we built a market liquidity score that classified metros into opportunity, neutral, and danger zones. Markets with strong buyer depth, acceptable close velocity, and messaging fit for speed-and-certainty offers received incremental budget. Markets where elevated prices and thin transaction liquidity inflated seller expectations without producing enough buyer competition were deprioritized. This gave the company a much more rational way to allocate capital than optimizing each market in isolation. The external housing indicators themselves pointed to a market that remained tight on affordability, uneven on activity, and highly sensitive to local conditions.

Once the signal and market logic were fixed, we rebuilt the account architecture. Search was restructured around distinct seller-intent clusters such as certainty and speed, avoid repairs, compare offers, inherited property, landlord exit, and pre-foreclosure urgency. Branded, non-branded, and competitor demand were separated cleanly. Geographic expansion was tied to the liquidity model, not just lead costs. Performance Max was retained only where downstream signal quality and asset governance were strong enough to support it. Meta and TikTok were repositioned from low-friction lead capture into higher-quality education and intent shaping, with campaigns built around seller scenarios and objection handling rather than generic “cash offer” creative.

Creative was rebuilt as well. We introduced localized AI creator videos and influencer-style explainers that translated a complicated marketplace promise into something more concrete and trustworthy. Instead of sterile direct-response ads, the company ran city-specific short-form videos featuring a consistent AI spokesperson who explained how the vetted-buyer network worked, why multiple offers mattered, what homeowners should compare beyond headline price, and when speed versus maximum proceeds was the right tradeoff. These assets were deployed across Meta Reels, TikTok, and YouTube Shorts to improve top-of-funnel education, reduce lead-form curiosity clicks, and lift branded search over time.

We also treated AI search as an emerging demand surface, not a future footnote. The company’s site architecture, city pages, buyer-vetting pages, FAQs, trust content, reviews, schema, and market explainers were rebuilt to increase the probability that large language models and AI-led search interfaces would surface the brand as a credible marketplace rather than just another generic cash-buyer site. A lightweight internal agent monitored LLM prompt outputs, review themes, search-query drift, and landing-page performance to identify content gaps, messaging opportunities, and negative-keyword additions. The objective was not to automate judgment. It was to make the system learn faster.

To prove what was actually driving lift, we also used synthetic control rather than relying only on platform-reported conversion deltas. The strongest use case was testing a new “vetted buyer competition” landing-page experience and AI creator video rollout across selected metro clusters. Because seller behavior moved with seasonality, rates, local inventory conditions, and buyer appetite, standard A/B testing at the market level would have overstated confidence. Instead, we built matched synthetic controls using untreated markets to estimate what would likely have happened without the intervention, then compared that counterfactual to actual verified-seller and gross-profit outcomes. That allowed the client to make scaling decisions with more confidence and less channel mythology. The approach closely follows the logic outlined in our synthetic-control lift case study, where causal inference was used to isolate impact from market noise. 

Finally, we addressed post-click friction. Seller-intake flows were redesigned to reduce abandonment, ask for higher-value information at the right stage, and dynamically route users based on address, property condition, urgency, and expected fit. Optimizely, Hotjar, and FullStory were used to improve the form path and surface where motivated homeowners were dropping off. The result was a cleaner handoff between media, qualification, and marketplace operations.