GA4 defaults to data-driven attribution (DDA) for all conversions, a substantial shift from Universal Analytics' last-click model. DDA uses machine learning to analyze your actual conversion paths and assigns fractional credit to each touchpoint—ads, organic visits, email clicks, direct traffic—based on how much each channel statistically influenced the outcome. If you don't have enough conversion volume (typically under 400 conversions per month for a given action and 20,000 ad interactions), GA4 falls back to last-click attribution until your data reaches the threshold. You can still switch to other models in the attribution settings: last click, first click, linear (equal credit to all touches), position-based (40% to first and last, 20% split among middle touches), and time decay (more credit to recent interactions). Most businesses leave it on data-driven because it's closer to reality than arbitrary rules, but ecommerce clients with very short buying cycles sometimes prefer last-click for simplicity, and lead-gen campaigns with long nurture sequences often test position-based to credit both awareness and closing touches. Crucial limitation: GA4's attribution only works within its own session and event data. If someone sees your Facebook ad on mobile, later Googles your brand on desktop, then converts via a direct visit three days later, GA4 ties that together through Google signals and user-ID tracking when enabled—but it still won't see offline touchpoints, phone calls, or un-tracked referrals. At Ottawa SEO, we cross-reference GA4 attribution reports with UTM-tagged campaign data and CRM records to fill those gaps, especially for clients running coordinated SEO, PPC, and email plays across Toronto and Ottawa markets. One watch-out: data-driven attribution recalculates continuously as new data comes in, so month-over-month channel performance can shift slightly in historical reports. If you're reporting to stakeholders who panic at retroactive changes, document that behaviour up front or export snapshots at month-end. The model is objectively better than last-click for multi-channel strategies, but it requires educating clients that attribution is probabilistic, not a fixed ledger.