How AI can reduce friction in KYC and onboarding
Every bank I talk to has automated some part of KYC. Document capture, identity verification, sanctions screening, risk scoring. Each piece works well enough on its own. Customers still walk away.
The average onboarding abandonment rate in identity verification sits at 34% this year, per Pactvera’s 2026 KYC and Identity Verification Trends Report. Established banks tend to run meaningfully lower. Unfamiliar fintech products tend to run well above it. Document upload alone accounts for 15 to 30% of single-step abandonment, the highest of any stage in the process. Sanctions and PEP screening drags in its own way further down the funnel: industry benchmarks put the false-positive rate on these checks as high as 95%, so nearly every flagged match still needs a human to clear it before the customer can move forward. We should stop treating these as a UX problem that a better camera interface will smooth over. They point at something structural.
Nobody owns the handoffs
I have watched banks automate identity capture, then risk scoring, then sanctions screening, and lose the same share of customers they lost before. The reason is simple. Each stage gets its own AI tool and its own owner. The handoff between them belongs to no one. A customer who clears identity verification cleanly can still stall for days, waiting on a risk assessment that never learns what identity verification already confirmed.
One in five onboarding applications gets abandoned primarily because of KYC and AML friction, largely on the retail side. Corporate onboarding has its own version of the same problem, on a different clock: Fenergo’s 2023 survey of corporate and institutional banks found the average KYC review took 95 days that year, up from 84 the year before. Think about that. The number moved in the wrong direction while automation spending moved up.
Where AI is already removing real friction
The gains are real where they exist. SGA’s own work on entity resolution for a global bank produced 2X attribute coverage, an 87% reduction in onboarding time, and a 50% drop in cost. We got there by enriching and reconciling entity data at scale instead of leaving it fragmented across systems. That is not a marginal UX fix. That is what happens when the data feeding every downstream stage becomes trustworthy before it ever reaches a human reviewer.
McKinsey’s analysis of corporate client onboarding found advanced digital banks cutting onboarding times by up to 85% while bringing roughly 80% of new clients on digitally. Behind numbers like these, the pattern repeats. The improvement never comes from one clever tool. It comes from treating identity, risk, and compliance data as one connected pipeline rather than three separate automations that happen to sit near each other.
Speed without trust just moves the risk downstream
Faster is not automatically better. A mid-sized European retail bank cut onboarding time sharply by relaxing document review thresholds. A few months later, SAR volume in downstream monitoring spiked. The friction had not disappeared. It had moved from the front door to the back office, where it cost more to find and took longer to catch.
I have written about this same lesson with enterprise AI generally. A model that speeds up a decision without being able to explain that decision is not removing friction. It is deferring it, with interest. In KYC specifically, a risk score a compliance officer cannot interrogate is a risk that eventually gets escalated, re-reviewed, and slowed down anyway, and regulators are making that non-negotiable: model risk management guidance like the Fed and OCC’s SR 11-7 already expects institutions to explain and validate the logic behind a risk decision, not just its output. Speed and trust have to be built together.
The compliance team has to be in the room, not handed the output
I have seen onboarding tools built entirely by technology teams, tested against clean synthetic data, then handed to compliance once they were ready to deploy. Compliance spends the first quarter finding every edge case the model was never shown. A politically exposed person with a common name, the exact kind of match our entity resolution work is built to resolve cleanly instead of flagging on a coin toss. A beneficial ownership structure spanning four jurisdictions, each with its own UBO disclosure threshold, so a stake that triggers reporting in one country sits below the radar in another, and that gap is where the ownership chain gets missed. That kind of edge case is a design problem, not a backlog problem, and it shows up long before any bank reaches a remediation mandate that makes headlines.
A model trained without the people who see those cases daily will not catch them. Adoption, not procurement, decides whether a KYC tool reduces friction or just relocates it. The compliance officer who has to defend a decision to a regulator needs to have shaped that decision’s logic. Inheriting it is not the same thing.
What good looks like
The onboarding flows that actually reduce friction share a few traits I keep seeing, and together they add up to what the industry now calls perpetual KYC. Identity, risk, and compliance data sit in one pipeline. Every AI-assisted decision carries an explanation a compliance officer can stand behind, not just a score. And the redesign starts with the people who will defend the outcome, not only the people who will build the model.
That is what separated the entity resolution work we did for that global bank from a typical point solution. The 87% reduction in onboarding time did not come from a faster document scanner. It came from fixing the data foundation every downstream stage depended on, so speed and accuracy stopped competing with each other.
Conclusion
The banks that win the next phase of onboarding will not be the ones with the most AI tools stacked across their KYC process. They will be the ones that removed the friction between those tools, put compliance in the room while building them, and made sure every faster decision is one they can still explain when it matters most.