# Why Relationship Reviews Take Two Days And What Commercial Banks Are Doing About It _Published: 2026-09-10T00:00:00.000-04:00_ A commercial relationship review can take two days. See how a dual workforce model gives analysts that time back and catches risk sooner. _Most of a relationship review is assembly, not judgment. Here's how agentic AI gives commercial credit teams that time back._ A single relationship review can take a commercial credit team two full days. On complex borrowers, a week. Multiply that across a growing book and the numbers stop working, because at most U.S. commercial banks loan volume is outpacing credit team headcount. The result is a backlog that costs banks twice: decisions on existing relationships move slower, and the rest of the book gets watched on a lag. The fix isn't more analysts. A growing number of banks are handing the repeatable parts of the review to AI and keeping their people on the judgment that protects the portfolio, part of a broader move toward [agentic AI in commercial credit](https://www.ncino.com/blog/agentic-ai-manual-data-assembly-commercial-analyst-judgment-work) now moving from concept to production. ## **The Backlog Costs More Than Speed** **The speed cost is visible**: when reviews back up, so do the decisions that depend on them. A renewal or a line increase for an existing borrower, the kind of growth that's supposed to be the easy part, waits behind a backlog. **The risk cost is easier to miss**. A review that's weeks stale can miss a covenant slipping or a borrower's liquidity thinning, and by the time the next cycle comes around, the problem is bigger and harder to work out. That second cost compounds. When review cadence falls behind loan growth, the whole book drifts toward being watched on a lag, which is exactly when a downgrade or a default arrives as a surprise. The effect scales with the book — the larger and older the portfolio, the more relationships sit between scheduled reviews at any given moment, each aging since its last real look. What good looks like is the opposite: a shift from reactive to proactive, catching a problem while there's still room to act, well before it turns into a loss. Get there and analysts spend less of the week fighting fires and more of it on the relationships that need real judgment. The hard part has always been capacity. On an annual cycle, trouble that surfaces in month two can sit unseen for 10 more, and a two-day manual review makes catching it any sooner impractical at scale. It traces back to how the day actually gets spent: [McKinsey notes the front line puts significant time](https://www.mckinsey.com/capabilities/risk-and-resilience/our-insights/embracing-generative-ai-in-credit-risk) into collecting information, running analyses, and writing memos for credit decisions. When most of a review is assembly, more loans without more analysts means a longer queue, not faster throughput. ## **What the Analyst Hands Off, and What Stays Human** Commercial banks can keep the same review depth and still cut the time it takes by changing who does which part of the work. An agent takes on the assembly: pulling and re-spreading the borrower's latest financials, testing covenants against the terms already in place, generating the relationship review when a borrower's risk status changes, and flagging the early-warning signals worth a closer look. The analyst picks up where judgment starts, reading the output, weighing the complex credits, and owning the decision. That division of labor is the [dual workforce](https://www.ncino.com/blog/ai-in-banking-dual-workforce-already-here), where the agent runs the routine, and the banker runs the call. The agent handles: The banker owns: Pulling the relationship's financials and spreading the latest tax statements Reading the assembled output Testing covenants against the institution's own criteria Weighing the complex credits Drafting a first-pass review and flagging early warning signals Making the final decision The split is easiest to see in a single covenant test. The agent pulls the borrower's latest financials, runs the calculation against the terms on file, and flags a debt-service coverage ratio that has slipped below its threshold, with the trend, the recent filings and the relationship history already attached. A breach can be a data-quality issue, a one-time accounting adjustment, or the first sign of real deterioration, and telling them apart takes a read of the borrower, the industry and the moment. The agent surfaces the signal; the banker decides what it's worth. [Automation has under-delivered in credit before](https://www.ncino.com/blog/what-makes-automation-intelligent-in-banking-a-guide-to-ai-powered-process-automation), so skepticism is warranted. But the current shift looks different in scale. [McKinsey](https://www.mckinsey.com/industries/financial-services/our-insights/banking-matters/agentic-ai-will-shake-up-banking-shrinking-global-profit-pools) groups robotic process automation with the tools that disappointed and describes the move to agentic AI as one that "appears to be for real," with early use cases cutting manual workloads by 30 to 50%. That range is an industry signal, not a single vendor's claim. ## **The Risk Decision Stays in Your Hands** You set the boundaries the agent works inside. The exposure limits and risk criteria are yours, and low-risk, repeatable relationships run on their own while anything above that line goes to a banker before it moves. In practice, that line follows rules the bank already uses: a routine review on a performing, well-within-covenant relationship can run and post for a banker's sign-off, while a covenant exception, a risk-rating change, or an exposure above a set threshold routes to a person first. You keep that control. What changes is how much of the routine ever reaches the team. When a review does reach a banker, judgment still runs the call. They read the covenant results, the spread and the draft, check it against what they'd have concluded, and either sign off or send it back. A person signs every credit decision, and the process keeps the same control points the bank already uses for manual review, so nothing about who approves what has to change. That's also what determines whether the team actually uses it. Credit teams have set aside tools before, usually because they couldn't see how the answer was reached. That caution is well placed. [The CFA Institute has found](https://rpc.cfainstitute.org/research/reports/2025/explainable-ai-in-finance) transparency and human oversight are essential and what makes these tools dependable. An agent handling the review has to show its work: the banker sees the same inputs they'd have assembled by hand and keeps the final say, so trust comes from checking the output, the same way they trust their own analysis. ## **What This Looks Like in Production** nCino built the [Analyst Digital Partner](https://explore.ncino.com/analyst-digital-part/) for exactly that division of labor. For an existing book, Analyst Digital Partner runs the review through a Credit Risk Manager sub-agent that monitors relationships continuously and re-tests covenants against the terms on file, generating a fresh review the moment a borrower's risk status shifts. The banker opens a finished draft ready to check, inside the lending workflow the team already uses. Early adoption is already underway: WaFd Bank is [piloting the Analyst Digital Partner](https://finxtech.com/early-adopters-of-agentic-ai-find-lending-efficiencies/) for credit risk assessment and financial analysis. While that pilot is just getting started, the measured results are already in. Across nCino customers using the Analyst Digital Partner within nCino Commercial Lending, the time to complete a relationship review has dropped [60 to 70%](https://investor.ncino.com/news-releases/news-release-details/ncino-analyst-digital-partner-cuts-commercial-relationship/). What sits behind that number is the assembly coming off the analyst's plate: the gathering, the spreading, the first-pass write-up. The hours that went into building a review now go into reading it and making decisions. Analysts spend their time on the relationships that move the portfolio, and the review keeps pace with the book. Because assembly no longer costs two days, the review stops being a once-a-year event. The same results let institutions move from annual or quarterly cadences to weekly or even daily ones, so credit deterioration surfaces as it forms, before the next scheduled review. That's the reactive-to-proactive shift in practice: the speed gain clears the decision backlog, and the cadence gain shortens the distance between when a credit starts to slip and when someone acts on it. On a weekly cadence, a liquidity slide that would have waited for the annual review surfaces while the borrower still has room to move, which can be the difference between adjusting a facility and writing part of it off. ## **A Realistic First Step** Starting doesn't take a platform overhaul — one enterprise-sized institution went live with Analyst Digital Partner in 36 minutes. The realistic entry point is a single workflow, and relationship reviews are the common first choice: high volume, heavy on assembly, already running on the lending workflow the team uses. Enablement is FI-led and self-service, so you can set the pace and the scope. - **Start narrow.** Turn it on for a defined set of low-risk, high-volume relationships. - **Prove the workflow.** Let analysts check the agent's output against how they'd have built the review by hand. - **Extend as confidence grows.** Widen the scope as the team trusts what comes back. By the time the scope widens, it's building on reviews your analysts have already checked and signed off. That's how the two days come back: one workflow at a time, at a pace the team sets. **** **What would your team do with that time?** [See the Analyst Digital Partner in action.](https://explore.ncino.com/analyst-digital-part/) --- [View sitemap](https://www.ncino.com/api/markdown/sitemap)