# Financial Services: Agents onboard loan portfolios in weeks, not months

> How a global credit manager cut portfolio onboarding time 92% and £14m a year in labour cost with an agentic migration pipeline that reviews every account.

**Published:** 18 September 2026
**Last Updated:** 18 September 2026

**Source:** [https://www.elsewhen.com/work/agents-onboard-loan-portfolios-in-weeks-not-months/](https://www.elsewhen.com/work/agents-onboard-loan-portfolios-in-weeks-not-months/)

---

Overview

Credit management firms grow by taking on loan portfolios, and every one arrives on a different legacy system in a different shape. A global credit management company was spending months and a team of 8 to 15 people on each, and could afford to check only 1 per cent of the accounts it moved.

Elsewhen built an agentic migration pipeline with AI at the bookends and a deterministic, auditable core, working as a prototype in the client's own cloud tenant within 2.5 weeks. It maps each portfolio onto the client's data model, checks every account and sorts out the exceptions before anyone looks at them. The payoff: 92 per cent off the time from taking on a portfolio to servicing it, and £14 million a year off labour cost.

| Metric | Value | Source |
| --- | --- | --- |
| in labour cost reduction, running in the client's own tenancy | £14m a year  |  |
| less source-to-servicing cycle time | 92%  |  |
| of accounts reviewed, up from a 1 per cent sample | 100%  |  |

Before

### Only 1 per cent of a migration ever got a human review

Each portfolio meant weeks of manual schema mapping, then validating records by hand, with onboarding capacity capped by headcount. Sampling was the only way to cope, which left a backlog, held back growth and left risk with the firm after go-live. Nobody had time to check whether the interest rate on each migrated loan matched its original terms, and the liability was real. A large acquisition on a separate system put a fixed deadline on all of it.

After

### Every account reviewed, every decision traceable

The firm now has a migration pipeline that runs in its own cloud, that it owns outright and that can replay any decision it has made. Every account in a portfolio is checked rather than one in a hundred, and the people who used to key in records now review the exceptions.

A language model drafts the mapping from the seller's system onto the firm's own. Plain code then tests every account against the business rules and regulatory gates, including each loan's interest rate against its original terms. An agent works through whatever fails those tests and passes only the real judgement calls to a person. The model never touches the reconciliation itself.

The same pipeline takes the next portfolio, and the one after that, so growth is no longer a question of headcount.