Rapid exploration · Financial services
Agentic AI tools that cut manual review by 70%, first agents working in a week.
QA automation, a knowledge base agent, and a company insights engine, designed, built, and put in front of the team fast enough to prove value on evidence, not slides.
The situation
A financial services firm was carrying a heavy load of manual review work: quality checks, knowledge lookups, and company research that consumed skilled people's time without using their skill. Like most firms, they'd heard every AI vendor pitch going. What they hadn't seen was working software in their own environment, on their own processes.
The brief was a rapid exploration: rather than commissioning a study about whether agentic AI could help, build the agents and find out.
The constraints
- Evidence over opinion: the firm wanted to make its investment decision on working output, not a feasibility deck.
- Financial services context: accuracy and auditability mattered more than raw speed. An agent that's confidently wrong is worse than no agent.
- It had to slot into how the team already worked, not require a re-platforming project before showing any value.
What I built
- QA automation: agents that handle the first pass of quality review, escalating to a human only where judgment is genuinely needed.
- Knowledge base agent: natural-language access to the firm's internal knowledge, so answers that took a search-and-skim session arrive in seconds with sources attached.
- Company insights engine: automated gathering and synthesis of company information that previously required manual research.
The first working agents were in the team's hands within one week. From there the work was iterative: watch how the team actually used them, tighten the escalation thresholds, and expand coverage where the value showed up.
The outcome
- 70% reduction in manual review effort on the covered processes.
- A decision made on evidence. The firm saw the agents working on its own data before committing further investment.
- A pattern the team understood and could extend, rather than a black box they depended on me to run.
Why it worked
Agentic AI is where the gap between people who talk about AI and people who build it is widest. Designing agents that know when to act and when to hand off to a human is an engineering judgment, not a slide, and it's the difference between a tool a regulated firm can actually use and a demo that never leaves the sandbox.
This maps to the Rapid exploration engagement on the services page. Typical 3–5 days to a working proof of concept, with a fixed or capped fee agreed upfront.