How a Fintech’s Fact Checker Toolmate slashed CFP review time
The Fact Checker GPT that Hillary built for our marketing team has improved efficiency in our content review process and provided impactful learnings for how we use LLMs in our day-to-day marketing efforts. Would highly recommend Hillary and her team.
A Marketing Lead at a Fintech Company
THE CHALLENGE
A US-based fintech publishes financial education content for goal-oriented investors — tax strategy, retirement planning, investment fundamentals. Every post requires CFP sign-off before it goes live. For them, accuracy isn't a nice-to-have. It's the product.
The problem wasn't the standard. It was the process. CFP reviewers were doing the same repetitive work on every single draft: hunting for outdated contribution limits, chasing missing citations, flagging disclaimer inconsistencies. There was no standard format, no shared source hierarchy, no way to know if a post was ready before it hit the reviewer's desk. Every article started from scratch.
The publishing calendar paid for it. CFP capacity is limited and expensive. When reviews ran long, content backed up. When content backed up, the team scrambled. The fintech didn't need more CFPs. They needed a system that made the CFP's job easier — before they ever opened the doc.
We built from the inside out — starting with how a CFP actually reviews, not with what a generic AI can do. The result is a Fact Checker Toolmate that functions as a structured first-pass reviewer: it flags, formats, and sources. The human decides.
Step 1: Workflow Discovery
We mapped the fintech's existing review process end-to-end: what CFP reviewers always checked, where writers consistently missed the mark, and which claim types carried the most compliance risk. From that, we defined the toolmate's scope — tax limits, contribution caps, APYs, deadlines, statutory definitions — and established a source hierarchy with clear tie-break rules for conflicting data.
Step 2: Context Engineering
We translated the CFP review process into a system the toolmate could execute reliably. That meant custom GPT instructions built around the actual workflow, a compliance disclaimer mapping guide (if X claim type → include Y disclaimer), and a knowledge base trained on the fintech's brand book, CFP playbook, and audience personas. So flagged suggestions aren't just accurate — they're on-brand.
Step 3: Structured Output + Pilot
We designed the output format for CFP speed: claim → recommended revision → source link → as-of date → footnote index → bibliography rollup, with a link count summary. Then we ran a 2-week pilot on real posts to validate accuracy on high-risk claims, citation completeness, and readability of suggested edits. After revisions and hand-off, the toolmate slotted directly into the existing pipeline — activated after draft, before final CFP review.
Step 1: Brand Diagnostic
We started by understanding what reusable brand assets Growth Friday actually had, and what was missing. Through discovery interviews, competitive analysis, market research, and a content audit, we mapped content gaps, strategic opportunities, and existing brand assets that could be repurposed or needed to be rebuilt from scratch.
This phase gave us a complete picture of where the brand stood and what needed to change to support the new positioning.
Step 2: Context Engineering
Next, we translated everything we learned into a system both humans and AI could use reliably. We codified Growth Friday’s new brand voice, audience insights, and best practices into reference documents, style rules, example libraries, and reusable templates.
This became the brand’s single source of truth for copywriting.
Step 3: Custom Brandwriter Toolmate
Only after building that foundation did we configure the toolmate. We drafted custom instructions that tied the AI directly to Growth Friday’s knowledge base—so every output reflected their actual voice, not a generic approximation.
The toolmate became a tool the team could trust because it was trained on their rules, not ours.
