Andrew Dunn

2021-2022

Caribou Financial

Senior SEO Manager, Growth Marketing. Auto refinance and insurance fintech, Series A through C.

verified
  • SEO
  • AI content
  • Design systems
  • Growth

Context

Caribou was an auto refinance and insurance marketplace moving from Series A to Series C. Growth marketing had to compound while the brand itself changed twice, including a move to the caribou.com domain.

A domain migration is the riskiest routine event in organic search. Done carelessly it costs a year. Doing one during a rebrand, at a company still finding its category, meant the migration plan and the growth plan were the same plan.

The second problem was content supply. Auto refinance is a money vertical, so every page carried legal exposure. Volume was needed, and the usual way to get volume was not available.

Owned

  • SEO and growth marketing through two corporate rebrands, including the caribou.com migration.
  • The AI content system, its legal review path, and the editorial standard it had to clear.
  • Growth roadmaps in scrum, with KPIs and OKRs, run against a shared backlog.
  • Frontend tooling: calculators, marketplace tools, and the component system the pages were built from.

Architecture

The content system is the part worth describing, because in 2021 generating regulated financial copy at volume was not a solved problem.

An attorney reviewed and approved prompt templates rather than reviewing output page by page. That inverted the bottleneck. Legal review moved to the template, which changed rarely, so generation could run against approved structures without a lawyer in the per-page loop. Editorial QA sat after generation and before publication, and anything that failed went back as a template problem rather than a copy problem.

Legal-vetted AI content pipelineAn attorney reviews and approves prompt templates rather than finished pages. Generation runs against approved templates, then editorial quality assurance passes or rejects each page before publication. A rejected page returns to the template, not to the copy, so a failure is fixed once for every page that template will produce.Draft structureAttorney reviewof the templateGenerationagainst templateEditorial QApass or rejectPublishRejected:fix the template
Review sits on the template, which changes rarely, so generation runs without a lawyer in the per-page loop. A rejection is fixed once for every page that template will produce.

The point of the design is where the human sits. Review at the template holds the standard and scales with the page count; review at the page holds the standard and does not scale at all. In a money vertical the standard is not the thing you trade away, so the only place left to find room is upstream.

The frontend side was a design system of 15+ component templatesSource note 5, plus custom JavaScript calculators for auto-loan and insurance scenarios. The calculators mattered for organic because they answered the question the searcher actually had, and the design system mattered because it made shipping a new page a composition problem rather than a build problem.

Decisions

  • Legal review at the template, not the page. The only way to get volume without lowering the bar. It also made the legal position auditable, because the approved artifact was a small set of templates.
  • Migration and rebrand planned as one project. Rejected sequencing them. Two consecutive disruptions would have doubled the window in which rankings were unstable.
  • A component system before a content push. Slower to start. It is why the page count could grow without the front end fragmenting.
  • Calculators over more articles. For an auto-loan query, a tool answers better than prose, and the tool earns links that prose does not.
  • Scrum with a growth backlog, not a marketing calendar. Put growth work in the same queue and the same review as product work.

Outcome

Total ranking keywords grew 672% (314 to 2,424)Source note 1 in the months after the migration. The framing matters more than the number: this was growth through the riskiest event in organic search, not growth in a quiet period.

The brand term reached #2 SERP, 135K monthly, 88 KDSource note 2, which for a company that had just changed its name is the clearest signal the migration held.

CTA testing raised lead conversion 5.7% lead conversionSource note 3, and running growth in scrum raised team velocity 22%Source note 4 through the funding rounds.

On the technical side: 90+ CWV, 9 technical SEO improvementsSource note 6. Product and engineering carried the parts that lived in the application, which is the honest division of credit on work that touches a live marketplace.

Revision

I would have instrumented the AI content system's quality earlier. Editorial QA caught what it caught, but it was human judgement without a measured baseline, so I cannot say now how much the template discipline was actually worth. On the migration, I would have kept the old domain's redirect map under version control from the first day rather than the day it became urgent.