Tasks
We designed, built and now operate the AI layer across five roles:
- SEO & content specialist — a 12-week, 36-article content programme across three market websites, researched from Ahrefs and Search Console data, written in Dutch, Swedish and Finnish, illustrated, and published straight into Webflow CMS. Includes interactive calculators embedded in articles and a recurring “rate audit” that crawls all three sitemaps and flags every tax rate, fee and threshold that has gone stale.
- Email marketer — a four-month newsletter plan for the Swedish base: campaign build, copy, scheduling, automated non-opener follow-up audiences, and a monthly performance report.
- Loan operations — applications are read from the brokerage platform, checked against each lender’s criteria and submitted into partner portals automatically; applicants receive templated SMS asking for the bank statements and open-banking connections when their file is missing.
- Sales operations manager — call data is pulled from the telephony platform, recordings transcribed in Dutch and Swedish, and each agent receives a monthly scorecard with connect rates, real talk time and coaching points.
- Business developer — market and pricing analysis, revenue models, partner-offer design and client-ready presentations.
Team set-up
- 1 AI employee with workspace-wide access to the company’s tools
- Client-side operations lead as product owner and approver of anything customer-facing
- Codepole: 1 engineer for integrations, automation scripts and monitoring
Results & Benefits
- Five roles covered by one AI employee, running on schedules rather than on someone’s memory.
- Around 20 automated jobs run every working day across loan operations and marketing, each reporting its outcome — including the runs where nothing needed doing.
- Content: a three-language publishing pipeline delivering 3 articles per week, live the same day. The first automated rate audit checked 311 pages and found 6 factual errors — including a VAT calculator whose widget used a different rate than its own surrounding text.
- Email: reporting corrected to measure opens on sent rather than delivered, which had been hiding a substantial deliverability loss, plus a follow-up audience of 8,405 mailable non-openers rebuilt automatically each cycle.
- Loan operations: 300–400 live applications re-checked several times per working day, a full pass taking about two minutes, with automatic SMS follow-up, lender submissions, policy enforcement and closure of stale files.
- Sales QA: 3,300+ calls analysed in the first review, surfacing 224 lost inbound calls in 30 days and near-zero call-recording coverage on the Dutch lines — both invisible in the telephony vendor’s own dashboards.
- The human team moved from doing this work to reviewing it: approvals, exceptions and judgement calls.
