# Case studies.

> HTML 版（正本）：https://voidvector.tech/cases/
> 語言：en · 來源：voidvector（虛空向量有限公司）

Selected work across AI applications, UX design, growth operations, and marketing growth.

- **GEO quantitative research, in progress: a reinforcement learning × transformer hybrid.** — Our in-house research program: which conditions keep content findable and cited by AI engines even after the engines change? We train a hybrid model — transformers for content representation, reinforcement learning on real citation feedback — instead of guessing at rules. The model's own numbers get published only when we can stand behind them, but the method has already done real work: one merchant's overall customer visits grew 109% in the first month after rollout. (https://voidvector.tech/cases/geo-quant-research/)
- **Tightening platform operations and service flow to lift a learning app's paid conversion and retention.** — A subscription language-learning app was adding subscribers faster than its service flow could follow: support volume climbed with the subscriber base, canned replies stopped coping, and nobody was there at the moments users were deciding whether to keep paying. We consolidated help, course, and plan rules into a retrievable knowledge base and rebuilt support and human handoff around it, so answers arrive at the moment of decision — paid conversion and retention rose with it. (https://voidvector.tech/cases/ai-support-retrieval/)
- **Redesigning a heavy system: UX and visual design together, so people stop paying to learn it.** — An ERP has every feature and still loses to a spreadsheet, because using it means learning its taxonomy first. We rebuilt our own Odoo as a portal: the things people actually touch daily sit on the first screen, the back office is one link away, and the interface was redesigned and re-skinned in the same pass — so the first time you open it, you already know where to click. (https://voidvector.tech/cases/erp-portal-redesign/)
- **Repositioning a B2B scheduling SaaS so renewals stop depending on rescue calls.** — A B2B scheduling SaaS sold to everyone and kept no one. We rebuilt the brand positioning around the customers who actually stay and replaced the pre-renewal fire drill with a retention loop — annual renewals climbing and customer LTV growing alongside them. (https://voidvector.tech/cases/b2b-repositioning-retention/)
- **Running SEO and GEO in parallel made a home fragrance brand both searchable and citable.** — A home fragrance brand had nothing on its site but product pages — invisible to search, nonexistent to AI engines. We rebuilt its content assets on twin SEO and GEO tracks: organic traffic climbing and citations by mainstream AI engines multiplying. (https://voidvector.tech/cases/seo-geo-dual-track/)
- **A fitness app's personalized plans: AI builds the workout, UX makes people follow it.** — A fitness app's AI-generated training plans were being ignored — half a model problem, half an experience problem. We ran the AI application and UX research as one project: day-7 retention and plan completion both climbing. (https://voidvector.tech/cases/adaptive-training-plan/)