The platform behind Selify
Selify is an AI-native operations platform for multi-channel commerce. Below: the architecture, the moat, the unit economics. No marketing-ese. For founders, operators, and investors who scroll.
1. Market
E-commerce operations tooling is fragmented across eight categories: support, marketing, SEO, sync, returns, content, analytics, and conversational commerce. Mid-market merchants stitch 5–8 tools together at $300–800/month per store and pay the integration tax with VA labor. We see the same recipe across every DTC brand we've talked to.
The structural shift we're betting on: AI-native consolidators replace stitched stacks, and per-outcome billing replaces seat-based pricing. Per-outcome scales with merchant value instead of fighting it.
Beachhead: 1000+ SKU apparel and beauty operators on Shopify Plus, where AI volume per merchant is highest and the pain is most acute. From there: adjacent verticals (beauty subscription, home goods, multi-brand fashion houses), then upmarket as the per-outcome margin curve proves out at production volume. Every channel we add — Amazon next, marketplaces after that — multiplies merchant value without multiplying engineering cost, because the agent runtime, customer profile, and knowledge graph are shared.
2. Architecture
Cloudflare edge
├─ SvelteKit (dash + admin + selify.ai)
├─ Kong gateway
│ ├─ FastAPI services (workflow, mcp, webmail, webhook-gateway)
│ └─ Rust services (auth-proxy, git-api)
├─ Temporal (5 worker queues — ai, commerce, messaging, internal-ops, product-generation)
├─ Modal (GPU pool — try-on, vision, embeddings, on-demand)
├─ Postgres + pgvector (per-tenant isolation, RLS)
└─ Supabase Auth
No request waits on a GPU job. Long-running operations (bulk product sync, training jobs, campaign generation) live in Temporal workflows with retries and signals. The dashboard never times out, the worker pool scales to zero, the customer never sees infrastructure.
3. The LLM moat
We run DeepSeek (chat + R1) on Modal, not OpenAI. For our task mix — structured customer support, listing generation, intent classification — DeepSeek's published per-token pricing is materially lower than the major US frontier APIs at quality parity for these tasks. We own the prompt and eval pipeline; switching providers is a config change. Self-host fallback is ready: every model the platform depends on can run inside a customer's VPC on the Enterprise tier. We don't publish internal cost-per-outcome benchmarks until design-partner production data hardens them.
4. Unit economics
Pricing is built around outcomes, not seats. 21 metered outcomes today (replies, try-ons, listings, syncs, recovery attempts, more). Tier subscriptions ($0, $79, $349, $999) bundle a monthly quota; overage is pay-as-you-go at meter rates. Result: revenue scales with merchant value, not user-count games.
5. Why per-outcome billing
Seat-based AI tooling penalizes the customer for using AI more — the exact opposite of what good incentives look like. Per-outcome billing aligns price with delivered value, makes upgrade paths obvious, and gives finance teams a clean line item per outcome type. It also lets us underprice incumbents on simple use cases without subsidizing edge cases.
6. Defensibility
Three things compound. First, integrations breadth: Shopify, Instagram, TikTok today; Amazon next; each new channel multiplies merchant value without multiplying engineering cost (shared customer profile, shared knowledge base, shared agent runtime). Second, the workflow layer: Temporal-native long-running jobs are hard to retrofit; competitors built on serverless-only stacks hit timeout walls at scale. Third, the customer data graph: every conversation, every try-on, every listing teaches the tenant's agent — outputs improve weekly, not monthly. None of this is a moat by itself; together they create switching cost.
7. Status
- live Shopify sync · Instagram bots · TikTok bots · Try-on · Product generation · Per-outcome billing · Stripe
- near Amazon listing generation · Storefront widget v2 · Self-host bundle
- planned Cart recovery agents · Knowledge agent · Multi-language
8. Where we are
All six pillars shipped. Per-outcome billing is wired end-to-end through Stripe with 21 active meters. Selected design partners are stress-testing in production against real catalogs ahead of public launch; the waitlist is open and prioritized for 1000+ SKU operators. We're heads-down on design-partner GA — first paid charges land at launch, and throughput numbers get published once design-partner production data validates them. Pre-revenue today, scaling on outcomes from GA forward.
9. Team
Operating from Toronto under Up Go Corp., a product studio that ships commerce, dev tools, and AI infrastructure. Current shape: ~5 engineering, 2 AI/ML, 1 design, 2 go-to-market. Lean by design. The full breakdown lives on about.
legal entity: Up Go Corp. (Canada) · open roles: [email protected]
10. The ask
Raising a seed round to take Selify from design-partner GA to validated per-outcome margins at production volume. Use of funds is concentrated on three things: operator-led GTM into Shopify Plus apparel, platform expansion (Amazon channel, storefront widget v2, self-host bundle), and engineering depth on the agent runtime. We're targeting partners who have lived inside a Shopify Plus operations team and understand why seat-based AI pricing is structurally wrong for outcome-driven workloads.
Specific round size, dilution, and use-of-funds breakdown live in the data room — request it: [email protected].