One platform for promotions, vouchers, loyalty and referral programs. Connect your stores, apps and POS, then let Claude-powered AI draft campaigns, pick the right offer for each customer, and explain the results in plain language.
API-firstMulti-tenantDeterministic rule engine
Promotions, vouchers, loyalty and referrals usually live in four tools that disagree about the customer. VyForge runs them on shared rules, shared budgets and one customer record, so a voucher knows the loyalty tier and a referral reward knows the budget cap.
Conditions, priorities, stacking order and budget caps, written once and applied identically at the till, online and in the app.
WHEN order.total ≥ 300,000₫ AND customer.tier IN (Silver, Gold) AND channel IN (app, pos) THEN discount 15% CAP 60,000₫ LIMIT 1 per customer · NO stacking · BUDGET 25,000,000₫
Bulk pools, partner distribution, reservation at checkout and reversal on refund.
Earn, burn, expiry and adjustments, auditable to the line. Tiers with their own rules.
Two-sided rewards released only when the referred order clears, with self-referral and device checks.
Every campaign carries a hard cap. Spend is tracked per redemption and paused automatically.
VyForge uses the Claude API where a model actually earns its place: drafting, ranking, explaining and answering. The rule engine and ledger stay deterministic, so every discount is still a decision you can audit.
The offer each customer is most likely to redeem, within the margin you allow.
Describe the goal in a sentence. Get a complete, rule-valid draft to review and launch.
Segments that explain themselves: who is lapsing, who only buys on discount, who would upgrade.
Suggested changes to depth, caps and timing, each with the expected effect on cost and uplift.
"Why didn't my code work" answered with the actual rule that blocked it, for staff and customers.
Ask about campaigns in plain language and get answers computed from the ledger, not guessed.
Which voucher campaign last month had the best uplift net of cost, and why?
Weekend bundle, 2-for-1 coffee had the best net result: +18.4% incremental revenue against a cost of 5.1% of basket. Redemptions clustered on Saturday mornings, and 61% came from customers who had not bought a bundle before.
Flash 20% storewide drove more redemptions but most went to customers who were already buying that week, so net uplift was only +2.3% at nearly three times the cost.
Suggestion: rerun the bundle with a 50,000₫ cap and a Saturday-only schedule. Want me to draft it?
No migration project. VyForge sits beside your commerce stack and takes over the promotion logic one program at a time.
Sync orders and customers through the API or webhooks. Point your POS, storefront and app at one validation endpoint.
Describe the goal. Review the AI draft, adjust the rules, set the cap, and launch across every channel at once.
Read uplift against a holdout, ask questions in plain language, and accept suggested changes with one click.
Every campaign reports incremental revenue against a holdout, cost per redemption and budget burn. The question is never "did people use it" but "was it worth it".
| Campaign | Program | Redemptions | Redemption rate | Incremental uplift | Cost / redemption | Budget used |
|---|---|---|---|---|---|---|
| Weekend bundle, 2-for-1 coffee | VOUCHER | 1,862 | 12.9% | +18.4% | 21,400₫ | 64% |
| Come back, we kept your table | PROMOTION | 311 | 12.9% | +9.7% | 38,900₫ | 41% |
| Bring a friend, both get 30k | REFERRAL | 540 | 6.1% | +11.2% | 60,000₫ | 77% |
| Flash 20% storewide | PROMOTION | 3,050 | 24.6% | +2.3% | 57,200₫ | 100% |
Sample data for illustration. Uplift is measured against a randomised holdout group per campaign.
No. The AI drafts campaigns and suggests changes. The deterministic rule engine applies the rules you approved, and every change goes through your approval flow and audit log.
Through a REST API and webhooks. Your POS, storefront and app call one validation endpoint and get the same answer. A sandbox tenant lets you test rules before launch.
Yes. VyForge is multi-tenant: each brand gets its own branding, limits, roles and budgets, with reporting across all of them.
Only what a task needs. Personal identifiers are stripped before any model call, context is isolated per tenant, and your data is not used to train models.
Each campaign keeps a randomised holdout group that does not receive the offer. Incremental revenue, cost per redemption and budget burn are reported against it.
Bring one real campaign you ran last quarter. We will load it, draft the AI version next to it, and walk through what the numbers would look like.
or write to contact@vyforge.com