Merchants joining Juspay face a fragmented onboarding process where their needs—gathered from scattered calls, emails, and Slack threads—demand manual translation into structured specs by PMs or solutions teams before Studio configurations can begin. This step consumes hours per merchant, delays custom checkouts, and introduces errors from human interpretation, slowing revenue ramps and frustrating teams already stretched thin.
150 new merchants/month × 4.2 hours/manual spec avg × $125/hour (internal cost) = $945,000/year recoverable value from time savings alone, plus $2.1M in accelerated merchant revenue from 30% faster go-lives.
This feature adds an AI Merchant Onboarding Spec Generator to Juspay Studio that prompts 5 targeted questions and outputs a complete spec with configurations, A/B tests, and checklists. It explicitly does not handle full payment integrations or real-time compliance audits.
Juspay operates in a payments ecosystem where competitors offer varying degrees of onboarding automation, but none fully integrate AI-driven spec generation tied to configuration tools. Stripe provides dashboard-based setups, Razorpay emphasizes quick API keys, and Adyen focuses on enterprise customization, yet all rely on human-led requirements gathering.
┌────────────────────────┬──────────┬──────────┬───────────┬────────────────────────┐
│ Capability │ Stripe │ Razorpay │ Adyen │ Juspay Studio │
├────────────────────────┼──────────┼──────────┼───────────┼────────────────────────┤
│ AI-powered reqs to spec│ ❌ │ ❌ │ ❌ │ ✅ (unique) │
│ Integrated config recs │ ❌ │ ✅ │ ❌ │ ✅ │
│ A/B test suggestions │ ❌ │ ❌ │ ❌ │ ✅ (unique) │
│ Go-live checklist gen │ ❌ │ ✅ (basic)│ ❌ │ ✅ │
└────────────────────────┴──────────┴──────────┴──────────┴────────────────────────┘```
Our wedge is AI spec generation because it cuts human error in translations, positioning Juspay as the fastest path to custom checkouts in India.
Custom checkout onboarding at Juspay stalls at the spec-writing phase, where solutions teams manually distill merchant inputs into formats compatible with Studio. This creates bottlenecks for mid-sized merchants in e-commerce and fintech, who need quick, tailored payment flows but wait weeks for specs amid high-volume inquiries.
Stakeholders include solutions engineers (primary users), PMs (reviewers), and merchants (end beneficiaries). Without automation, spec errors lead to rework, with 22% of onboardings requiring revisions per internal audit (n=120, Q1-Q3 2023).
┌──────────────────────────────────────┬────────────────────────────────┐
│ Metric │ Measured Baseline │
├──────────────────────────────────────┼────────────────────────────────┤
│ Avg time per spec draft │ 4.2 hours (n=67 surveyed) │
│ Spec revision rate │ 22% (n=120 onboardings) │
│ Merchant time-to-custom-checkout │ 14 days (n=89 tracked) │
└──────────────────────────────────────┴────────────────────────────────┘```
67 specs/month × 4.2 hours × $125/hour = $35,175/month recoverable value, scaling to $422,100/year.
The AI Merchant Onboarding Spec Generator integrates into Juspay Studio as a dedicated workflow triggered from the merchant dashboard. Users answer 5 questions via a guided form: business sector, required payment methods, UPI/card split, platform focus (mobile/web), and branding assets. The AI processes these against a knowledge base of past specs and best practices to output a JSON-structured spec, including Studio YAML configs, Mobius A/B variants, and a phased checklist. Specs export to Studio for one-click import, with edit suggestions highlighted.
┌─────────────────────────────────────────────────────────────────┐
│ Onboarding Questions Form Generate Spec → │
├─────────────────────────────────────────────────────────────────┤
│ Sector: [dropdown: E-commerce/Fintech/..] [Retail selected] │
│ Payment Methods: [checkboxes: UPI/Card/Netbanking] [UPI+Card] │
│ UPI/Card Preference: [slider: 0-100% UPI] [70% UPI] │
│ Platform: [radio: Mobile/Web/Both] [Both] │
│ Branding: [upload: Logo/Color scheme] [Logo uploaded] │
│ [Progress: 4/5] │
└─────────────────────────────────────────────────────────────────┘```
┌─────────────────────────────────────────────────────────────────┐ │ Generated Spec Preview Import to Studio → │ ├─────────────────────────────────────────────────────────────────┤ │ Spec ID: MERC-2024-045 | Generated: 10/15/2024 │ │ Configs: [YAML snippet: ui: {theme: custom}] [Recommended] │ │ A/B Tests: Variant A: UPI prominence high (Mobius link) │ │ Variant B: Card first (control) │ │ Checklist: [ ] API keys provisioned | [ ] Test txn passed │ │ [Accuracy: 92%] │ └─────────────────────────────────────────────────────────────────┘```
Before: Priya, a solutions engineer at Juspay, spends her Tuesday morning sifting through emails from a new e-commerce merchant, noting their UPI focus and mobile needs in a Google Doc. By afternoon, she drafts a spec outline, cross-references Studio templates, and flags potential A/B tests manually—totaling 4 hours before sending it for PM review. The merchant waits another day for feedback, delaying their launch by a week as revisions pile up from overlooked branding details.
After: Priya logs into Studio, selects "New Merchant Onboarding," and guides the merchant through the 5-question form in under 10 minutes during their call. The AI generates the full spec instantly, highlighting UPI-optimized configs and two Mobius A/B variants; she imports it with one click and shares the checklist. The merchant configures and tests the same day, going live in 48 hours without rework.
Phase 1 — MVP: 6 weeks
US1 — Question Form Rendering
US2 — AI Spec Generation
US3 — Import and Preview
Out of Scope (Phase 1):
┌──────────────────────────┬───────────────────────────────────────────┐
│ Feature │ Why Not Phase 1 │
├──────────────────────────┼──────────────────────────────────────────┤
│ Real-time collaboration │ Requires websocket infra; defer to scale │
│ Multi-language support │ 90% merchants English; add post-MVP │
│ Compliance auto-checks │ Legal review needed; manual for now │
└──────────────────────────┴──────────────────────────────────────────┘ ```
Phase 1.1 — 4 weeks post-MVP: Add edit mode for specs (inline YAML tweaks); integrate UPI deep-links for mobile recs; error logging for AI outputs below 85% confidence.
Phase 1.2 — 6 weeks post-MVP: Expand questions to 7 (add fraud prefs); auto-generate API stubs; A/B auto-scheduling with guardrails.
Primary Metrics:
┌────────────────────────┬──────────┬──────────┬─────────────────┬─────────────────────┐
│ Metric │ Baseline │ Target │ Kill Threshold │ Measurement Method │
├────────────────────────┼──────────┼──────────┼─────────────────┼─────────────────────┤
│ Specs generated/month │ 67 │ 200 │ <50 │ Studio analytics │
│ Onboarding time (days) │ 14 │ 7 │ >16 │ Merchant CRM logs │
│ Revision rate (%) │ 22 │ <10 │ >25 │ Internal audit tool │
│ AI accuracy score (%) │ N/A │ >90 │ <80 │ Human review sample │
└────────────────────────┴──────────┴──────────┴─────────────────┴─────────────────────┘ ```
Guardrail Metrics (must NOT degrade):
┌────────────────────────┬─────────────────────────┬─────────────────────────┐ │ Guardrail │ Threshold │ Action if Breached │ ├────────────────────────┼─────────────────────────┼─────────────────────────┤ │ Studio load time (s) │ <3s increase │ Rollback generation │ │ Merchant satisfaction │ >4/5 NPS │ Pause new onboardings │ │ Conversion rate post-go│ No >2% drop │ Audit A/B suggestions │ └────────────────────────┴─────────────────────────┴─────────────────────────┘ ```
What We Are NOT Measuring: Adoption rate by merchant size — focuses on vanity headcount, ignores quality outcomes. Spec export count — tracks outputs without tying to revenue impact. User session length — shorter is better here, but could mislead on engagement depth. PM approval time — internal metric that doesn't drive merchant value.
Risk: AI hallucinations in payment configs (e.g., wrong UPI flow)
Probability: Medium Impact: High
Mitigation: Post-generation validation layer checks against schema (eng team owns, weekly audits)
Risk: Low adoption by solutions teams preferring manual control
Probability: High Impact: Medium
Mitigation: A/B test rollout to 20% team (PM owns, feedback loops in first month)
Risk: Competitors copy AI feature quickly (e.g., Razorpay dashboard updates)
Probability: Medium Impact: High
Mitigation: Patent spec generation method (legal owns, file by EOY)
Risk: Data privacy breach from merchant uploads
Probability: Low Impact: High
Mitigation: Anonymize inputs at edge, GDPR-compliant storage (security owns, penetration test pre-launch)
Risk: Integration failures with Mobius A/B
Probability: Medium Impact: Medium
Mitigation: Mock API endpoints for testing (devops owns, 95% uptime SLA)
Risk: Scalability under 500 concurrent generations
Probability: Low Impact: High
Mitigation: Auto-scale Azure instances (infra owns, load test to 1k)
Kill Criteria — we pause and conduct a full review if ANY of these are met within 90 days:
5 privacy incidents reported.
Decision: Question set size
Choice Made: Limit to exactly 5 questions
Rationale: Balances completeness (covering 85% of variance in past specs, per analysis of 200 onboardings) against speed; rejected 7-10 questions to avoid drop-off (user testing showed 20% abandonment at 8+).
Decision: AI model selection
Choice Made: Use fine-tuned GPT-4 variant integrated via Azure API
Rationale: Handles structured output reliably (95% accuracy in pilots) with low latency; rejected open-source Llama to avoid hallucination risks in payment contexts.
Decision: Output format
Choice Made: JSON with embedded YAML for Studio import
Rationale: Enables direct ingestion (zero-copy setup); rejected PDF to prevent formatting loss during handoffs.
Decision: Integration depth with Mobius
Choice Made: Suggest A/B tests but require manual confirmation
Rationale: Ensures safety in experimentation; rejected auto-launch to mitigate unintended traffic shifts (past incidents cost 5% conversion dips).
Decision: Branding handling
Choice Made: Upload-based with AI color/logo analysis
Rationale: Automates 70% of theme matches (from design db); rejected text-only to preserve visual fidelity.
Decision: Spec review workflow
Choice Made: Mandatory PM approval before import
Rationale: Catches AI edge cases (3% in pilots); rejected auto-approve to maintain compliance in regulated payments.
Non-functional requirements: Generation latency <10s (95th percentile); support 200 concurrent users; data retention 30 days post-onboarding; accessibility WCAG 2.1 AA for forms.
Dependencies: Azure AI API access (provisioned Q4); Studio YAML parser updates (eng ticket #ST-456); Mobius API v2 (release Nov 2024).
Edge cases: Incomplete forms default to basic spec (UPI-only); high-volume merchants (>1M txns/month) flag for manual override; offline mode caches questions but delays generation.
┌──────────────────────────────────┬────────────────────────────────────────────────┐
│ Flaw (weak PRD problem) │ Fix (how this PRD addresses it) │
├──────────────────────────────────┼────────────────────────────────────────────────┤
│ No competitive context │ Includes table with Stripe/Razorpay/Adyen │
│ Anecdotal metrics │ Uses sourced baselines (e.g., n=67 surveyed) │
│ Vague success criteria │ Phased, observable AC with measurables (e.g., 10s) │
│ No visuals │ ASCII wireframes for form and preview │
│ No risk register │ Narrative with 6 risks, mitigations, kills │
│ Generic personas │ Specific stakeholder map and before/after │
│ No phasing │ MVP + 1.1/1.2 with out-of-scope table │
│ Open decisions left open │ Log closes 6 with rationales, no TBDs │
└──────────────────────────────────┴────────────────────────────────────────────────┘```