The Apps SDK Just Raised the Stakes: How Indian SaaS Should Re-Think AI Distribution in October 2025

Yesterday — October 6, 2025 — OpenAI's DevDay shipped the Apps SDK and announced ChatGPT is at 800 million weekly active users. Sam Altman's framing was direct: ChatGPT is now an OS for apps, and developers can ship inside the conversation. Spotify, Booking.com, Figma, Canva, Coursera, Zillow are live on day one. For Indian SaaS founders, this is the biggest distribution shift since the App Store. Your product page is no longer the funnel entry. This post is the 90-day plan we are now running with our SaaS clients to compete inside ChatGPT — and the pricing reset that comes with it. ## The answer in 60 words OpenAI's Apps SDK lets developers ship interactive apps inside ChatGPT, accessible to 800M weekly users. For Indian SaaS, this collapses three funnel stages — discovery, evaluation, first-action — into one conversation. The 90-day plan: ship an MCP-compliant Apps SDK build of your top use case, reposition pricing around in-app actions not seats, and rebuild your top-of-funnel content for AI conversational discovery instead of Google search. ## Why this matters now (October 7, 2025) The Apps SDK announcement is not a new chatbot integration. It is a distribution platform reset. VentureBeat's coverage framed it correctly: ChatGPT is becoming the new App Store, with the SDK letting developers build interactive applications that render inline in conversation. OpenAI's own announcement describes it as built on the Model Context Protocol (MCP) — the open standard already adopted by Anthropic, Cursor, and others — so the bet here is on MCP-compliant integration, not on a proprietary lock-in. For Indian SaaS, three numbers matter. ChatGPT has 100M+ weekly users in India (per OpenAI's February 2026 figure — confirming India as one of their largest markets). 42% of new SaaS launched in 2025 already build on OpenAI models — your competitors will be there. And the consumer surface is now conversational: a Bangalore SMB owner asking "find me a logistics tracker" inside ChatGPT will get an inline app from a vendor whose Apps SDK build is best-aligned with the query. If that is not your build, you do not exist in the answer. ## The funnel collapse — what changed Your old funnel: Google search → SERP listing → click → product page → signup → in-product trial → conversion. Five steps, four drop-off points. Industry-average click-through from a top-3 SERP listing is ~12%; product-page-to-signup conversion ~3%; trial-to-paid ~22%. Multiply: roughly 0.08% of search-intent eventually pays. Your Apps SDK funnel: ChatGPT query → inline app render → in-conversation action → upsell to standalone product. Two steps, one drop-off point. The discovery, evaluation, and first-action collapse into one conversation. The user never leaves ChatGPT to "evaluate" your offering — they evaluate it by using it inline. ## Pricing reset — seats are dead, actions are alive Your old pricing: ₹990/month per seat, 14-day trial. Worked when "user signed up" was a meaningful commitment. In an Apps SDK world, the user did not sign up. They asked a question. They got an answer. They paid for the answer (or did not). The pricing model that works for inline apps is per-action — ₹X per booking, ₹Y per design generated, ₹Z per transcription minute. Stripe and Razorpay both support this model natively; OpenAI's Apps SDK exposes payment hooks. The math changes dramatically:

Metric	Old SaaS funnel	Apps SDK funnel
CAC (₹)	2,400 (paid + content)	180 (SDK listing + low CPC promo)
Time to first action	4-7 days (signup + onboarding)	under 60 seconds (inline)
Pricing model	Subscription per seat	Per-action or freemium-action
ARR per user (year-1)	~₹11,880 (₹990 x 12)	~₹2,800 (action-priced) but 14x users
Funnel steps	5	2

The trade: lower per-user revenue, dramatically lower CAC, and a much wider funnel. Total revenue can be 2-4x higher in modeled scenarios — the cost is rebuilding your billing system around per-action pricing and your product around inline rendering. ## The 90-day plan we are running ## When Apps SDK is NOT the right move Three cases. If your product is enterprise-only with average deal sizes above ₹15 lakh per year, the Apps SDK is the wrong distribution channel — enterprise sales is still a salesperson-led motion, not a conversational one. If your product depends on deep workflow integration (CRMs, ERPs, custom dashboards), the inline-render constraint will frustrate users — they need a full canvas, not a chat panel. If your moat is your UI (design tools, video editors), the SDK's component library will flatten your visual differentiation; ship a conversational adjunct, not a replacement. The Apps SDK fits best for utility SaaS with quick-action use cases: scheduling, transcription, conversion utilities, simple data lookups, content generation. We have nine SaaS clients in our portfolio; we are recommending Apps SDK builds for four of them this quarter and explicitly recommending against it for the other five. ## Common mistakes (from the first 10 days of Apps SDK builds) Symptom: app gets discovered but no one completes the action. Cause: the inline app demands too much input ("upload your brand kit, fill 12 form fields"). Fix: design for the smallest possible action that delivers value. Iterate up. Symptom: your app rarely gets triggered for category queries. Cause: your manifest description is generic ("AI-powered productivity tool"). Fix: write the manifest description as the literal phrase a user would type. "Track shipments in real time" beats "Logistics intelligence platform." Symptom: user completes the action but never visits your standalone product. Cause: no upsell hook in the inline result. Fix: every action result should end with a "want this in your team workflow? open in [your product]" link. Symptom: pricing per-action looks great but margins crash. Cause: each action triggers expensive backend work (e.g., 3 LLM calls). Fix: cache aggressively. Pre-compute. Quantise the model to a smaller variant for the per-action call. ## Real example — what we are shipping for a Bangalore SaaS client A Bangalore SaaS client of ours runs a transcription product (₹699/month, 200 minutes included). Their user base is 1,400 paying customers. CAC was ₹1,800 from Google paid + content. Funnel conversion: 0.7% of organic search to paid. Our Apps SDK plan for them, week-by-week: - Week 1-2: identified "transcribe this audio file in under 60 seconds" as the inline use case. - Week 3-5: built MCP server wrapping their transcription API. Inline app accepts a file upload, returns a transcript, ends with "open the full transcript in [product] for editing and team sharing." - Week 6-7: shipped per-action pricing — ₹4 per minute transcribed, free first-3-minutes, no signup required. Razorpay handles billing. - Week 8-9: submitted to OpenAI's directory. Listed as "Audio Transcription" with Hindi + English support. We are not 90 days in yet, so the outcome data is partial. Week-1 telemetry: their app triggered on 0.4% of "transcribe audio" queries in ChatGPT (low — manifest description needs work). Action completion rate among triggered: 31% (high — the inline UX is clean). Of action-completers, 6% clicked through to the standalone product. Numbers will move as we iterate; the structural learning is that the Apps SDK funnel is faster but the optimisation surface is different. Your old SEO playbook does not transfer. ## What this means for your top-of-funnel content If you publish blog content for SEO, the Apps SDK changes the substrate. Buyers researching your category will increasingly do it in ChatGPT — and ChatGPT cites pages that look like answers to questions, not pages that look like marketing brochures. We covered the patterns in our 60-site Indian B2B audit: question-format H2s, original numbers, structured schema, monthly publishing. If your buyer's first interaction with your category is a ChatGPT query, your blog needs to (a) be cited in the answer, and (b) document your MCP server / Apps SDK app so the buyer can move directly into trial. The new top-of-funnel page is "Use [product] inside ChatGPT — here's the manifest URL." ## Pre-launch checklist for an Apps SDK build

• Single use case identified that delivers value in under 60 seconds
• MCP-compliant server built and deployed
• Inline UI components built using OpenAI's design system
• Per-action pricing wired (Stripe, Razorpay, or OpenAI payments)
• Free-action allowance set (1-3 free actions recommended)
• Manifest description written as the literal user-typed phrase
• Upsell hook on every result ("open in [product]")
• Telemetry: trigger rate, completion rate, upsell rate dashboards
• Standalone product page documents "use inside ChatGPT" with manifest URL
• Backend cost-per-action calculated and margin verified before launch

## A counter-take we hold The Apps SDK could plateau if OpenAI's directory becomes a SEO-style competition for visibility — same dynamics as the App Store, where most apps starve because the top-100 captures attention. Our hedge for clients: build the Apps SDK app and keep your standalone product strong. The SDK is a discovery layer, not a replacement for product. Treat it that way and the downside is bounded; treat it as your only channel and you are renting your business from OpenAI's algorithm. ## FAQ ### Is the Apps SDK only for US developers? No. OpenAI announced it as globally available on October 6, 2025. Submission process is open to any developer. The launch partners include Booking.com (Netherlands HQ) and Expedia (US) — global from day one. ### What is MCP and why does it matter? Model Context Protocol — an open standard for connecting AI models to tools and data, originally introduced by Anthropic and adopted by OpenAI for the Apps SDK. MCP-compliant means your integration works across multiple AI engines, not just ChatGPT. Anthropic, Cursor, Sourcegraph already use it. ### Can my existing API just plug in, or do I need to rewrite? Your existing REST API can be wrapped in a thin MCP server in 1-2 days. The bigger work is the inline UX and the per-action pricing model. Plan for 2-3 weeks of senior engineering for a single use case. ### What pricing does OpenAI take from Apps SDK monetisation? OpenAI has not published final commercial terms. Early indications suggest a revenue-share model on payments processed through their integration, with developer-direct billing also supported. Treat the commercial structure as evolving over the next 90 days. ### Does the Apps SDK work for B2B SaaS or only consumer? Both. B2B works for utility-grade use cases (transcription, scheduling, conversion utilities, data lookups). Enterprise SaaS with deep workflow integration is a poor fit — the inline render is too constrained. ### Will my existing SaaS subscription model die? Not immediately. The likely outcome is a two-tier model — per-action pricing for inline / first-touch users, subscription pricing for power users who upgrade for team features and integrations. Coexistence, not replacement. ### How does this affect my SEO strategy? The buyer-discovery surface shifts from Google to AI conversation. Optimise for being cited in ChatGPT answers (see our prior posts on schema and GEO patterns) and document your Apps SDK manifest prominently on your standalone product pages. For the founder perspective on why this matters specifically for Indian SaaS, our founder Vivek Singh writes on the same beat. We covered the broader generative-engine landscape in our zero-click GEO post, and our AI automation team ships these integrations for clients. As Hrishikesh, our CTO, points out — see his team page — MCP is the standard worth betting on, not any single AI engine. We documented similar inline-discovery work for TalkDrill's ChatGPT integration. Discussion on the launch is active in Hacker News threads and on r/OpenAI. Email contact@softechinfra.com with your product brief.