Table of Contents
Augmented Reality has moved past the gimmick phase. IKEA Place has processed over 2 million furniture placements. Google Lens handles 15 billion visual searches monthly. Snap's AR commerce features drive $100M+ in annual try-on interactions. But for every success story, a hundred AR projects launched with impressive demos and died when users didn't come back after the first session.
At Pillai Infotech, we've built AR solutions for retail product visualization, industrial maintenance workflows, and educational content delivery. The pattern is clear: AR succeeds when it solves a real problem faster than the non-AR alternative. Not when it's "cool." This guide covers what works, what doesn't, and how to build AR that people actually use.
1. AR Development Platforms Compared
The AR platform landscape has consolidated around three major ecosystems. Your choice depends on target audience, feature requirements, and whether you need native performance or web accessibility.
| Platform | Ecosystem | Key Capabilities | Device Coverage | Best For |
|---|---|---|---|---|
| ARKit (Apple) | iOS/iPadOS, visionOS | LiDAR scanning, object occlusion, body tracking, room mapping | iPhone 12+, iPad Pro, Apple Vision Pro | Premium experiences, retail, LiDAR-dependent features |
| ARCore (Google) | Android, Web (via Scene Viewer) | Environmental understanding, light estimation, depth API, Geospatial API | 250M+ Android devices supported | Mass-market Android reach, location-based AR |
| WebXR | Browser-based (Chrome, Safari, Edge) | Hit testing, plane detection, anchors, hand tracking (limited) | Any modern mobile browser | No-install AR, marketing campaigns, product visualization |
| Unity + AR Foundation | Cross-platform | Wraps ARKit + ARCore, plus custom features | iOS + Android + headsets | Complex AR apps, games, cross-platform from single codebase |
| Snap AR (Lens Studio) | Snapchat, Camera Kit | Face/hand/body tracking, world effects, try-on | Snapchat users + Camera Kit in any app | Social AR, brand campaigns, face filters, try-on |
Platform Selection Decision Tree
Need maximum reach with zero friction? WebXR. Users tap a link, AR loads in browser. No app download, no App Store approval, no update cycle. Conversion rates are 3-5x higher than "download our app" flows. Need advanced features (LiDAR, body tracking, persistent anchors)? Native with ARKit/ARCore via Unity AR Foundation. LiDAR-based room scanning and mesh generation are ARKit-only features that enable furniture placement and interior design use cases. Need location-based AR at scale? ARCore's Geospatial API (powered by Google Street View data) provides centimeter-accurate positioning in 100+ countries. ARKit has no equivalent. Need social/viral distribution? Snap AR or Instagram AR (Meta Spark). Built-in sharing and discovery. But you're locked into their ecosystem.
2. WebXR — AR Without the App Store
WebXR is the most underestimated AR platform. While native AR gets the headlines, WebXR quietly powers the majority of commercial AR interactions — product visualization, try-on, and marketing experiences where "download our app" is a conversion killer.
Why WebXR Wins for Commerce
The math is simple. A native AR app has these friction points: see ad → visit App Store → download (50-200MB) → open app → onboard → find feature → use AR. Each step loses 40-60% of users. Total conversion from ad to AR experience: 2-5%. WebXR: see ad → tap link → AR loads in 3-5 seconds. Total conversion: 15-30%. For e-commerce product visualization, that 5x conversion improvement is worth millions.
WebXR Technical Stack
Three.js + WebXR API is the standard stack. Three.js handles 3D rendering; WebXR API provides device pose tracking, hit testing (placing objects on real surfaces), and light estimation. For simpler use cases, model-viewer (Google's web component) lets you embed AR product views with a single HTML tag — no JavaScript required. For complex interactions, use A-Frame (Mozilla's web framework built on Three.js) or Babylon.js (Microsoft-backed, better TypeScript support).
3D model optimization is critical. Mobile browsers have strict memory limits. Target: GLTF/GLB format, under 5MB per model, under 50K triangles, compressed textures (KTX2 with Basis Universal). Use Draco compression for geometry (60-80% size reduction). A furniture model at 500K triangles will crash Safari on older iPhones. The same model decimated to 30K triangles with baked lighting looks nearly identical and loads in 2 seconds.
3. Retail and E-Commerce AR
Retail AR has the clearest ROI in the AR ecosystem. Shopify reports 94% higher conversion rates for products with 3D/AR views. Wayfair reports 3.4x higher conversion and 50% fewer returns for furniture purchases made with AR visualization.
Product Visualization (Try Before You Buy)
The killer use case: place a product in your real space before purchasing. Works for furniture, home decor, appliances, electronics (see TV size on your wall), and fashion accessories. The technical pipeline: Product photography or CAD model → 3D artist creates optimized model → Texture baking and PBR materials → GLTF export with multiple LODs (Level of Detail) → Integration with e-commerce platform via model-viewer or native SDK.
Pillai Infotech case study: We built an AR product viewer for an Indian furniture brand (120 SKUs). Implementation: WebXR via model-viewer for zero-friction access from product pages. Each furniture model was created from manufacturer CAD files, optimized to under 3MB. Results after 6 months: 28% higher average order value when AR was used, 31% reduction in returns (customers knew the actual size), and 4.2x more time spent on product pages with AR enabled. The 3D model creation was the biggest investment — Rs 3,000-8,000 per SKU depending on complexity.
Virtual Try-On
Face AR (sunglasses, makeup, jewelry) is mature and commercially proven. Body AR (clothing) is still emerging. Face tracking accuracy is excellent on modern phones (60+ landmarks, real-time). Body tracking for clothing requires either depth sensors (LiDAR) or sophisticated pose estimation ML models. Current body try-on solutions work well for loose-fitting garments but struggle with fitted clothing, draping, and fabric physics.
| Try-On Type | Maturity | Accuracy | Platform Options | Cost per SKU |
|---|---|---|---|---|
| Eyewear | Production-ready | 95%+ fit accuracy | ARKit, ARCore, Snap, WebXR | Rs 2,000-5,000 |
| Makeup/cosmetics | Production-ready | 90%+ color accuracy | ARKit, ARCore, Snap, ModiFace | Rs 500-2,000 |
| Jewelry | Production-ready | 85-90% (reflections are hard) | ARKit, ARCore, WebXR | Rs 3,000-8,000 |
| Footwear | Early production | 80-85% (foot tracking improving) | Snap, Wanna, custom native | Rs 5,000-12,000 |
| Clothing | Emerging | 60-75% (fabric physics limited) | Custom native + ML | Rs 15,000-40,000 |
4. Industrial and Maintenance AR
Industrial AR delivers the highest per-user ROI in the AR ecosystem. Boeing reported 25% faster wiring harness assembly with AR guides. Porsche reduced diagnostic time by 40% with AR-assisted technician workflows. The value proposition: reduce errors, speed up procedures, and capture expert knowledge for less-experienced workers.
AR-Guided Maintenance Workflows
The standard industrial AR workflow: Technician scans equipment (QR code or visual recognition). AR overlay shows step-by-step instructions anchored to the physical equipment. Each step highlights the specific component, bolt, or connection point. Sensor data overlays show live readings (temperature, pressure, vibration) without switching to a separate monitoring app. Completion of each step is logged for compliance and audit trails. Remote expert can join via video call with shared AR annotations.
Pillai Infotech case study: We developed an AR maintenance guide for a pharmaceutical plant's packaging line (12 machine types, 150+ maintenance procedures). Technicians use iPads with ARKit to scan machines and receive step-by-step visual instructions overlaid on the equipment. Results: 35% reduction in mean time to repair (MTTR), 60% reduction in procedure errors by junior technicians, and the system captured 2,000+ expert knowledge entries that would otherwise retire with senior staff. The biggest challenge wasn't the AR technology — it was digitizing 150 paper-based maintenance procedures into structured, AR-compatible formats.
Remote Assistance
AR remote assistance connects a field technician's camera feed with a remote expert who can draw annotations anchored to the real world. The field tech sees the expert's markings overlaid on their environment. This is particularly valuable for: specialized equipment where the expert is in another city, reducing travel costs for service calls (one expert supports 10-15 field techs instead of traveling to each site), and knowledge transfer from senior to junior staff.
5. Education and Training AR
Where AR Genuinely Improves Learning
AR in education works when the subject matter is inherently spatial or when physical materials are expensive, dangerous, or unavailable. A meta-analysis of 87 studies found AR improved learning outcomes by 0.68 standard deviations — significant, but only when used for spatial understanding tasks.
Strong use cases: Anatomy (explore 3D organs without cadavers), chemistry (visualize molecular structures and reactions), engineering (exploded views of machinery), geography (terrain and geological formation visualization), and vocational training (practice procedures on AR equipment before touching real machinery).
Weak use cases: Reading comprehension (AR adds distraction, not value), math (equations don't benefit from 3D), history (timeline AR is novel but not more effective than traditional teaching), and any subject where the AR is decorative rather than functional.
Building Effective Educational AR
The technical approach: use WebXR for accessibility (no app install requirement means teachers can share AR links directly), design for 3-5 minute interactions (attention spans in educational contexts are shorter), include assessment checkpoints within the AR experience (not just visualization), support offline mode (schools may have unreliable connectivity), and ensure the 3D models are scientifically accurate (partner with subject matter experts, not just 3D artists).
6. Technical Deep Dive: Building AR Features
Plane Detection and Object Placement
The foundation of most AR experiences is placing virtual objects on real surfaces. Both ARKit and ARCore provide plane detection — identifying horizontal (floors, tables) and vertical (walls) surfaces. The quality varies significantly by device. LiDAR-equipped devices (iPhone 12 Pro+) detect planes instantly with millimeter accuracy. Non-LiDAR devices need 2-5 seconds of scanning and achieve centimeter accuracy. The UX implication: always show a visual scanning indicator and don't let users place objects until plane confidence is high enough.
Light Estimation and Rendering
Making virtual objects look "real" requires matching the lighting of the physical environment. ARKit and ARCore provide ambient light intensity and color temperature. ARKit 6+ adds HDR environment probes (directional lighting). For product visualization, the difference between basic light estimation (flat, fake-looking) and HDR environment mapping (realistic shadows and reflections) is the difference between "this looks like a game" and "is that real?" Invest in PBR (Physically Based Rendering) materials and environment-mapped reflections for commercial AR.
Performance Optimization
AR apps must maintain 60fps while running camera processing, ML models (plane detection, tracking), and 3D rendering simultaneously. Budget: camera + AR tracking takes ~30% of GPU. That leaves 70% for your 3D rendering. Optimization rules: use instanced rendering for repeated objects, implement LOD (swap to simpler models at distance), keep draw calls under 100, use compressed textures (ASTC on mobile), and profile on the lowest-spec target device, not your development phone.
7. AR in India: Market Reality
Device Landscape
India's AR development must account for the device reality. Only ~15% of Indian smartphones have ARCore support with depth API. The majority run on budget Android devices (Snapdragon 4-series, MediaTek Dimensity 700) that support basic ARCore but struggle with complex AR scenes. LiDAR is limited to iPhone Pro models — less than 3% of Indian smartphones. WebXR works on ~70% of Indian smartphones (Chrome 90+ on Android, Safari 15+ on iOS).
The implication: Design AR for mid-range Android first. Test on Redmi Note series and Samsung Galaxy A series — they represent 40%+ of Indian smartphone sales. If your AR experience stutters on a Redmi Note 12, it won't work for most of your Indian users. Use progressive enhancement: basic AR for all devices, advanced features (occlusion, LiDAR scanning) for capable devices.
India AR Use Cases with Traction
| Use Case | Indian Companies | Market Size | Status |
|---|---|---|---|
| Jewelry try-on | Tanishq, CaratLane, BlueStone | Rs 5,00,000 Cr jewelry market | Production, driving conversions |
| Furniture visualization | Pepperfry, Urban Ladder, HomeLane | Rs 40,000 Cr furniture market | Growing adoption |
| Real estate walkthroughs | NoBroker, Housing.com, 99acres | Rs 12,00,000 Cr real estate | Early production (mostly 360° today) |
| Education | Byju's, Unacademy, Physics Wallah | Rs 10,000 Cr EdTech market | Experimental features |
| Beauty try-on | Nykaa, MyGlamm, Lakme | Rs 15,000 Cr beauty market | Production on Nykaa, growing |
Cost and Timeline for India
Realistic budgets for AR development in India: WebXR product viewer (50 SKUs with 3D models): Rs 8-15 lakhs, 8-12 weeks. Native AR try-on (face-based, 100 SKUs): Rs 15-25 lakhs, 12-16 weeks. Industrial AR maintenance (10 machine types, 50 procedures): Rs 20-40 lakhs, 16-24 weeks. AR educational app (single subject, 20 modules): Rs 12-20 lakhs, 12-20 weeks. The biggest cost driver is 3D content creation, not engineering. Budget 40-60% of total cost for 3D models, textures, and optimization.
Frequently Asked Questions
Should we build a native AR app or use WebXR for our product visualization?
Start with WebXR for product visualization — it covers 80% of use cases with zero install friction. WebXR via model-viewer lets customers see products in AR directly from your website with a single tap. Conversion rates are 3-5x higher than asking users to download an app. Go native only if you need LiDAR-based room scanning (interior design), advanced body tracking (clothing try-on), or persistent AR anchors (multi-session experiences). For Indian markets specifically, WebXR is even more important because users are reluctant to download apps for one-time shopping experiences, and storage space on budget phones is limited. A hybrid approach works well: WebXR for first-time product exploration, then push users to a native app only after they've had a positive AR experience and see ongoing value.
How much does it cost to create 3D models for AR product visualization?
3D model costs vary dramatically by complexity. Simple objects (bottles, boxes, basic electronics): Rs 2,000-5,000 per model from CAD or reference photos. Medium complexity (furniture, shoes, bags): Rs 5,000-15,000 per model. High complexity (jewelry with gem reflections, fabric with realistic draping): Rs 15,000-40,000 per model. For scale (100+ SKUs), photogrammetry pipelines (automated 3D capture from photos) reduce per-unit cost by 60-70% after initial setup (Rs 5-10 lakhs for the capture rig and software). At Pillai Infotech, we've helped brands set up in-house photogrammetry — a turntable, 3 DSLR cameras, and processing software — that produces AR-ready models in 2 hours per SKU after training. The pipeline pays for itself at around 50 SKUs. Always budget for optimization after creation: raw 3D models need polygon reduction, texture compression, and AR-specific LOD variants.
Will AR work well on the budget Android phones that most Indian customers use?
Basic AR works on most Android phones sold after 2020 (Snapdragon 600-series and above, MediaTek Dimensity 700+). ARCore supports 250M+ Android devices in India. However, performance varies significantly. On a Redmi Note 12 or Samsung Galaxy A14, expect: plane detection in 3-5 seconds (vs instant on flagships), 30-40fps for simple scenes (vs 60fps on flagships), and occasional tracking jitter in low light. Design for this reality: keep 3D scenes under 30K triangles and 3 texture maps, implement progressive loading (show low-poly first, then enhance), provide a "2D fallback" for devices that can't run AR (still show the 3D model in a viewer without camera), and test on Rs 10,000-15,000 phones, not just developer devices. WebXR is more forgiving than native AR on budget devices because the browser handles resource management. For Indian deployment, always have a non-AR alternative — even a 360-degree product spin — so no user is left with a blank screen.