THE PRE-IPO SCORECARD: RATING ZEPTO'S AI READINESS BEFORE THE DRHP

Eight categories, a 40-point card, and the parts of the quick-commerce story that AI either rescues or wrecks before the bankers file.

Quick-commerce in India is now a five-year-old category and a two-horse race. Blinkit sits inside a listed parent. Zepto is the next big standalone DRHP that the market has priced into its calendar. By the time the red herring lands, the AI line items in that prospectus will be read more carefully than the EBITDA bridge. This is a pre-DRHP scorecard — eight categories, five points each, 40 in total — to answer one question: how AI-ready is Zepto when the bankers file?

This is not a buy or sell view. Zepto is unlisted. The intent is to give the operator and the public-market investor a vocabulary for the AI parts of the story before the lock-up of the DRHP makes that vocabulary harder to apply.

The Card

━━━━━━━━━━━━━━━━━━━━

▸ 1. Demand forecasting & dark-store assortment — 4 / 5

▸ 2. Pick-pack-dispatch optimisation — 4 / 5

▸ 3. Personalisation & search relevance — 3 / 5

▸ 4. Pricing & margin engine — 3 / 5

▸ 5. Customer support automation — 3 / 5

▸ 6. Fraud, abuse, and rider safety — 2 / 5

▸ 7. Data infrastructure & MLOps — 3 / 5

▸ 8. AI governance, DPDP posture, and disclosure quality — 2 / 5

TOTAL: 24 / 40. A respectable B-grade.

The number itself matters less than where the points are missing — and where the company is asked to defend the gaps in the DRHP.

━━━━━━━━━━━━━━━━━━━━

Category 1 — Demand Forecasting (4/5):

The hardest problem in quick-commerce is not delivery. It is inventory. Every dark store carries 3,000–4,000 SKUs in 2,000–4,000 sq ft and must turn each one in days, not weeks, without stocking out on the 200 items that drive 70% of trips. Zepto's per-cluster forecasting model retrains nightly and is sensitive to weather, festivals, and local micro-events. This is the right architecture for India. The lost point: forecast error vs. baseline is not publicly disclosed.

Category 2 — Pick-Pack-Dispatch (4/5):

The 10-minute promise is an optimisation problem stacked on a routing problem stacked on a workforce problem. Public claims around pick paths and dispatch hold against industry benchmarks. The lost point sits on rider-side ETA prediction conditioned on rider profile, weather, and order weight — where Blinkit has a longer data history.

Category 3 — Personalisation & Search (3/5):

LLM-assisted search shipped in 2025. Homepage personalisation is visibly better than it was 18 months ago. But the moat is the data flywheel — order frequency × basket diversity × users — and Zepto is still building that vs. Blinkit's two-year lead. The DRHP will be read for repeat-rate and basket-size trajectory.

Category 4 — Pricing & Margin Engine (3/5):

Dynamic pricing is live on the long tail. The unanswered question: how much of the take-rate lift is the model and how much is mix shift toward private labels. Public-market investors will want this disaggregated.

Category 5 — Customer Support Automation (3/5):

The tier-1 agent is live. Hindi, Tamil, Marathi, Telugu rollouts in progress. Cost-to-serve trajectory in the DRHP will be the proof. The point gap is on tier-2 and refund automation, where Indian quick-commerce contact reasons are dominated by quality, missing items, and delivery experience.

Category 6 — Fraud, Abuse, Rider Safety (2/5):

The lowest-scoring category for a reason. Returns and promo abuse are category-wide problems. The rider-onboarding KYC stack is where deepfake risk now sits — synthetic video onboarding with a real-world handler operating multiple accounts. The DRHP should disclose what fraction of rider onboards are flagged for additional liveness, and the trend line.

Category 7 — Data Infra & MLOps (3/5):

The warehouse stack is modern. The point gap is on MLOps governance — model versioning, A/B test infrastructure, drift detection, rollback playbooks. None of this is sexy. All of it is the difference between a public company that ships AI changes safely and one that issues a clarification a quarter later.

Category 8 — AI Governance & Disclosure (2/5):

This is the category that will swing the most before the DRHP. As of today: DPDP consent and breach-notification posture is in place. What is not externally visible is an AI governance framework — model inventory, board oversight, third-party AI vendor risk. The MeitY draft AI advisory and steady RBI/SEBI signalling on algorithmic accountability point to a near-term DRHP norm: a paragraph on AI governance, by name, in risk factors. Most quick-commerce DRHPs globally have not shipped this — that gap becomes a discount.

What to Ask in the Roadshow

━━━━━━━━━━━━━━━━━━━━

Three questions are enough.

▸ What is your forecast error vs. a naive baseline, by store, by category, over the last 12 months?

▸ Which AI-driven decisions, today, are not behind a human override? How many run per day?

▸ Who on the board owns AI governance, and what was your last AI incident?

A management team that can answer in 5 minutes has a 30/40 story. A team that needs the slide pack has a 20/40 story. The market is no longer paying for the difference in promise — it is paying for the difference in operating maturity.

The DRHP Filter

━━━━━━━━━━━━━━━━━━━━

When the DRHP lands, read the AI sections through this scorecard. Categories 1, 2, 3 will read well — customer-facing wins that get press time. The categories worth the second read are 6, 7, and 8 — fraud, MLOps, and governance. They are where IPO-grade operating discipline is either visible or absent. The AI premium has already been priced into private rounds. The DRHP is where it gets re-underwritten by a much harder room.

QUICK-COMMERCE WILL NOT BE VALUED FOR ITS AI PROMISE IN THE DRHP — IT WILL BE VALUED FOR THE AI IT CAN SHIP SAFELY THE QUARTER AFTER LISTING.

━━━━━━━━━━━━━━━━━━━━

Related reading: Swiggy at +18 Months — siddheshj.com/blog/swiggy-plus-18-months and The FY27 Compliance Map — siddheshj.com/blog/fy27-compliance-map.

Sources: Zepto public engineering and product talks (2024–2025), founder media interviews, public job postings, MeitY draft AI advisory, RBI working group on ethical AI in financial services. All numbers and capability claims rely on public statements only.

Drafted with Claude assistance; edited by Siddhesh Joglekar.

#IPOWatch #QuickCommerce #AIinIndia #DRHP #IndianMarkets