Where Delhivery's Missing Profit Went

Where Delhivery's Missing Profit Went

If you handle 55% more parcels than you did the same quarter last year, grow revenue by nearly 28%, and still watch your profit fall by roughly two-thirds, where did the money go?

That is the puzzle sitting inside Delhivery's Q1 FY27 results, released on 8 August 2026: express parcel shipments up to about 322 million, revenue from services at ₹2,931 crore, EBITDA at ₹156 crore and up only about 5%, and consolidated profit after tax at ₹32 crore, down about 65% from the year before. The company is doing more work, being paid more for it, and keeping much less. Some of that gap is the ordinary friction of higher fuel and labour bills. Some of it is the tail of Ecom Express integration, which management says has now concluded. The part I want to sit with is the part every Indian logistics deck this quarter is selling around, and none of them are yet costing out in the same breath.

Delhivery's own investor materials for the quarter name three AI initiatives shipped in the last twelve months: SmartNDR, an AI service that helped 500-odd e-commerce clients improve delivery success rates by up to 10%; Delhivery Maps, described as "an AI-native mapping suite for enterprises and developers"; and a first commissioned Automated Storage and Retrieval System at a client warehouse. None of those are bad ideas. Two of them may turn out to be real businesses. All three, this quarter, are cost lines waiting for a corresponding revenue line to arrive.

Now widen the frame.

On the pitch deck this week

Adani Ports announced in mid-June that it would commit up to $100 million with the American terminal-software firm Kaleris over the next several years, rolling AI-augmented container-handling tools across 15 terminals at nine ports at home and abroad. Kaleris tells the market it can lift Rubber Tyred Gantry crane productivity by up to 20% and terminal-truck productivity by up to 14%. Those are vendor figures on vendor slides, the sort of number a serious operator reads once and asks for the sample size on. They are also the kind of number that gets a board vote.

Just up the coast, JNPA (formerly JNPT) crossed 10 million TEUs in a single financial year for the first time, and its own promotional material puts a digital twin at the centre of the story: the port's control room using a live model of vessel traffic, terminal moves and truck appointments to make hour-by-hour operating decisions. Ekart, Flipkart's supply-chain arm, opened its pan-India network to outside merchants at the end of July, and pitched the move around AI-driven demand prediction and near-real-time inventory placement across more than 20 warehouse locations. Every one of those announcements came with a growth number attached.

I do not want to be a scold about it, because the growth numbers are not fictitious. Locus, the Bengaluru-founded route-optimisation firm acquired by IKEA's parent Ingka in October 2025, reports 1.5 billion optimised deliveries and about $320 million in savings for customers over its lifetime, with concrete lifts in on-time delivery rates from the 82–88% band into the 94–97% band, and fleet utilisation from 65–72% to 80–90%. Those are their numbers, not an audited third party's, and you should keep the same salt cellar handy that you keep for Kaleris's cranes. But Locus has been in production long enough at long enough client lists that the direction is fairly clear. Route optimisation delivers. It is the archetype for the small set of Indian logistics AI applications that pay for themselves.

What sits underneath the growth line

Now hold the two things in the same frame. A handful of applications — route optimisation, port-side crane orchestration, warehouse pick-and-place — reliably pay back. Everything else on the average pilot list, in most Indian shops I have any visibility into, does not.

The MIT Media Lab's NANDA initiative published a working paper in August 2025 called The GenAI Divide: State of AI in Business 2025, based on 52 executive interviews, 153 leader surveys, and analysis of 300 public deployments. Its headline finding — battered around badly enough in the year since to warrant the salt — was that only about 5% of enterprise generative-AI pilots produced measurable P&L impact. The other 95% did not. That figure is broad, it is generative-AI-specific, and it lumps in many experiments that were never going to move a line item on their own. It is also directionally consistent with what Gartner has been reporting inside supply-chain organisations, where 47% of respondents named low data quality as their top blocker to scaling AI and only 17% had moved past pilot into production.

Read those two paragraphs against a logistics P&L and the picture sharpens. The proven applications get counted twice. Their savings show up in the operations line. Their vendor invoice sits inside a technology line that also carries every experiment which has not yet paid back. And the experiments are the thing that has grown fastest in this cycle.

The slop debt, plainly named

Every Indian carrier I know of has, right now, more of one thing than it did two years ago: half-finished proofs of concept, evaluation harnesses no one runs, retrieval-augmented systems with no clear owner, prompt libraries that drift because no engineer is paid to prune them, and fine-tuned models that no one can quite reproduce. Some of it will graduate into production. Most of it will sit on the balance sheet in a spot the CFO's chart of accounts does not yet name. Call this the AI slop debt: the accumulating liability of pilots whose cost hit in Q3 and whose benefit was going to arrive in Q1, and then Q2, and then, well, "in the second half of next year." Under that name or another, it is coming for the margin.

The FinOps arithmetic is unforgiving in the specific case. Take SmartNDR at Delhivery's quarterly volumes. Assume, generously to the model, a per-shipment inference cost in the low single-digit paise range from a modern hosted API, running against 322 million parcels a quarter. You are in low-to-mid single-digit crores of direct model spend on that one product line, before engineering, before the data platform, before the observability stack that a fleet of production AI services actually needs, and before any of the pilots that never shipped. The 10% delivery-success uplift Delhivery is quoting for the 500-odd SmartNDR clients is, at their aggregate volumes, worth substantially more than that inference bill. That is the good news. The bad news is that SmartNDR is one product line of many, most of the others will not show a comparable uplift, and the composite bill is what actually lands on the P&L.

A fair reading of the other side

BCG published a note in early 2026 called AI Is Already Moving the Logistics Industry Forward. It is a serious piece, and its central claim, that a small number of AI applications in transport and warehousing are producing measured productivity gains, is one I would not push back on. Where I part company is the extrapolation. The presence of a working use case at one operator does not tell you the industry is moving forward on aggregate. It tells you the leaders can convert. On the ground in Indian logistics right now, the leaders are converting. Everyone else is running twelve pilots to catch three that pay, and paying for the other nine out of the same margin line the good ones just refilled.

What a dispatcher sees tonight

If you run a mid-tier carrier's control tower in Gurugram or Chennai or Mundra tonight, the interesting story in Delhivery's quarter is not the AI headline and it is not the volume headline. It is that the ceiling on margin has quietly lowered for the whole industry in the same period the vendor pitch decks have gotten louder. The volume growth is real. Competition on price is real. The technology bill has become a fixed cost you cannot easily walk back. Some of that spend will earn out; a larger part will not, and the arithmetic of which is which is not yet public information at any of the listed carriers.

The honest reading of Delhivery's quarter is that the company is still doing the fundamentals — 55% more parcels, revenue up, cash-flow positive, integration finishing — and paying, in real rupees this quarter, for the industrial reordering that will make the next five years different. The missing profit is not gone. It is being spent on the next generation of the business, and on a fair pile of stuff that will turn out not to be it. Which of those two piles turns out larger will still be the whole story of Indian logistics in 2028, and no one looking at this quarter can honestly tell you the ratio yet.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan, an Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.