Variant Perception

1. Where We Disagree With the Market

The sharpest disagreement is this: the consensus is paying for a productized-IP layer that has not shown up in the only metric that would prove it. Sell-side is unanimously Buy (8 of 9, average target ₹1,859 implying +37%, Nomura ₹2,220 at +64%) and is implicitly pricing eClerx against the India BPM peer set (Coforge 36×, LatentView 30×, Datamatics 19×) where 18× trailing looks cheap. But revenue per employee has been flat at ₹20.5 lakh for three years against 91% time-and-materials revenue, no quantified agentic-AI revenue line exists, and global peer Genpact separately discloses $1.2B (23.7% of revenue) as Advanced Tech Solutions. Strip out the productized-services premium baked into consensus targets and eClerx is a high-quality generalist KPO whose right anchor is Genpact at 10× — not Coforge at 36×. We are not contrarian on the asset; we are contrarian on the multiple denominator the market is using to value it.

A second, narrower disagreement sits inside the FY26 margin print. Consensus reads the H2 FY26 27% margin as evidence the 24–28% band is structural; management itself called Q2 FY26 28.8% margin 200 bps INR-aided unprompted. The variant read is that consensus is anchoring base-case FY27 EPS on a margin that includes a currency tailwind that may not repeat.

2. Variant Perception Scorecard

The score is moderate, not extreme. Consensus is clear and well-documented (eight Buys, Nomura just raised FY27 EPS 2.4%, MarketScreener notes opinion "improved significantly over the past four months") — that is what makes the disagreement audit-able, but it also means a lot of smart sell-side capital is on the other side. Evidence strength is medium-high because the load-bearing data points — flat revenue per employee, the Genpact AI revenue benchmark, the bunched analyst-target distribution, the management-unprompted INR call-out — are each documented in the upstream tabs. Resolution is concentrated in a narrow 6–9 month window: Q1 FY27 print (Aug 6), the FY26 annual report (Jul-Aug), and the first quantified AI revenue disclosure (target Q1-Q3 FY27).

3. Consensus Map

Two of these consensus views (margin band; bonus-issue mechanics) are well-grounded; we do not contest them. Two more (top-10 trajectory; FCC tail) are observation-based but rest on a single interpretation of an ambiguous signal. The most monetisable disagreement sits where consensus is highest — the multiple anchor and the productized-IP credit baked into it.

4. The Disagreement Ledger

Disagreement #1 — the multiple anchor. Consensus would say: eClerx earns 25.5% Op EBITDA and 35% ROCE versus India peers averaging 17-20% margins and 17-23% ROCE; that earns the specialist multiple, and 18× trailing is cheap because 22-25× is the historical mean. Our evidence disagrees because the specialist label has not shown up in revenue per employee for three years, the 91% time-and-materials book is not different in mix from generalist BPM, and the global pure-play peer Genpact discloses $1.2B AI revenue separately while eClerx discloses zero. If the productized-IP layer fails to print as quantified revenue, the multiple should converge to Genpact's 10× over an AI-deflation cycle. The market would have to concede that the "specialist" frame is supported by historical returns (true) rather than by a measurable IP layer (unproven). The cleanest disconfirming signal is the first quantified outcome-priced revenue line ≥5% of group revenue.

Disagreement #2 — the top-10 trajectory. Consensus would say: top-10 falling from 64% to 59% during +22% growth is textbook diversification. Our evidence disagrees because the same number is equally consistent with quiet wallet-share migration — Capgemini's July 2025 acquisition of WNS placed a Tier-1 integrator directly inside the BFSI niche, the CEO refused to engage on the Q1 FY26 call ("I wouldn't want to comment on the competition"), and Personiv's largest client repeated the FY16-style $8M loss pattern in FY24. The market would have to concede that the headline trajectory is consistent with two opposite stories and the CEO's silence on the competitive question is informative. The disconfirming signal is two consecutive quarters of top-10 ticking back above 62% or a named Tier-1 bank disclosing a Capgemini-WNS bundled win.

Disagreement #3 — the FX-cleaned margin. Consensus would say: FY26 25.5% operating EBITDA with H2 at 27% is the new sustainable band. Our evidence disagrees because management itself called Q2 FY26 28.8% margin "200 bps INR-aided" unprompted — the cleanest possible self-disclosure that part of the recovery is currency. With ~85% USD-linked revenue and the FY27 hedge book pinned at ₹89/USD, an INR rally or even normalisation strips ~80 bps from structural margin. The market would have to concede that the run-rate is closer to 24-25% than 26-27% — putting consensus at the band floor, not the band middle. The disconfirming signal is the Q1 FY27 print on Aug 6: margin held above 25% on a wage-cycle low quarter with normalised INR would refute the variant; a print below 24% confirms it.

Disagreement #4 — implementation drag. Consensus does not address this because consensus is sell-side, which has no implementation constraint. Our evidence is that 20-day ADV is ₹3.13 Cr against a ₹12,706 Cr market cap, the promoter block is structurally non-lendable, and the stock is not on the NSE F&O list. The bull setup may be analytically right and still produce no marginal buyer at size — capping the multiple-rerate path that ₹1,859 average targets imply. The disconfirming signal is sustained ADV doubling above ₹6 Cr per day for three months post-bonus or F&O inclusion.

5. Evidence That Changes the Odds

The strongest single piece of evidence is the Genpact disclosure benchmark juxtaposed with eClerx's silence on AI revenue. The most fragile is the FX call-out — useful as a disclosure signal, but uncertain because INR could continue weakening and mask the underlying margin question. The most durable is the flat revenue per employee: it is the metric on which consensus's bull case and our variant case will both be settled, regardless of how INR or the multiple debate evolves.

6. How This Gets Resolved

7. What Would Make Us Wrong

The fairest version of the bull case is that we are wrong on three points at once.

First, consensus may be right that revenue per employee is the wrong metric. A genuine productized-services pivot would show up first in segment-margin disclosure and outcome-priced contracts, not in the blended revenue/headcount ratio. The unified AI organisation under John Flowers (announced May 20, 2026), the Q4 FY26 large-scale agentic-AI contract close, and the absence of any forensic flag in the unbilled work suggest the IP layer is being built — it is just not yet at scale. If the Q1 FY27 call discloses a productized-services revenue line above 5% of group revenue, our central variant collapses fast, and the India peer-anchor multiple is the right denominator after all. The cleanest concession would be that we under-weighted management's stage-setting and over-weighted the absence of disclosure.

Second, the top-10 concentration trajectory could simply be diversification, full stop. ACV +23% YoY to $170M, with quarterly bookings sustained at $40-50M, is hard to reconcile with quiet wallet-share leakage at the top end. If Q1 and Q2 FY27 print top-10 continuing to drift toward mid-50s while ACV holds, the wallet-share-leakage variant fails the data — and the CEO's Q1 FY26 non-answer was simply CEO discipline about not commenting on a specific competitor. We would have over-read silence as disclosure.

Third, the FX-cleaned margin debate may be a non-issue if the INR continues to weaken. Most India IT and BPM names have benefited from secular INR depreciation for two decades; treating the FX tailwind as transitory rather than structural may itself be the wrong frame. If Q1 FY27 prints above 25% margin on continued INR weakness, the disagreement still holds technically but loses operational relevance for the next 12 months.

The implementation drag is the variant we are most confident about because it does not require any forecast — ADV, F&O listing, and promoter lock are observed states, not predictions. The risk is that the post-bonus float expansion organically doubles liquidity over the next 6-9 months and the F&O committee adds eClerx to the permitted list. Either would weaken this point.

The first thing to watch is the Q1 FY27 print on August 6, 2026 — specifically whether the earnings call discloses a quantified outcome-priced or agentic-AI revenue line. The margin number resolves the FX sub-debate; the disclosure (or its absence) resolves the multiple-anchor debate that drives every other disagreement on this page.