Damage Math

Damage Math

DocuSign's drawdown took roughly $12.5B of enterprise value out of the stock — a 59% cut — while the near-term earnings line barely moved. FY2026 revenue and EPS beat the consensus that stood before the trigger, forward FCF and EPS estimates rose, and full-year revenue guidance was raised on the very quarter the stock fell 19%. With the numerator near zero, the entire re-rating is a judgment on long-run earning power under AI. Whether that judgment is right is the trial's question, ruled temporary at 0.63 — contested, judges spanning 0.42 to 0.68.

The numerator: how far the near-term numbers actually fell

The framework's canonical setup is Centene — consensus EPS cut roughly two-thirds, the stock down two-thirds. DocuSign is the opposite arithmetic. The triggering event, the 5 June 2025 first-quarter print, missed on one metric only: billings grew 4% to $740M, "slightly below our guidance range due to lower-than-expected early renewals," a change management attributed to its own sales-compensation redesign and called of "negligible impact on revenue" [1]. The stock fell 19.0% the next day. The profit-and-loss guidance went the other way: management raised full-year FY2026 revenue guidance to $3.151–3.163B from the $3.129–3.141B set in March, while trimming the billings midpoint about $15M for renewal timing [2] [3].

Sources: consensus levels from CapIQ estimates, vintage 2026-07-23 (data/sp/estimates.json, n=22–23 analysts); guidance change from the Q1 FY2026 and Q4 FY2025 earnings calls [4] [5].

Not one quarter of the drawdown carried a consensus cut. Every quarter from the trigger forward beat, and the beats were largest on the bottom line.

Note: the 1Q26 print (5 Jun 2025) also beat — revenue +2.0%, EPS +10.5% — the miss was billings, a forward-timing metric not in this consensus. Source: CapIQ actual-vs-consensus surprise history (data/sp/estimates.json, beat_miss).

Forward estimates did not just hold — they climbed, including straight through the February 2026 AI-driven selloff. Consensus FY2027 EPS rose 10.0% and FY2028 EPS rose 12.4% over the trailing six months; forward FCF consensus rises every year through FY2029.

Source: CapIQ estimate momentum, as-of dates 2026-01-24, 2026-04-24, 2026-07-23 (data/sp/estimates.json, momentum). Vintages begin after the June-2025 trigger; the pre-trigger anchor is the quarter-by-quarter beat record above.

The near-term hit to revenue, EPS, and FCF is, on the record, approximately zero — and if anything positive. That is the numerator of the whole tab, and it is empty.

The denominator: what the price took out

Against a numerator of zero, the price took out more than half the company. From the $106.99 peak of 6 December 2024 to $47.04 on 23 July 2026, the shares fell 56.0%; at the 23 February 2026 trough of $41.75 the decline was 61.0%.

Enterprise value = market cap − net cash. Market cap at peak ≈ $106.99 × 210.3M shares = $22.50B, less ~$1.15B net cash. Current market cap $9.84B (fit_features.market_cap, price 2026-07-23), less $1.02B net cash — $814.2M cash and short-term investments plus $209.9M long-term investments, no debt [6]. Peak/trough prices from fit_features.capitulation_gauge.

Stated side by side: consensus FY2026 EPS ended 1.6% higher than the pre-trigger consensus, and the market cap fell 56.3%. The earnings the market can see did not fall; the price it will pay for them did. Enterprise value shed roughly $12.5B while free cash flow rose from $920.3M (FY2025) to $1,058.6M (FY2026) [7]. The current enterprise value is 8.3× realized FY2026 free cash flow and 7.8× the $1,133M consensus expects for FY2027.

The NPV arithmetic, conservatively

Because the near-term line did not move, none of the $12.5B can be a discounted-cash haircut on the next year or two. All of it is a change in the assumed terminal earning power. A transparent perpetuity test isolates exactly how much long-run change the price is claiming.

The workings are deliberately simple and visible. Value of a growing perpetuity is V = FCF₁ / (r − g), discount rate r = 10% (a profitable, no-debt software business; the sensitivity to r is modest and noted below). Two free-cash bases are shown, because the framework and the market disagree on which one is real:

Reported FCF ≈ $1,133M — the consensus forward figure the sell side and the 8× multiple use.

Adjusted FCF ≈ $437M — reported FY2026 FCF of $1,058.6M less $622.3M of stock-based compensation [8]. This is closer to the owner-earnings basis the framework insists on; the full adjusted-yield computation is owned by the Yield tab.

Derived: V = FCF₁ / (r − g), r = 10%. Reproducible from the two FCF bases above. Current enterprise value ≈ $8.8B; peak ≈ $21.4B. At r = 9% every reported-FCF figure rises ~15–25% and at r = 11% falls a similar amount; the ordering versus current EV is unchanged.

The grid reads two ways, and the honest answer is that it depends on which column you trust:

The gap, in numbers. If the impairment is temporary — earning power intact, the business stabilizes at a modest +4% terminal on reported cash — fair EV is ~$18.9B against ~$8.8B today: the price destroyed on the order of $10B (≈53%) more value than a temporary reading supports. If the impairment is permanent — AI erodes the base toward a −3.5% terminal — the current $8.8B is roughly fair and the gap closes to about zero. And on the framework's SBC-adjusted basis, the gap is absent under either scenario. So the mispricing Ruchir hunts is present only if both conditions hold: the damage is temporary and one values on reported rather than adjusted cash. The first condition is the trial's; the second the framework decides against. This is not the clean Centene gap, and it should not be dressed as one.

The trial: temporary or permanent

The temporary-versus-permanent question was argued by two opposing, corpus-cited briefs and ruled on by three blind judges. Both cases are strong; neither is a straw man.

The case for temporary

The move is multiple compression, not earnings destruction. The trigger was a timing metric, not a cash flow: early-renewal pull-forward is a working-capital phenomenon that annualizes away in one renewal cycle, and it did — FY2026 billings reached $3.4B, up 10%, with Q4 billings exceeding $1B for the first time in company history and the year producing over $1B of free cash flow for the first time [9]. The company beat the guidance it set before the crisis. Underlying demand is improving, not eroding: Q1 FY2027 dollar net retention was over 102% and rising, consumption reached multi-year highs across most segments, and customers spending over $300,000 grew 12% — the first double-digit reading in that metric in three years [10]. The AI threat is being monetized as distribution, not lost as share: IAM reached 12.6% of ARR, up from 10.8% the prior quarter, across 40,000 companies, guided toward ~18% by year-end, and DocuSign now connects inside the feared platforms via MCP to Claude, Gemini, and ChatGPT [11]. Meanwhile the balance sheet converts the discount now: 32% operating margin, 35% FCF margin, and a $318M buyback — the largest in company history — cut diluted shares 8% year-over-year to 196.5M [12] [13].

The case for permanent

The impairment is a lasting loss of expansion velocity, not a one-quarter miss. Revenue growth reset from 45% in FY2022 to roughly 8% by FY2026, and consensus embeds no snapback — FY2028 revenue growth of only about 7.5%. The land-and-expand engine broke and has not recovered: dollar net retention fell from 119–123% in FY2021–22 to about 102% in FY2025–26, a 17–21 point structural impairment, and management itself said customers were "expanding at a slower rate relative to their peak levels" even before COVID cohorts rolled off [14]. Remaining performance obligations stalled at $2.4B across FY2025 and FY2026. IAM is an attempted replacement engine, still only 12.6% of ARR and explicitly early — management called the enterprise renewal-cohort sample "pretty small" — and it bundles the new signature offer, so its ARR is not purely incremental. The latest filings say eSignature remains the "substantial majority" of revenue [15], and that AI is making core agreement capabilities "cheaper and easier to replicate" by generic LLMs and general-purpose agents [16]. Free cash flow and buybacks lift per-share value, but they do not prove pricing power or expansion power has returned.

The judges ruled the impairment temporary with probability 0.63 — but the ruling is contested: the three seats landed at 0.68, 0.63, and 0.42, a 0.26 spread that straddles the coin toss.

Source: adversarial-trial ruling (ruchir/trial/tally.json): p_temporary 0.63, three-judge mean 0.577, spread 0.26, contested. Reading order matters — the two temporary-first-mean vs permanent-first-mean readings differ by 0.155, so the diagnosis is order-sensitive as well as split.

The reading is genuinely two-sided and should be carried as such. The one judge who read the permanent brief first and still leaned temporary (0.63) and the one who flipped to permanent (0.42) disagree on the same evidence; the case does not resolve to a clean answer, and rounding it up to one would misrepresent the record. What decides it going forward is observable: dollar net retention breaking below 100% or consumption turning negative year-over-year would signal genuine AI substitution (permanent); total ARR re-accelerating above 11% organically with IAM crossing 25% of ARR would confirm incremental demand (temporary).

Which line broke, and whether it self-corrects

Three separate lines are tangled in the drawdown, and they self-correct on very different clocks.

Sources: billings reversal [17]; DNR history [18] and Q1 FY2027 [19]; AI risk factors [20] [21]; AI de-rating dates and price moves confirmed via web research, no corpus filing.

The line that actually broke in the numerator — billings — has already self-corrected: it is a working-capital timing swing that reversed within the fiscal year, and management flagged the go-to-market change one quarter before the strongest billings print in company history. The line the market is pricing — terminal earning power — did not appear in any reported number; it entered on two headline AI events (OpenAI's agreement agent on 30 September 2025, −12.2%, and Anthropic's legal tools on 3 February 2026, −11.4%). The repricing mechanism, if temporary, is mechanical: at roughly 8× free cash flow, retiring 8–13% of the share count a year is itself most of the return, and it works while the market waits to see whether DNR and consumption confirm the base is intact. The company concedes the structural risk in its own words — competitors "may incorporate AI into their products more quickly or more successfully than us" [22]. The Durability tab weighs that moat question on the longer clock; here the arithmetic is only this: the price destroyed value the near-term numbers did not, and the gap is real only under a reading of the future the trial rates a contested 0.63.