Recurring customers can finance better owner conversion—but not every buyback

The answer: a supported mechanism, not a completed transition

A recurring customer franchise can generate more cash while continuing to pay for selling, product development and talent. Freshworks supplies evidence of that mechanism over 2023–2025. It does not yet establish that the latest compensation and acquisition pattern can be financed indefinitely while continuing shareholders also receive more distributable cash. The distinction is measurable, rather than an objection to equity compensation in principle.

The annual operating improvement is not just a disappearing accounting charge. Between 2023 and 2025, Freshworks revenue rose by $242.377 million and operating income improved by $183.377 million. Recognized stock-based compensation (SBC) expense fell by $63.888 million. Removing that recognition change from the comparison still leaves $119.489 million of operating improvement before recognized stock compensation. A selling-resource proxy that replaces commission amortization with current capitalized commissions fell from 49.30% to 42.29% of revenue, while its dollar amount increased. That is evidence of scale permitting a lower cost share, not evidence that continuing distribution costs vanished. [3]; calculations

The latest matched half-year is less reassuring about a smooth continuation. Operating income improved by $17.072 million, but recognized stock compensation fell by $19.175 million: operating income before that recognized expense deteriorated by $2.103 million. Selling resources grew faster than revenue. Meanwhile, unvested RSU/PRSU units increased from 21.936 million to 31.084 million. A lower expense and a larger outstanding award inventory can therefore occur together. [4]; calculations

Repurchases nevertheless made a real difference to ownership. Across 2025 and the first half of 2026, basic shares fell by 40.572 million, while the combined inventory of basic shares, unvested RSU/PRSU units and outstanding options fell by 31.526 million. The latter is a gross unit inventory, not a fair-valued fully diluted capitalization. Cash repurchases were $593.717 million, against $327.999 million of operating cash after cash property/equipment and software investment, before employee settlement cash and acquisitions. The matched treasury account shows how accumulated financial resources financed the difference. Neither the cash spent nor the ownership improvement should be erased. [3]; [4]; calculations

A deliberately limited, consistently costed comparison preserves a favorable case. At the reported 2025 average repurchase price, allocating cash to neutralize the year's net addition to that gross inventory would leave $54.309 million after employee cash settlement and the acquisition advance, before repurchase taxes, additional reserves and other distributions. Reserving the entire recognized interest-income amount reduces that sensitivity to $16.128 million. Applying the same convention to the observed first-half 2026 units and historical execution price instead requires $70.570 million more free cash flow, including that half's acquisition payment. These are allocation requirements under disclosed historical quantities, not estimated expected award values, normalized earnings, forecasts, or price targets. calculations; claims / cash requirement

Salesforce supplies a meaningful favorable contrast: operating income improved while stock compensation and development spending increased in absolute dollars. It also supplies a warning against treating all buybacks as internally earned distributions. Its acquisition and financing account remains relevant even after a large recurring business has developed. The transferable mechanism is operating leverage with paid renewal—not Salesforce's margin as a target assigned to Freshworks. [8]; [9]

Whole-question consequence: the evidence strengthens selective, paid-for value creation at the application layer. It weakens both “cash conversion is entirely fictitious” and “a rising adjusted margin proves that owners receive the improvement.” It does not establish Freshworks, Salesforce or the entire AI sector as a bubble, and it does not certify the prices studied in the earlier ownership investigation. The remaining empirical burden is the joint persistence of customer value, complete service costs, new compensation claims and funded investment. [13]; method and interpretation

Scope, periods and what counts as owner cash

The last reported period is not the current bank account

Freshworks' September quarter had ended by the inspection date, but the company's 8 October announcement scheduled its results for 3 November 2026. Its investor-relations overview still identified Q2 as the latest quarterly report. June therefore remains the latest complete published financial period used here. A June award count is not a 17 October capitalization; an announced transaction is not inserted into June cash. The period boundary is intentional, not a claim that nothing happened afterward. [7]

The early history establishes the transition from the IPO-era business. The detailed cost and cash decomposition uses 2023–2025, because one later annual filing provides aligned three-year statements. Separate matched first halves prevent seasonality from being mistaken for a year-on-year trend. The 18-month cash account adds 2025 plus H1 2026 only; it does not add H1 2025 a second time. Salesforce's January fiscal year is explicitly different from Freshworks' December year. [1]; [3]; [4]; [8]

The earlier customer-capture report supplies the customer-versus-product distinction and the signed working-capital reconciliation. The ownership report supplies the opening-award/future-grant convention and its genuinely costed favorable cases. Neither is replaced by another valuation exercise. The buyer and full-service studies constrain claims about avoided work and continuing costs; the substitution study does not provide a measured Freshworks hosting saving. [10]; [11]; [12]; [13]; [14]

Four different measures are needed

Operating income includes the disclosed cost of delivery, product work, distribution, administration, restructuring and recognized compensation. Operating cash flow (CFO) adjusts accounting income for noncash items and payment timing; it is not gross customer collections and includes non-franchise items such as portfolio income and taxes. Unadjusted free cash flow (FCF) here is CFO less cash purchases of property/equipment and capitalized software at Freshworks. Acquisitions and financing cash are then shown separately. The analogous Salesforce measure subtracts its reported capital expenditures, with financing-obligation principal explicitly retained as a further use. [3]; [4]; [8]

Cash potentially available to a continuing ownership claim needs a specified compensation convention and investment boundary. This report first shows actual cash uses and actual gross unit movements. It then tests a convention that maintains the opening gross claim inventory: pay employee withholding, credit actual employee exercise/purchase receipts, and allocate repurchase cash to the net new units. Recognized stock compensation is not deducted again in that convention. Additional owner distributions, taxes on repurchases, acquisitions and required resources still use cash.

A buyback pays a seller. A holder who continues holding may own a larger proportion afterward, but also owns a proportion of a company with less cash. Those are not two independent gains. A computed remaining budget is not an observed dividend, permission to distribute, or a cash payment already received by the continuing holder.

The analysis does not manufacture maintenance capital expenditure or a minimum talent bill. Existing accounts mix maintenance, improvement and growth. Preserving the actual cost base in a historical comparison is different from showing that precisely that base can sustain future revenue. The latter remains a conditional operating proposition.

Customers: recurrence is supported; the organic and AI-only attribution is not complete

A growing franchise, with a less uniform expansion signal

Sources: [1], [2], [3], [5] and [6]. “Not used” is not a zero or a claim that the issuer never reports the quantity. Half-year retention refers to the period-end trailing cohort measure, not retention during just six months.

The larger-customer count rose materially, as did its share of the franchise. That supports demand for substantial recurring contracts. It does not establish that the same individual customers moved upward, that new acquisition spending became more productive, or that the incremental contracts were all organic. A threshold count can change with new customers, expanded spending, exchange rates and acquisition. [2]; [3]

Net dollar retention measures the change in an entering customer's ARR, including expansion and contraction. The issuer's definition also permits expansion associated with acquisitions. It is not the fraction of customers renewing, realized customer profit, or a cash collection rate. Converting both period-end ARR measures at a stated exchange-rate convention removes one influence, not every acquisition or mix effect. [5]; [6]

A new source conflict prevents a falsely favorable annual interpretation. The 2025 annual filing describes constant-currency retention of 104% as an increase from the prior year. The full-year release explicitly pairs 104% with 105% a year earlier, consistent with the 2024 filing's prior observation. The table uses that explicit numeric pair and retains the prose inconsistency. This is not an issuer-confirmed correction. It does not overturn the separately dated June comparison preserved in the earlier customer dollar and renewal investigation: reported retention weakened from 106% to 104%, while the constant-currency pair improved from 104% to 105%. Annual and June windows should not be merged into one direction of travel. [2]; [3]; [5]; [6]; [10]

Revenue growth cannot be reduced to one cohort multiplier

The issuer's rounded 2023 revenue bridge attributes $62.2 million of growth to existing customers and $36.2 million to new customers. The 2025 discussion attributes approximately $80 million to the existing base and $38.4 million to new customers and relevant acquired business. H1 2026 attributes approximately $34.8 million to the existing base and $30.3 million to new customers. The descriptions and acquisition perimeters do not provide a complete longitudinal organic cohort decomposition. They nevertheless show that both expansion within the franchise and replenishment matter. [1]; [3]; [4]

Do not multiply total prior-year revenue by period-end retention to “solve” new-customer revenue. ARR is a point-in-time measure; revenue is recognized throughout the period. Contract starts, acquisitions, foreign exchange, usage and delivery timing differ. Such a calculation would give a precise-looking allocation that the source does not establish.

Device42 entered during 2024; FireHydrant closed in January 2026. Their acquired customer relationships and product capabilities are part of the consolidated franchise, not free organic expansion. The annual filing contains Device42 purchase accounting and supplemental pro forma information, but the pro forma totals are not an observed organic revenue series. FireHydrant replacement employee awards are excluded from purchase consideration; their units are not separately allocated out of the aggregate award table. The purchase and compensation analyses must therefore remain joined but not conflated. [3]; [4]

The source mix also argues against assuming universal cross-sell improvement. Reported multi-product customers moved from 26% of customers in 2023 to 36% in 2024 and 31% in 2025; their ARR shares were 47%, 46% and 45%. The definition includes suites and bundles, so those movements are not a controlled cross-sell experiment or a count of product losses. Monthly subscriptions fell as a share of ARR from 17% to 14% to 13%. Longer contractual persistence may help collections while still leaving the next voluntary buying decision to be earned. [2]; [3]

Freddy AI's reported $25 million ARR and the more-than-$40 million Device42 and enterprise-service-management milestones are useful product-scale observations, not matched product cash margins. No portion of all Freshworks salary, hosting or subscription revenue is relabeled AI-only economics. [5]

Interpretation: revenue recurrence has an empirical basis. More efficient selling is plausible because an installed customer relationship can support expansion and renewal. Its durability is not settled by aggregate retention alone: migration burden, product bundles, discounting, service quality and acquired customers remain important competing explanations. the earlier buyer demand and budget substitution investigation's distinction between a priced purchase, contractual persistence and the next value-based decision remains applicable. [11]; method and interpretation

Complete costs: genuine annual leverage, a mixed latest half

The longer cash and profit trajectory

Sources: [1], [3] and [4]; calculated investment and FCF columns. Acquisitions, employee financing cash and buybacks are not deducted in this table; they appear below. The early IPO-era accounting is not assumed to be a recurring grant policy.

Positive cash is not by itself the entire owner result. Equally, the historical transition from low or negative FCF to a substantial positive amount should not be dismissed merely because it includes compensation addbacks. The task is to determine what else was financed and what claims remain.

Sources: [3] and [4]. The source dash for 2023 restructuring is a reported zero. The final row removes an accounting recognition component for decomposition; it is not a fully paid profit measure and is not used as owner cash.

For 2023–2025, the additional $242.377 million of revenue less $59.000 million of additional complete operating costs produces the $183.377 million operating improvement. Of that improvement, $63.888 million comes arithmetically from lower SBC recognition and $119.489 million remains before that change. Within 2024–2025 alone the corresponding amounts are $151.815 million, $69.887 million and $81.928 million. This establishes that the annual improvement cannot be explained solely by the SBC decline. It does not establish which management action caused the remainder. Nor does it make the positive 2025 GAAP result a recurring profit baseline: removing only the disclosed $38.7 million specific-award reversal would change operating income from $13.205 million to −$25.495 million. That single-credit sensitivity is not a complete normalized compensation policy. [3]; calculations

The latest half-year reverses that inference at the margin. Additional revenue of $65.059 million was accompanied by $47.987 million of additional complete operating costs. The $19.175 million SBC reduction is larger than the resulting $17.072 million improvement. Without that recognition change, the remaining operating account deteriorates by $2.103 million. This is a seasonally matched comparison, not an annualization of June. calculations

Current selling investment is not its amortization

Freshworks capitalizes qualifying commissions, generally amortizing initial and expansion contract costs over three years. Current capitalized commissions rose from $26.962 million in 2023 to $34.524 million in 2024 and $40.071 million in 2025; amortization was $23.965 million, $28.556 million and $31.702 million. The resulting asset rose from $39.675 million at the start of 2023 to $57.009 million at the end of 2025. H1 2026 added $21.290 million and amortized $17.533 million, reaching $60.766 million. No expenditure becomes free because recognition is deferred. [3]; [4]

Define the commission-timing-adjusted selling proxy:

S = GAAP sales and marketing − recognized S&M SBC − commission amortization + current capitalized commissions.

This proxy isolates a useful direction without calling itself cash customer-acquisition cost. Other accruals, acquired-intangible amortization, shared marketing, reseller payments and employee compensation remain incompletely allocated. There is no measured cohort CAC or payback period. The substitution of commission investment for amortization occurs once; FCF already includes the cash-flow effect of that investment, so the proxy is not deducted again from FCF.

The favorable mechanism is concrete: selling resources increased by $60.715 million over two years, yet the ratio fell about seven percentage points because revenue grew faster. The latest half adds a constraint: the proxy grew $35.956 million and its ratio rose about 2.06 percentage points. The filing attributes H1 S&M growth to salary/personnel, sponsorship and advertising, tools, professional services and reseller commissions. Cutting that burden without sacrificing new customers, expansion or service remains an operating hypothesis—not a ratio that can simply be reset to 2025. [4]; calculations

Product renewal and delivery still consume resources

A complementary partial resource proxy, R&D minus its recognized SBC plus cash software capitalization, rises from $106.503 million in 2023 to $128.563 million in 2024 and $144.135 million in 2025. It is not pure cash R&D: other accruals remain, and software amortization elsewhere is not reclassified away. Its purpose is to show that a favorable scale story need not assume no continuing product work. These amounts are not an extra deduction after the cash statement. [3]; calculations

H1 2026 cost of revenue rose $9.764 million; the issuer attributes approximately $6 million to cloud/hosting and $1.9 million to software amortization. R&D and S&M increased even while G&A fell. The G&A decline is materially associated with the compensation change, including the earlier executive-award recognition schedule. No evidence here identifies AI automation as the causal source of the aggregate savings. [4]

The November 2024 restructuring and the May 2026 restructuring also have different observation windows. The latter produced $7.032 million of H1 charges, with approximately $5.6 million paid and $1.4 million still accrued. Excluding a paid restructuring charge may help compare management's chosen adjusted measure, but cannot make that cash available to owners in the same period. Reduced future staffing may lower costs; whether retained capability supports the customer franchise is not independently measured. [3]; [4]

From operating performance to cash: timing, taxes and acquisitions

The cash improvement has a source, not just a margin label

For 2025, the signed cash-statement reconciliation is:

183.723 net income + 57.362 noncash adjustments + 1.285 working-capital movement = 242.370 CFO.

The noncash adjustments include the removal of the $151.738 million deferred-tax valuation-allowance release. That release helps reported earnings, but is not cash collected from customers. Relative to 2024, CFO improved by $81.724 million even though the net working-capital contribution deteriorated by $8.447 million. The annual increase therefore is not explained by a more favorable aggregate working-capital inflow. It still contains stock-compensation addbacks, taxes and portfolio returns that must be handled in their own accounts. [3]; calculations

The H1 comparison is different:

−1.571 net income + 126.542 noncash adjustments − 4.059 working capital = 120.912 CFO.

The corresponding H1 2025 working-capital movement was −$10.349 million. Its improvement of $6.290 million exceeds the $4.348 million increase in CFO. Cash investment rose $4.668 million, leaving FCF $0.320 million lower. The earlier customer dollar and renewal investigation already established the current cash total and the conflict between the filing's favorable working-capital prose and its signed outflow components. The new contribution here is the longer history and its connection to current investment and claims, not rediscovery of that sign issue. [4]; [10]; calculations

Cash-flow stock compensation is $81.770 million in the H1 2026 cash statement, while the function-level expense note totals $81.769 million. The cash reconciliation uses the former and the operating decomposition the latter. The $0.001 million difference is not silently made into another saving. Capitalized stock compensation is also kept distinct from the cash software line; it is not a second cash outlay.

Treasury income is consequential. Recognized interest income fell from $51.696 million in 2024 to $38.181 million in 2025 and from $21.449 million in H1 2025 to $13.255 million in H1 2026. Management links lower portfolio balances to repurchases. Thus distributing accumulated assets can reduce the future income that helped the reported cash measure. The recognized-interest line is not a separately verified interest-cash receipt: securities discount accretion, accrual timing and tax effects prevent simply calling FCF minus that line pure franchise cash. Section 7 treats that subtraction as a transparent sensitivity, not an accounting normalization. [3]; [4]

A new acquisition reconciliation closes an earlier timing gap

The prior ownership investigation could not fully join FireHydrant's consideration, acquired cash, holdback and H1 acquisition payment. The 2025 annual statement now supplies an $18.432 million advance for a business combination, and its subsequent-event note identifies the advance as related to FireHydrant. Combined with the June purchase-accounting record:

18.432 + 56.913 + 4.349 + 9.043 = 88.737.

This resolves the specific $18.432 million timing join, not every acquisition economics question. The 2025 restricted cash with the payment administrator is a balance, not another purchase expense to add. New employee replacement awards are excluded from consideration and remain within the compensation question. FireHydrant's standalone contribution, return on integration costs and the allocation of replacement award units remain unmeasured. [3]; [4]; [13]

Device42's $213.905 million net acquisition cash in 2024 is also outside the displayed FCF. It is included in the matched annual treasury account. A growing product franchise can rationally buy complementary capability; it cannot claim the acquired revenue while indefinitely presenting the acquisition resource as irrelevant to owner conversion. Conversely, subtracting the entire acquisition price again every later year would charge the same historical purchase repeatedly. Future acquisition needs require an explicit policy, not either extreme. [3]; method and interpretation

Compensation: recognition fell, but claims and cash must be followed separately

Four measures answer different questions

Grant-date accounting value prices an award for recognition. Recognized expense allocates that value over service and performance conditions, with modifications or reversals when applicable. Employee realization depends on vesting, exercise, market prices and tax. Employer cash includes withholding remittances, exercise and purchase proceeds, and any repurchases. None is a reliable substitute for all the others.

The older IPO-related recognition and the chair's modified performance awards make a naive expense trend particularly unreliable. The 2024 activity table includes approximately 0.9 million modified chair awards among grants and cancellation of 6 million old awards among forfeitures. The 2025 note reports a $38.7 million reversal on a specific forfeited award, while the release describes an approximately $41 million broader reversal. Those are distinct disclosed scopes, not interchangeable amounts to use in a supposed exact normalization. Future ordinary forfeitures cannot simply be assumed to repeat those unusual cancellations. [1]; [2]; [3]; [5]

Follow the actual units before assigning a future policy

Sources: [1], [2], [3] and [4]. Units follow the source tables; they are not current market values or the maximum payout on every performance award.

H1 2026 granted more units than full-year 2025, but multiplying those first-half grants by two would be unjustified. Annual refresh timing, hiring, retention, acquisitions and performance grants all affect the pattern. FireHydrant's replacement awards are not separately allocated in the aggregate table. The records establish a larger inventory, not the expected annual economic cost of an unchanged workforce. [4]

Performance terms matter. The 2026 executive PRSUs use revenue and non-GAAP operating-income measures, with a range of outcomes and continued service conditions. A target unit is not a certainty of vesting, nor necessarily the maximum right. The increased recognition backlog—$298.923 million at December to $352.621 million at June—is a schedule attached to existing rights, not another liability to add to those rights. [3]; [4]

The exact realized value of all employee awards is not measured here. Option exercise, grant-date fair values and the issuer's vesting disclosures have different valuation conventions. The report does not multiply every grant by the current stock price and call that cash paid to employees. Cash remittances and share movements provide a usable alternative convention without inventing option fair values, expected forfeitures or a required talent-cost floor.

The opening claim enters once; vesting does not create it again

Let B be issued common shares, A unvested RSU/PRSU units, and O outstanding option units. Define gross inventory I = B + A + O. This deliberately omits option valuation, contingent excess performance units and unexercised ESPP rights not represented by issued shares. It is not a fully diluted equity valuation or an upper bound on every possible right.

Sources: [3] and [4]; same rounded table precision throughout. Exact program-share notes and the equity table have small rounding/scope differences; they are not mixed to manufacture an exact-share claim.

The common-share identities are:

2025: 303.382 + 0.227 option exercises + 6.148 net RSU issuance + 0.566 net ESPP issuance − 26.896 repurchased = 283.427.

H1 2026: 283.427 + 0.004 + 3.020 + 0.403 − 24.044 = 262.810.

Vesting moves an old right into common shares. Net settlement removes some shares for taxes; exercises also move a pre-existing option claim into issued shares. Neither should be charged as a newly granted claim again. The option inventory falls through exercise and cancellation; cash exercise receipts are separately credited. Actual withholding cash need not equal contemporaneous withheld units multiplied by a rounded program repurchase price.

The 18-month decline in B is 13.37%, versus 9.62% in I. A statement that compensation caused net unit dilution over this interval would be wrong under the displayed convention: repurchases more than offset the remaining new inventory. A statement that continuing owners received the full basic-share reduction without using resources would also be wrong. The gap between B and I matters particularly in the latest half, when unvested units grew by 9.148 million. calculations

Join the ownership change to the treasury account

Sources: [3] and [4]; arithmetic reproduced from the individual statement rows. Repurchase cash is the cash-flow measure, not trade-date purchase consideration. Securities turnover is a net cash source, not an estimate of investment profit or proof that each purchase was financed by a specific maturity.

The identities reconcile, with no new debt or equity-financing plug. The issue is the use of accumulated resources, not a failure to account for cash. The closing amount contains restrictions and is not certified distributable liquidity. A repeat of the same cash uses would need more ongoing resources or changed uses; it cannot repeatedly withdraw the same opening portfolio.

This account is favorable in one important respect: a liquid balance sheet made an actual ownership contraction possible while the business continued investing. It is limiting in another: cash returned to selling holders exceeded contemporaneous FCF, and withholding and acquisition payments made the resource claim larger. The cash and the shares must stay in the same explanation.

A constrained future-conversion requirement, not another valuation grid

The convention and its limits

A continuing shareholder can ask a narrower question than “what is the stock worth?”: could this operating cash finance the recorded new gross claims without consuming the opening financial assets, while leaving room for another distribution? Answering that question needs no discount rate or stock-price target.

Under the displayed inventory convention, the net addition before buybacks is:

q = grants − award forfeitures − option cancellations − (gross RSU vesting − net RSU shares issued) + net ESPP shares issued.

For these two periods the option tables contain no new option grant line requiring an additional term. More generally new option grants, acquisition shares and other newly created rights must be added; their absence must not be assumed in a future forecast.

The result is q = 6.843 million units in 2025 and 12.571 million in H1 2026. Existing awards enter I at the opening and flow through vesting or cancellation, so their future expense recognition is not separately charged. Actual employee purchase/exercise cash and withholding are handled once. Maintaining I then requires purchases of max(q, 0) common shares at the stipulated price, within this gross-unit convention.

Write E for employee cash receipts, W for withholding, A_cash for acquisitions, p for the execution price, D_owner for additional owner distribution and R for other required reserves or cash costs. The requirement is:

FCF ≥ W − E + A_cash + p × max(q, 0) + D_owner + R.

This is a resource requirement, not a claim that every repurchase was made to neutralize compensation. It is not equivalent to expected fair-value compensation cost. Contingent performance outcomes, unallocated acquisition replacement awards and future ESPP options are unresolved. Forfeiture releases of old awards can help one period without supporting a permanent policy. The requirement is useful precisely because those future choices have to be stated rather than hidden in a declining accounting percentage.

What the observed quantities allow

For the following comparison, use the issuer's historical program average prices, $14.29 and $8.50—not a current quote. Set additional distributions and reserves to zero only to measure the remaining capacity for them. The historical prices exclude or do not fully resolve excise tax and payment-timing effects; therefore the results are before those additional costs.

Sources: [3] and [4]; new calculations. These are not observations of owner distributions or full economic profits. The interest reserve is an accrual-based sensitivity, not measured cash-interest normalization. H1 is not annualized.

The positive 2025 result matters. It funds the actual period's selling, delivery, cash product investment, employee settlement and acquisition advance under the selected unit convention. It does not require paying employees nothing, holding every cost fixed or assigning Salesforce's margin. Even the explicit interest-reserve sensitivity leaves positive room before further costs. The much larger actual repurchase program chose to use more resources than this neutralization allocation; that is a capital-allocation choice, not proof that the smaller funded path was impossible.

The latest half imposes a different constraint. Under its observed quantities, maintaining the gross inventory would require another $13.657 million of FCF even before its acquisition cash, or $70.570 million including it, before additional costs. The $9.043 million unpaid acquisition holdback is not deducted from these historical cash payments; when payable, funding it would add that amount to the requirement in its own payment window. It is neither assumed paid at June nor forgotten as a remaining use. The interest-reserve sensitivity raises the displayed H1 requirement to $83.825 million. A sustainable plan could alter cash generation, grants, cancellations, acquisition cadence or distributions; the calculation does not declare that any one lever is operationally costless.

What the favorable continuation must fund: enough selling and customer support to replenish and expand the base; continued product and security work; hosting at the service level customers buy; compensation sufficient to retain the needed workforce; taxes and commission investment; and explicitly chosen acquisitions or other expansion. Revenue growing faster than this paid resource base can increase conversion. Repeated executive-award reversals, portfolio drawdowns and a permanent continuation of one favorable working-capital movement cannot substitute for that operating proposition.

The evidence does not identify a steady-state floor for new grants. Nor does it show that reducing H1 unit grants to the 2025 rate would preserve retention, hiring and acquired capability. Thus the comparison supplies a constrained requirement and an observed favorable reference, not an estimated probability of sustainable conversion. This is the empirical extension of the earlier ownership convention, not a rerun of its growth or compensation roots. [13]

Salesforce: a mature mechanism contrast, not a borrowed margin

Better conversion need not require declining compensation

Source: [8]; calculated rows use the same expense/commission convention as Freshworks. Dollars in millions. These are consolidated measures, not Agentforce product economics. Cash capital expenditure is the issuer's disclosed aggregate; the analysis does not assert an identical software-capitalization perimeter to Freshworks.

The mechanism is materially different from Freshworks' latest compensation-recognition decline. From fiscal 2024 to 2026, Salesforce operating income improved $3.320 billion despite a $0.722 billion increase in recognized SBC. Its operating income before that recognition expense improved $4.042 billion. R&D increased $1.087 billion. This is direct evidence that a mature recurring business can fund more product work and compensation while improving its operating result. It is not proof that every new dollar of development earned its cost. calculations

The selling proxy also prevents an overly smooth account of maturity. Its ratio improved between fiscal 2024 and 2025 but rose again in fiscal 2026. Capitalized contract acquisition costs increased to $2.811 billion, compared with $2.197 billion of amortization. Current plus noncurrent deferred contract costs rose from $4.446 billion to $5.060 billion. Even an established distribution franchise pays for new and renewed business. [8]; calculations

There is a real comparability limit: Salesforce amortizes qualifying new contract costs over four years and renewal costs over two years, versus Freshworks' three-year initial/expansion convention. Replacing amortization with current capitalized costs removes one recognition difference; it does not standardize sales channels, product breadth, customer sizes, payroll benefits, license commissions or the economic lives of acquired customer relationships. The percentage comparison is a mechanism check, not a league table or a measured target for Freshworks.

Customer persistence is useful evidence at its actual boundary

Salesforce reports approximate 8% attrition, with exclusions for Slack self-service and current-year acquisitions. That supports persistence within the stated population; it is not Freshworks' net dollar retention and should not be converted into a comparable “92% renewal rate.” Annual advance billing and the timing of renewals also affect collections. A high first-half cash share is not automatically a new permanent cash margin. [8]

Its business mix differs too. Subscription and support is about 95% of revenue, but that category includes term licenses, approximately 6% of subscription/support revenue in fiscal 2026, whose recognition can be upfront rather than ratable. Informatica contributes $388 million of fiscal-2026 subscription/support revenue after its November acquisition. Product labels containing “Agentforce” do not turn the whole category into newly incremental AI revenue. These differences are why the contrast tests funded scale and retention rather than assigning a mature margin to Freshworks. [8]

The compensation account also warrants restraint. The rounded restricted-stock activity table moves from 26 million units through 14 million time-based grants, 2 million performance grants, 3 million cancellations and 13 million vesting to 26 million. Options fall from 8 million to 6 million through exercise. The common-share account moves from 962 million through 17 million issued and 50 million repurchased to 929 million. A separate reservation table lists 25 million restricted award units rather than 26 million. The representation/rounding difference is left unreconciled; no exact fully diluted Salesforce claim is inferred from it. [8]

This does not invalidate the observed operating contrast. It limits the additional conclusion that its cash and new compensation obligations have been fully priced on the same basis as Freshworks' more detailed gross-unit account.

A mature franchise still faces an allocation constraint

For fiscal 2026, Salesforce's $14.402 billion cash residual after capital expenditure falls to $13.818 billion after financing-obligation principal. Adding $1.039 billion of employee-plan proceeds and deducting $0.351 billion of withholding leaves $14.506 billion. Cash repurchases of $12.596 billion and dividends/equivalents of $1.587 billion leave $0.323 billion before acquisitions and strategic investment. The same year includes $9.268 billion of net acquisition cash and $1.774 billion of net strategic-investment purchases. Debt proceeds, marketable securities and cash therefore remain part of the account. [8]; calculations

The $0.323 billion is not economic profit after expected new grants; it is an actual cash-allocation remainder at an explicit stopping point. Actual repurchases and employee settlement have already been charged, so deducting the recognized SBC expense again would not produce a coherent cash-plus-claim measure.

A later observation reinforces the distinction. The indexed original July 2026 quarterly filing identifies $25 billion of March note principal, approximately $24.8 billion of net proceeds and a $25 billion accelerated-share-repurchase prepayment. This later financed transaction is not inside the January annual totals, and its indexed account's expected final settlement is not promoted into a verified October settlement. The full interim body could not be opened in this inspection; only the specific original indexed financing passage supports this observation. [9]

What transfers: recurring customer relationships can support increasing full operating profit and large cash generation while continued development and talent are funded. What does not transfer: the same margin, acquisition requirement, grant policy, renewal rate, collection seasonality or ability to repeat any selected buyback amount. Mature distribution is a credible favorable mechanism—not a blanket exemption from capital allocation.

Sustainable conversion: what is supported, what would change the answer

Favorable and adverse explanations can be distinguished

The favorable explanation has several observed foundations. Freshworks has a larger recurring franchise, stronger annual operating performance even before the SBC recognition change, and a lower multi-year selling-resource ratio while dollar spending increased. The actual buybacks reduced both basic shares and the displayed gross inventory. Salesforce demonstrates a mature operating improvement that did not require a declining recognized compensation bill. These observations make a paid-for scale transition more credible than a model that simply assumes away customer acquisition and talent costs.

The contrary evidence is specific rather than categorical. Freshworks' latest half has a higher selling-resource ratio, more unvested units, weaker operating income before recognized SBC, and essentially flat unadjusted FCF. The longer treasury account shows substantial use of accumulated resources. Some apparent earnings improvement is noncash tax recognition or unusual compensation reversal. The annual constant-currency retention pair does not support its accompanying favorable prose. None of those facts alone demonstrates loss of customer usefulness, but together they limit a claim that the transition has already become self-sustaining.

A durable favorable path would look different from a temporary accounting improvement: existing customers would keep adequate paid service; new and expansion sales would grow faster than the fully paid selling burden; required development and hosting would remain funded; the net creation of compensation rights would fit within cash available at an explicit ownership convention; and acquisitions or distributions would not repeatedly draw the same opening financial assets. A lower cost percentage could coexist with a rising absolute cost. It need not imply poor service or underpaid staff.

The adverse path is not “equity is fictitious.” It is that a business spends enough to sustain service and customer growth, yet continuing owners do not receive the claimed improvement because new rights, acquisitions or the required resource base absorb it. Reducing those uses can improve near-term cash while undermining future demand. The available aggregate data cannot causally allocate that trade-off to a specific restructuring or AI feature.

The highest-value observations are now concrete

These are discriminating observations, not additional completion gates or demands for inaccessible private customer data. Public future reports can materially improve the account even without exact product-level profitability. The current study does not wait for that evidence to explain the constraints already measured.

Consequence for the accumulated bubble assessment

The earlier ownership study treated a useful Freshworks franchise as capable of supporting a stronger, fully costed outcome and did not establish the company as a bubble. That favorable path survives. This investigation adds an empirical basis for annual scale economies, while showing that the latest expense trend is not the same as a lower recurring claim burden. It narrows which improvements can reasonably be treated as already observed. [13]

The buyer studies explain why customers can rationally purchase without producing equivalent new supplier revenue; the full-service study explains why a cheap component must still finance the rest of the service. The present account follows those costs into the actual application business and then into employee and shareholder claims. It does not assign the substitution study's cloud savings to an undocumented Freshworks workload or turn application cash into the upstream model developer's return. [10]; [11]; [12]; [14]

The supported whole answer is selective value creation with a still-conditional distribution to continuing claims. Useful AI-enabled products, profitable providers, genuinely received seller cash and adequate later ownership returns are connected but different outcomes. Freshworks' history supplies more evidence for the first parts of that chain and a measurable constraint on the last. Salesforce shows that the favorable operating transition is possible in a mature business, not that any application company's current price or future grant policy is thereby validated.

No new valuation grid, stock recommendation, sector-wide loss estimate or crash date follows. Earlier financing, customer, continuation and ownership findings retain their own dates and perimeters; this report changes the recurring-business conversion assessment, not the meaning of those earlier measurements.

Methods, source status and reproducibility

What is measured and what is inferred

The source-linked ledger contains 551 numeric observations. Each record has a company, metric, unit, observation window, source ID, original URL, locator, inspection date and qualification. Financial values from Freshworks' thousand-dollar tables are converted to millions; Salesforce's million-dollar tables retain their unit. Award activity uses millions of units. A source dash may be a reported zero; unavailable values are not turned into numeric records. Threshold ARR observations are identified as thresholds in their notes.

Numbers can be repeated in different measurement roles without creating independent evidence. The annual comparative statements and interim comparisons are not separate samples of customers. The data ledger retains both the H1 2026 cash-flow SBC line and its slightly different expense-note total because they support different reconciliations. Annual and half-year records overlap for comparison but are only summed in the explicitly non-overlapping 18-month account.

The calculation companion reproduces the operating statements, signed CFO components, commission stocks, financing and investing cash, award and common-share rolls, acquisition join, cash/claim account and conditional neutralization comparison. It uses decimal arithmetic, explicit units, source validation and rejection of blank/non-finite values. The source-linked inputs are observations; ratios, decompositions and resource requirements are this report's calculations. Successful arithmetic does not verify a bank payment, prove causality or price a contingent award.

Three conventions deserve emphasis. First, OI plus recognized SBC is only a decomposition of accounting change. It is not cash profit. Second, the selling proxy replaces commission amortization with current capitalized costs and removes recognized S&M SBC; it is not cohort CAC and is never deducted again from FCF. Third, the gross-inventory cash convention starts with old rights already included, credits actual employee proceeds and charges withholding and stipulated purchases once. It cannot be combined indiscriminately with a second SBC expense deduction or a second dilution charge. [13]; method and interpretation

Cash interest, share-based tax benefits, ordinary tax payments and acquisition taxes are not fully allocated into a normalized franchise cash series. The recognized-interest reserve is deliberately labeled a sensitivity. There is no estimated maintenance/growth split, independent customer-acceptance sample, private payment audit, empirical talent-cost floor or forecast probability.

Source conflicts and access limits that survive the result

The principal annual and interim financial statements and compensation notes were inspected as original SEC HTML. Issuer releases supplied explicit retention comparisons and product observations. No external PDF was used for the numerical analysis. HTML tables, not scanned images or OCR, carry the financial evidence.

The material unresolved source distinctions are: Freshworks' annual constant-currency retention prose versus the explicit numeric pair; the $0.001 million H1 SBC cash-versus-expense difference; the narrower $38.7 million award reversal versus the broader approximately $41 million release observation; small equity-table versus exact program-share rounding differences; and Salesforce's 26 million restricted-unit activity balance versus its 25 million reservation table. None is silently “fixed” by guessing a missing transaction. The first changes the interpretation of the annual retention direction; the others limit precision or normalization without preventing the displayed reconciliations.

The specific FireHydrant advance join is resolved, but replacement compensation and complete acquisition returns are not. Company-level growth, customer thresholds and mixed product measures do not produce an exact organic or AI-only profit account. Complete new-grant expected value cannot be inferred from aggregate target units and grant-date recognition.

The full body of Salesforce’s July 2026 interim filing was not retrieved. Its three financing observations are identified in the ledger as indexed-source evidence and is not used as a full latest-period operating or ownership account. The principal Salesforce contrast instead uses the fully inspected annual history. The Freshworks results-announcement page was readable through indexed original text, while a direct open failed; the source status is retained. No access refusal was treated as permission to seek credentials, bypass a restriction or pay for data.