Actionable insights — Software Strikes Back (CRWD · CRM)

1. Rank recurring-revenue metrics by how early they turn — net new ARR and cRPO lead, reported revenue lags

1. Build the hierarchy before reading the print. Net new ARR (recurring revenue added this quarter, net of churn) turns first. cRPO (contracted revenue to be recognized within twelve months) turns second. Ending ARR is a smoothed sum of past quarters. Reported revenue is the slowest of all — it is largely last year's bookings arriving.
2. Judge an inflection on the leading metrics and use revenue only to confirm it two to four quarters later. A company whose revenue decelerates while net new ARR accelerates is inflecting up, not down.
3. Read the second derivative: net new ARR growth accelerating (+32% → +51%) is a different event from net new ARR merely being positive.
4. Interrogate the composition — new-logo net new ARR versus expansion, and the direction of gross and net retention. Acceleration carried entirely by upsell into existing accounts is a narrower signal than acceleration accompanied by record new logos and improving retention.
5. Cross-check management's own guidance on the leading metric, not on revenue. A guidance revision to net new ARR is a far bigger disclosure than an equivalent revenue raise.

• Net new ARR growth decelerating while ending ARR still looks fine (the smoothing hides it for two or three quarters); retention rolling over; and the first quarter a company stops disclosing net new ARR or cRPO separately.

2. Run the redefinition audit — a metric that improves in the same quarter its definition widens has not improved

1. For every headline KPI, check the footnote for a change in scope, definition, or accounting treatment during the period.
2. If the definition widened, the sequential and year-over-year comparisons are broken. Ask explicitly for the like-for-like figure; if it isn't given, treat the metric as unproven rather than as evidence.
3. Estimate the direction and rough size of the distortion — what was added, and how big were those products standing alone before the merge.
4. Note the timing. A widening that lands exactly when the market most wants to see the metric accelerate is a framing choice, not a coincidence.
5. Downgrade the metric permanently in your model: once redefined, the series is discontinuous, so build the thesis on an unedited neighbouring metric (bookings, cRPO, segment revenue).

• Whether a restated prior-quarter figure ever appears; further scope creep next quarter; and any metric that only exists on the "adjusted" or company-defined basis with no auditable anchor.

3. Strip the acquisition out of both the print and the guidance raise before calling a reacceleration

1. For any acquisitive company, find the disclosed dollar contribution of the acquired business and subtract it from revenue to get an organic growth rate. That number, not the headline, is the demand signal.
2. Do the same for forward-looking metrics. If management discloses the acquisition's revenue contribution but not its contribution to backlog or cRPO, say so explicitly — the leading indicator is then only partially readable.
3. Decompose every guidance raise into its parts: acquired revenue, FX, and genuine organic improvement. Report the organic residual, which is usually a fraction of the headline number.
4. Check whether the acquired business also sits inside the fastest-growing reporting bucket — if so, the segment growth rate is inflated on the same basis as the total.
5. Watch for the "hard to read cleanly" problem compounding: each new deal adds another quarter before organic growth is legible again, which is itself a reason to demand more disclosure.

• The quarter Informatica annualizes into the base (organic and reported converge, and the comparison finally becomes honest); cRPO holding ~14% cc once it does — App Economy names that as "the main KPI to watch"; and further deals arriving just as legibility returns.

4. Prize the pricing model that captures the budget before the product choice is made

1. Identify how the vendor takes money: per seat, per module, per unit of consumption, or as an upfront pooled commitment the customer draws down later.
2. Rate the pooled-commitment model highest in a consolidating category. Once the budget is committed, each additional module competes at zero marginal budget against rivals who must find new money — the vendor wins the default.
3. Measure adoption three ways: the committed cohort's ARR growth, its share of total ARR, and the average uplift for customers who convert. All three must move together for the flywheel to be real.
4. Confirm the mechanism produces platform consolidation rather than just a discount — module count per customer, cross-sell attach rates, displacement of named competitors.
5. Then stress it: what does the model cost in a downturn, when a committed pool is drawn down and not renewed at the same level? The upfront commitment that accelerates growth also front-loads the risk.

• The renewal cohort — the first large Flex pools coming up for re-commitment at a lower level; Flex ARR growth slowing as the easy conversions run out; and whether the 40% average uplift compresses as later, more price-sensitive customers convert.

5. Underwrite the AI beneficiary that needs no AI revenue — count the surface area it creates for someone else

1. For any AI-adjacent business, ask whether the thesis requires the company to sell AI, or merely requires other people to deploy it. The second is a much lower bar and a much more durable exposure.
2. Name the unit of expansion precisely — in security, an autonomous agent that "can access files, call APIs, and interact with critical systems" is functionally a new identity, and identities are what the incumbent already charges for.
3. Find the specific product line that monetizes the new unit and demand a disclosed number for it, however small; a structural argument without a line item is a narrative.
4. Track that line's sequential growth rather than its absolute size early on — tripling from a small base is the evidence; the size comes later.
5. Ask what the incumbent's existing distribution is worth here: whether the new product is sold into an installed base already under contract (fast) or requires new buyers (slow).

• AIDR's sequential growth rate as the base gets larger; whether CrowdStrike starts disclosing AIDR ARR in dollars rather than multiples; and the competing claim from identity vendors (Okta, the CyberArk/"Idira" stack inside Palo Alto) that agent identity belongs to them, not to the endpoint vendor.

6. When agents threaten to take the interface, value the meter instead — and check the vendor is wiring itself into rivals' front ends

1. Separate the vendor's two assets: the interface people click through, and the substrate underneath — data, permissions, workflow logic, system of record.
2. Assume the interface can be disintermediated by an agent. Ask what remains chargeable if it is, and whether the pricing model can follow the value there (consumption or platform tiers rather than per-seat).
3. Treat a partnership that lets a rival's assistant reach the vendor's data as evidence for the substrate thesis, not against it — provided the vendor still meters the interaction. Confirm the meter, or the partnership is just a giveaway.
4. Watch the seat-based revenue line separately from the platform/data line; the thesis requires the second to grow faster than the first shrinks.
5. Re-underwrite the multiple on the substrate business, which is stickier but usually slower-growing and lower-priced than the application business it replaces.

• Disclosure of what a Claudeforce-style interaction actually bills; the seat-based application lines decelerating faster than the platform lines grow; and whether the agent vendors start caching or replicating the data rather than calling it, which would remove the meter entirely.

7. Strip strategic-investment gains out of EPS — and hold the same standard on the way up as on the way down

1. When adjusted EPS beats by an unusual margin, check whether gains on strategic or equity investments are included in the "adjusted" figure. Many companies exclude some marks and keep others.
2. Subtract them and restate. Investment marks are portfolio outcomes, not operating results, and they reverse.
3. Apply the identical treatment to investment losses — a miss caused by marks is not an operating miss, and the market frequently punishes it as one.
4. Anchor the valuation on operating income and free cash flow instead, which the marks don't touch.
5. If marks recur every quarter, treat headline EPS as structurally unreliable for that company and build the model on operating lines permanently.

• Marks recurring in size each quarter; the company adjusting out losses while keeping gains; and free cash flow diverging from adjusted net income, which is where the distortion shows up first.

8. Price the gap between non-GAAP and GAAP as dilution, not as an accounting artifact

1. Put the non-GAAP operating margin beside the GAAP one and identify the bridge. In software the bridge is usually stock-based compensation.
2. Express SBC as a percentage of revenue and treat it as a real cost — it is compensation the company chose to pay in equity, and it dilutes existing owners whether or not it is "adjusted."
3. Check whether buybacks are offsetting the dilution or merely mopping it up, and track share count, not just per-share metrics.
4. Compare the SBC ratio to the peer group and to the company's own history; a rising ratio during an acceleration means the growth is being partly bought with equity.
5. Fold it into the valuation explicitly — a triple-digit multiple on an adjusted number that excludes a quarter of revenue in stock is a much bigger number than it looks.

• SBC as a share of revenue trending down as the company scales (it should); diluted share count growth; and any quarter where GAAP operating income turns positive, which is the real profitability milestone.

9. Don't price a sector as one basket — locate the growth before accepting the narrative

1. When a sector-wide thesis is in force ("AI will eat SaaS"), resist applying it uniformly. Ask which specific businesses the mechanism actually harms and which it feeds.
2. Sort the sector by exposure: seat-based application vendors are most exposed to agent substitution; infrastructure, security and data-substrate vendors are fed by the same trend.
3. Test each name against the mechanism rather than the label — the question is whether AI reduces the number of things the vendor charges for, or increases it.
4. Look for the de-rating that applied the narrative indiscriminately; that is where the mispricing sits if a print contradicts it.
5. Keep the narrative alive as a hypothesis rather than settling it on one quarter — two good prints do not resolve a structural argument.

• Seat-count disclosures across application vendors; whether any of them reports declining seats with growing revenue (the substitution thesis confirmed and survived); and the next quarter's prints from the same cohort, which are what actually settle it.