Actionable insights — This Week in Visuals (#521)

1. Separate AI adoption from AI monetization — count the customers on the meter, not the customers using the feature

1. For every AI claim in a print, sort the disclosed numbers into two buckets: adoption (users, customers touching a feature, MAUs, deals closed, penetration) and monetization (ARR, ACV, revenue, customers on a paid plan or usage meter).
2. If a company discloses only adoption metrics, assume the revenue is immaterial until it says otherwise — a company with meaningful AI revenue reports it, because it is the better number.
3. Where both are given, compute the conversion ratio directly: paying customers ÷ using customers. That single fraction is the entire state of the monetization effort.
4. Read management's stated reason for a low ratio. "Deliberately prioritizing adoption" is a strategy with a timetable — demand the timetable. Note whether the giveaway is reversible: usage customers have to accept a bill later, and habituation to free is a real cost.
5. Then check what the AI is defending against. A per-seat vendor deploying agents that reduce headcount is cannibalizing its own billing unit, so the consumption meter is not upside — it is the replacement for revenue it is about to lose.
6. Finally, prefer the disclosure that is cohort-scoped (penetration within customers above a spending threshold) to a company-wide total, which mixes free trials into the same figure as production deployments.

• Workday's Flex Credits customer count each quarter — that number, not agent adoption, is the thesis; the first quarter a company stops disclosing the using-customer count (usually because the ratio got embarrassing); and any per-seat vendor whose seat count flattens before its consumption revenue starts.

2. Prefer the AI beneficiary whose revenue needs other people to deploy AI — and price the second-order effect, not the product

1. When a company launches an AI product, ask which line of its P&L actually benefits: the new product's own revenue, or the existing product that AI deployment forces customers to expand.
2. Test for the second-order effect with three questions: does deploying agents create a new problem the incumbent already solves? Does it raise the deal size of the existing platform? Does it pull forward a modernization project that was previously deferred?
3. Look for the corroborating metrics on the old business, not the new one: attach rate of newer modules to bookings, average deal size uplift, growth in the largest-customer cohort, and the contracted-revenue metric.
4. Treat management candour about immaterial AI revenue as a positive signal, not a negative one — it means the growth being reported is coming from somewhere verifiable.
5. Size the opportunity off the installed base being modernized, which is knowable, rather than off the new category's TAM, which is not.

• The new-product share of bookings stalling (the modernization wave has passed); deal-size uplift fading; and the moment a platform vendor starts guiding on a standalone AI revenue line, which converts a durable second-order story into a competitive first-order one.

3. Strip the acquisition out of the growth rate — then insist on the organic number the company was criticized for last quarter

1. When a headline growth rate jumps, first ask what was bought. Find the acquisition's revenue contribution and restate growth without it.
2. Identify which specific organic line was the bear case in the previous quarter, and check whether the company disclosed that exact line again. A company that quietly stops breaking out the number it was criticized for is answering the question.
3. For an acquisition, separate three distinct claims: revenue contribution (arithmetic), margin impact (does it dilute, and is the offset specified?), and cross-sell (the only one that creates value, and the last to appear).
4. Check whether integration has produced joint products yet, not just combined revenue — a shipped combined product is the first real evidence the thesis is more than a bolt-on.
5. Note the dilution defence precisely: "operating leverage and sales efficiencies fully offset the dilution" is a testable claim, so put it beside the unchanged margin guidance and check it next quarter.

• The quarter the acquisition annualizes into the base and headline growth collapses to organic; margin guidance quietly trimmed a quarter or two after the "fully offset" claim; and cross-sell still being evidenced by one anecdote three quarters later.

4. Strip the tax item and the below-the-line gain before you read EPS at all

1. Whenever EPS is startlingly large relative to revenue, walk down to the tax line and the other-income line before comparing to consensus.
2. Common one-offs to isolate: a valuation-allowance release or other non-recurring tax benefit, investment or fair-value marks, gains on consolidation, legal settlements, tariff or duty refunds.
3. Re-underwrite on operating income and free cash flow, which these items do not touch, and treat the headline EPS as uninformative rather than good.
4. Apply it symmetrically — a non-operating loss depressing EPS is equally meaningless, and is far more likely to be mispriced because the market reacts to the miss.
5. Where the item is a tax benefit, check whether it changes the go-forward tax rate; if not, next year's EPS comparison will invert for purely mechanical reasons.

• The following year's EPS comparison inverting on the missing benefit; a company quoting the boosted EPS in its own headline; and repeat "non-recurring" items, which stop being non-recurring by the third appearance.

5. When a company cuts its own price, separate the deliberate reset from the eroded moat — and audit any metric redefinition arriving with it

1. Find management's stated diagnosis of churn. A company that names price as its number one churn reason has just told you its pricing power is being tested, whatever the strategy it announces next.
2. Separate the two possible readings and write down what would distinguish them: a deliberate reset should come with a specified reacceleration date, a target customer count, and a monetization path for the cheaply-acquired users; structural erosion comes with a lowered guide and no such specifics.
3. Quantify the sacrifice explicitly — the growth rate of the reset product versus the group — so the cost of the strategy is on the page rather than buried in a range.
4. Check whether the same playbook is being extended to other products. Repetition is evidence of a considered strategy; it is also evidence the same pressure exists in more than one market.
5. Independently, run the redefinition audit: if a headline metric's definition changes in the same print, refuse to compare it to prior consensus until a bridge is published. Note the direction — a redefinition that makes the company look worse (adding a real cost back in) is far more credible than one that flatters.

• The FY27 TurboTax growth rate against the 2%–3% guide (undershooting turns "deliberate" into "structural"); free-tier conversion rates actually disclosed rather than the free user count; Mailchimp's flat-to-down trajectory; and whether the SBC-inclusive EPS bridge is ever published.

6. Decompose any growth number into price and volume — inflation-led growth is borrowed

1. For any beat driven by one category, get the two components separately: average selling price and unit volume. Growth from price alone is a different asset than growth from units.
2. Identify who caused the price change. Price rises the company chose (mix, premiumization) behave differently from price rises passed through from an input cost — the pass-through kind stops the moment the input normalizes and can reverse demand while it lasts.
3. Trace the input cost upstream. Component inflation visible in a supplier's earnings will surface one or two links later in a retailer's comparable sales; the same shortage can read as a positive in one company's print and a fragility in another's.
4. Check the comparison base — growth achieved while lapping a one-time event is worth more than the same number on an easy base.
5. Then locate the structurally better dollars: marketplace fees, advertising, services — revenue that does not consume inventory — and ask whether they are growing fast enough to matter when the price effect ends.

• PC unit volumes turning positive (the growth becomes real) or ASPs flattening while units stay negative (the comp collapses); memory prices normalizing; and Marketplace + Ads growth against the $1.3B target as the computing comparisons harden.

7. When a supply constraint caps a business, track the adoption rate of the workarounds — that separates a timing problem from a demand problem

1. Name the physical constraint precisely and identify the step in the sales motion it blocks (here: a customer must buy new servers before it can deploy the software).
2. List every path the company offers that bypasses that step — running on hardware the customer already owns, a cloud-hosted version, a partner integration — and demand a disclosed growth rate for each.
3. Test with the leading recurring metric, not revenue: if ARR accelerates during the constraint, demand was never the issue and the workarounds are converting.
4. Separate the constraint's expected duration from its effect. "Lead times stabilizing" but "shortages and price increases persisting" with "customers delaying contract start dates" is a timing statement — pushed revenue, not lost revenue — provided bookings keep growing.
5. Re-check next quarter whether the workaround mix is a permanent shift (which may change margins and competitive exposure) or a temporary detour that reverses when supply normalizes.

• ARR decelerating while the workarounds keep growing (which would mean the detour is cannibalizing, not converting); contract start dates slipping further right; and whether NC2's cloud mix carries a structurally different margin than the on-prem motion.

8. Read the recurring-revenue metrics in the order they turn — and trust the backlog over the revenue line

1. Rank the disclosures by how early they move: net-new ARR and bookings first, then cRPO / current backlog, then total RPO, then reported revenue (which is largely last year's bookings arriving).
2. Judge an inflection on the leading metrics and use revenue only to confirm. A company reporting 13–14% revenue growth on 8% backlog growth is reporting the past.
3. Watch the second derivative on the leading metric: ARR growth accelerating (not merely high) is rare as a base scales, and is the strongest single signal in this family.
4. Compare growth rates across the stack in the right direction — cRPO above revenue means the business signed more than it billed; the reverse means the reported number is being harvested from an older book.
5. When a company guides growth down and margin up in the same breath, treat it as a declaration of which line it will defend, and model the margin promise as the reliable half.

• Workday's backlog growth converging with (or staying below) revenue growth; Rubrik's ARR growth as the base crosses $2B; Elastic's cRPO holding above 20% now that the reacceleration is guided back-half weighted; and any company that changes which backlog metric it emphasizes.

9. Ask what is producing a record margin — cost removed or revenue added — then check the compounding engine underneath

1. When margin hits a record, attribute it: price increases, mix shift, operating leverage on growth, or costs structurally removed (automation, headcount, service deflection).
2. Automation-driven margin is real but finite — a call deflected once cannot be deflected twice. Ask what percentage of the addressable cost has already been automated; the closer to the ceiling, the less runway remains.
3. Then find the engine that compounds — the asset base, the float, the installed base — and read its growth rate. Efficiency gains are a stock; engine growth is a flow.
4. If the engine is decelerating while margin expands, the earnings growth has a horizon. Model the point where automation savings are exhausted and earnings converge back toward the top line.
5. A record margin printed alongside a falling share price usually means the market has already made this distinction — go looking for the decelerating engine rather than assuming an overreaction.

• The AI deflection rates approaching their ceiling (85% cannot become 170%); HSA asset growth stabilizing or falling further; Marketplace revenue finally becoming material; and margin expansion decelerating faster than revenue, which is when the story ends.

10. When regulation removes a structural advantage, re-underwrite the business model — not the growth rate

1. Identify whether a business's unit economics depended on a rule rather than on an operating capability — a duty exemption, a licensing gap, a classification loophole, a subsidy.
2. If so, treat the rule's removal as permanent and re-derive the unit economics without it, rather than modelling a temporary growth dip.
3. Read management's mitigation for what it costs structurally, not just in dollars. Building local warehousing to replace direct-from-origin shipping converts an asset-light model into an asset-heavy one — lower returns on capital even if revenue recovers.
4. Check the domestic/core business separately; a company under regulatory pressure abroad and competitive pressure at home has no engine left to carry the transition.
5. Distinguish capacity to fund the transition from evidence the transition works. A large cash pile answers solvency, never returns — the same test this source applies to AI capex.

• Whether Temu discloses margin or fulfilment cost per order in the localized markets; Online Marketing Services growth (the domestic health check, and the higher-margin half); the first-party brand initiative getting a stated timeline; and any further jurisdictions adding low-value import fees.

11. When a beat-and-raise still falls, ask what the guidance added — not what it said

1. Reconstruct what the market already assumed before the print: the forward multiple, and specifically which announced opportunities were already inside the prior guidance.
2. Compute the incremental information: a raise that merely confirms a previously-embedded opportunity adds enthusiasm without adding numbers.
3. Read partnership announcements for whether they change the forecast or re-describe it. Look for the structural commitment (a warrant, a take-or-pay, a multi-year contract) as the part that is genuinely new.
4. Put the multiple beside the growth durability rather than the growth rate — at a high multiple, the question is how many years of the current rate are already paid for.
5. Then decide explicitly whether your disagreement with the market is about the business or about the price; they demand different evidence, and only the second is answered by valuation.

• The H2 FY27 custom-silicon acceleration actually appearing on schedule, and FY28 custom revenue doubling as promised; whether the warrant translates into disclosed program wins; and the forward multiple compressing on a flat price, which is how an over-owned name digests its own good news.