Cash and working capital: the three levers that move the balance sheet

Cash levers get less attention than margin levers because they do not compound the same way, and they deserve more, for two reasons. Days sales outstanding is a number the company can move inside a quarter with a purchased tool, which makes it the right first proof at the day-one-hundred review. And in a vertical whose customers bill a third party, the customer's revenue cycle is the company's product: an agency that gets paid faster because of the software will pay for the software.

C1 · Billing and collections

What it is. Pre-billing checks that reconcile usage, contract terms and price changes before the invoice goes out, because inaccurate invoices are the leading cause of late payment; collections agents that confirm delivery, send personalized reminders, capture disputes, propose payment plans and escalate; cash application and dispute handling by agent; and, for card-billed revenue, retry logic and card-account updating.

The line it moves. Working capital (days sales outstanding); bad debt; involuntary churn on card-billed subscriptions.

Typical range and the evidence. Strong, on an independent case. DSO from the mid-fifties toward the high thirties is the case-study range, against APQC's benchmarks of a 38-day median and 30 or under for top performers.¹ Nucleus Research's 2026 case for HighRadius at an unnamed office-solutions provider reports DSO "reduced ... by approximately 27 percent, from 80 days to 58," an independent analyst's figure with an anonymous customer.² Tesorio's customer pages report Couchbase cutting DSO by ten days, Seismic by about 30%, and Discovery Education's K-12 segment "from 128 days to 43 days" over a year.³ Chaser reports an average 14-day DSO reduction across its users in 2024, aggregate telemetry with no baseline and the selection effect that companies with the worst DSO adopt first.⁴ On card-billed revenue, Stripe reports its retry logic recovering "$9 in revenue for every $1 customers spend on Billing," with Deliveroo recovering more than £100 million and Retool more than $600,000 using it with other recovery tools.⁵ The most useful line in the evidence is old and mechanical: at Zapier in 2021, machine-learning retry logic lifted authorization rates by 1.24% and the card-account updater by 2.76%; the plumbing beat the model by more than two to one.⁶

Conditions. A billing system that can produce accurate invoices from usage and contract data (O6's general ledger and P7's warehouse); collections communications reviewed for tone and never sent to accounts with open service disputes; a human on every payment plan and every write-off.

Kill criteria. DSO not down five days after two quarters; or any collections message sent to a disputed account.

Horizon and stage. Deploy. Day one hundred to year one.

How it translates. A: agencies pay slowly because payers pay them slowly; C2 is the deeper lever. B: enterprise customers on annual invoices; the pre-billing check matters most. C: card-billed merchants; retry logic and the account updater carry the card. D: self-serve card billing at volume; the largest involuntary-churn effect. E: large annual invoices with milestone billing; the dispute capture matters most.

C2 · Revenue cycle in the vertical

What it is. Claims scrubbing, eligibility and benefits checks, prior-authorization assembly and submission, denial prediction and appeal drafting, run for the customer inside the product. It is a product lever (R2) whose value is measured in the customer's cash, and it belongs in the cash chapter because that is what the customer buys it for.

The line it moves. The customer's cash and denial rate; for the company, product revenue (R2) and retention (R5).

Typical range and the evidence. Strong for the size of the problem, weak for vendor outcomes, and the gap between the two is the lever. KFF's August 2026 analysis of insurer-reported prior-authorization metrics found appeal overturn rates of 67% in Medicare Advantage, 47% in Medicaid managed care and 43% in the Marketplace: roughly half of denials are wrong, and appealing is a labor problem.⁷ The peer-reviewed side of the same story is that payers adopted the technology first: UnitedHealthcare's post-acute denial rate rose from 10.0% in 2020 to 22.7% in 2022, and its algorithm is alleged in litigation to have had "over 80% of the prior authorization denials being reversed on appeal."⁸ Experian's 2025 survey found "41 percent of providers now face denial rates of 10 percent or higher."⁹ HFMA and AKASA's 2025 survey of 519 finance leaders found 80% of health systems exploring, piloting or implementing generative AI for the revenue cycle, and the adoption gap that is the value-creation thesis in one line: 64% of large systems piloting or implementing against about 20% of small ones.¹⁰ The vendor outcomes are thin: Availity says 80% of prior-authorization requests through its product are "touchless"; Waystar reports appeal-package creation "cutting time from 38 hours to two" and early adopters "overturning 40% more denials," anonymously; Cleveland Clinic's flagship 2025 partnership announced document-processing speed and no financial outcome; and Olive AI, once valued at $4B for revenue-cycle automation, sold its assets to the incumbent it was built to displace.¹¹ No independently verified outcome data exists for the prior-authorization and denials AI category as of September 2026, and the card says so.

Conditions. Payer rules and portal access; a covered model endpoint, because every record is protected health information; a human signing every appeal and every medical-necessity claim; the state disclosure rules from X2; design-partner customers on a measurement protocol.

Kill criteria. First-pass acceptance at design partners not up two points after two quarters, which for an agency is worth more than the subscription.

Horizon and stage. Invent. Years one to two.

How it translates. A: the archetype and the largest module in its bridge. B: the regulatory equivalent is the filing, not the claim; a cousin of this lever. C: invoicing and collections for the merchant are the vertical's revenue cycle, and R8 carries them. D, E: not applicable.

C3 · Payments timing and float

What it is. Settlement timing, instant-payout pricing, reserve policy and the interest on funds in transit, once R8 exists and the platform sets the merchant's terms.

The line it moves. Working capital and a fintech revenue line.

Typical range and the evidence. Medium. Stripe's published platform pricing is the only primary source: an instant-payout markup of "1% of payout volume," funds routing at "0.25% of payout volume," and "$2 per monthly active account" where the platform handles pricing.¹² Those are the unit prices of the lever, and the platform's take on them is what the merchant will pay for speed. Toast's disclosure that non-payments fintech contributed 11 basis points of take rate in a quarter includes some of this.¹³ No platform below Toast's scale publishes float economics.

Conditions. R8 at facilitator or near-facilitator terms; a treasury policy; regulatory review of reserve and payout practices by state; the merchant's terms disclosed.

Kill criteria. Any month in which reserves held exceed policy, or any regulatory finding on payout practice.

Horizon and stage. Invent. Years two to three.

How it translates. C only.