Insights · FP&A and forecasting

The SaaS metric set is the patient funnel wearing different clothes

Jatinder Purewal · ACMA, CGMA · Executive MBA · 18th August 2026 · 19 min read

Cohort curves are persistence curves, churn is discontinuation, and a total addressable market is epidemiology measured with less discipline. I built the patient version inside pharma and I run the consumer version on my own P&L today, and the skeleton they share is a working tool rather than a figure of speech.

In the patient-based forecasting piece I made a promise: that the same cascade would travel into consumer, subscription and technology businesses properly, and that the software-as-a-service (SaaS) metric set and the patient funnel would turn out to be closer relatives than either side tends to admit.

This is that piece. The claim is not that patients are customers. The claim is structural: both worlds forecast revenue by building it from the unit that causes it, through a chain of rates, and the chains map onto each other at the stages where the money is actually decided. Learn the skeleton once and you can read either industry's numbers quickly. The mapping is not an equality: several stages break in instructive ways, and I name them.

A note on scope: this piece is written from practice on both sides, and the two sides are not symmetrical. I built patient-based models inside pharmaceutical businesses for more than a decade, and I run the consumer version today on my own company's numbers, which is a direct-to-consumer durable-goods business rather than a subscription one. So where subscription-specific metrics appear below, I am reading them with pharma eyes and against my own cohort and unit economics, not claiming to have carried a software annual recurring revenue (ARR) book.

Recurring revenue itself is where I started, twice over. My first group finance seat was at IMS Health, where the product pharmaceutical companies bought, MIDAS among them, was itself a recurring data subscription, and revenue recognition on that book sat in my remit. Then came subscriber-level value models at O2 in the early 2000s, across pre-pay customers and post-pay contracts, when a subsidised land grab only worked if finance could tell which subscribers were worth the subsidy, what mobile then called subscriber acquisition cost, and when churn and average revenue per user (ARPU) were operating metrics in mobile before software borrowed the vocabulary.

The funnel I describe is the self-serve consumer kind; enterprise SaaS adds procurement, contracts, implementation and seat deployment, and I say so where it changes the answer. The metric conventions belong to their industries and their originators, and I cite the standard sources at the foot; none of the frameworks here are mine. What is mine is the mapping, the places where I have watched it break, and the judgement calls, which I flag as judgement.

The skeleton, briefly

A demand-unit model refuses to start from the revenue line. It is what FP&A practice calls driver-based planning, with the driver chosen honestly: the unit that causes the revenue, not the line that reports it. The model builds the number from a population, a set of rates that turn the population into paying units, and a value per unit. In pharma the units are patients and the chain runs from epidemiology through diagnosis, treatment and brand share down to net revenue per patient. In a subscription or consumer business the units are customers, and the chain runs from an addressable market through acquisition, activation and retention down to revenue per customer. I set out the pharma version in full in the patient-based piece; this article walks the same spine on the other side of the street.

The reason the mapping matters is not intellectual tidiness. It is that every line of the map is a question a finance leader already knows how to ask in one industry, and can therefore ask on day one in the other.

The key mapping: epidemiology to TAM, diagnosis and treatment to activation and conversion, brand share of the patients being started or switched to win rate on new business, persistence to cohort retention, net revenue per patient to ARPU and margin-based LTV - the same skeleton, two vocabularies

Population: epidemiology and the addressable market

Pharma starts from epidemiology: how many people have the condition, measured by science that is imperfect but usually external to the sponsor, published, and argued over in public by people whose incentives at least differ from the sponsor's. Commercial models rarely run on pure registry science: syndicated vendor epidemiology, commissioned burden-of-illness work, claims data and expert judgement all feed in, and in rare disease the estimates are soft enough that two credible sources can differ by a multiple.

A subscription business starts from the total addressable market (TAM), and here is the first honest difference: TAM is usually measured by whoever is trying to raise money on it. The discipline transfers exactly; the data quality does not. When I read a TAM slide now I ask the question epidemiology taught me: who counted this population, with what definition, and what would they have had to believe for the number to be wrong? A population nobody independent has counted is not a population. It is a hope with a denominator.

Both numbers sit at the top of the cascade, and each is the least controllable and most-argued cell in its model. Neither, though, is the number the model actually uses. Pharma narrows epidemiology to the diagnosed, eligible and addressable population, which is the stage that decides whether a model is credible, and the subscription version narrows TAM to the customers who fit the profile and can be reached through channels you can actually buy.

Then the mapping breaks. Underlying biology usually changes slowly, at the pace of disease and demography; the operational definition can change overnight, and the definition is what the model consumes. The 2017 American College of Cardiology and American Heart Association (ACC/AHA) blood-pressure guideline moved US hypertension prevalence from roughly 32 per cent of adults to roughly 46 per cent in a single publication, reclassifying about 31 million people overnight. A TAM can be re-cut just as fast by a category shift or a change in pricing, regulation or market definition, so both versions need re-basing more often than either side likes to admit.

The middle rates: diagnosis and treatment, activation and conversion

In the patient cascade, having the condition is not the same as being diagnosed, and being diagnosed is not the same as being treated. Each step is a rate, each rate has an owner, and the gaps between them are where commercial effort goes. The subscription equivalents are direct: awareness is not sign-up, sign-up is not activation, activation is not payment. The funnel leaks at every joint, and the model's job is to name each joint so that effort and money can be pointed at the leakiest one.

The structural difference is who decides. In pharma the decision at the critical joint is distributed rather than singular: a prescriber writes it, but guidelines, formulary, payer and patient all constrain what is written and whether it is filled, and in most of the world the manufacturer cannot promote the brand to the patient at all, since direct-to-consumer advertising of branded prescription medicines is broadly permitted only in the United States and New Zealand.

In self-serve consumer subscription the decision is mostly the end user's own, made in seconds, against alternatives one tap away; in enterprise SaaS it looks far more like pharma, with procurement, a business case and an implementation standing between the decision and the revenue. The self-serve funnel is faster, cheaper to test, and far more volatile; the pharma funnel is slower, dearer to move, and stickier once moved in chronic therapy, though oncology, acute treatment, tolerability or a formulary change can end it quickly. Same skeleton, different physics, and a forecast that ignores the physics will be wrong in a predictable direction: consumer models under-estimate volatility, pharma models under-estimate inertia.

One structural nuance keeps the mapping honest. Pharma's diagnosis and treatment rates are market-level: they happen before any brand is chosen, and the brand then competes only for its share of the treated pool. A self-serve funnel is brand-level from the first click, because the customer discovers the category and the product in the same session, so category penetration and brand share collapse into one funnel. That is why the two industries' conversion rates are not comparable: they are not measuring the same span of the decision.

Two funnels, one shape: population to diagnosed to treated to on-brand in pharma; addressable to acquired to activated to paying in subscription - the joints leak in both, and every joint is a rate with an owner

The heart of it: persistence is retention

This is the mapping that made me trust the whole exercise: each side has something real to teach the other.

Pharma splits staying-on-therapy into named parts. Adherence is the umbrella process, and it runs in three phases: initiation, whether a prescribed patient ever takes the first dose; implementation, whether the doses that follow are taken as prescribed, which the older literature calls compliance; and discontinuation, when the patient stops. Persistence is the measure that spans them, the length of time from initiation to the last dose before discontinuation. A patient-based model that assumes one flat adherence rate across a portfolio is indefensible, because the true number is a function of the condition, the regimen, the population, the measurement method and the horizon over which it was measured.

Subscription businesses split the same behaviour the same way: activation is initiation, engagement is implementation, cancellation is discontinuation, and retention measures the same duration persistence does. Their best practitioners run it with a rigour pharma would recognise instantly: cohort vintages, not a blended average. A cohort retention curve is a persistence curve. Gross revenue retention (GRR) is that curve with a wallet attached: how much of an opening revenue base is still there at the end of the stated period after churn and downgrades, and like a persistence curve it cannot exceed one hundred per cent.

One caution matters more than it looks. As reported, GRR and NRR are struck on the whole revenue base as it stood at the start of the period, usually trailing twelve months: a single blended number, one point read off a mixture of tenures, not a curve. The true twin of a persistence curve is revenue retention by acquisition cohort against tenure; a headline GRR is a portfolio-wide one-year persistence rate, which no serious patient modeller would accept on its own either.

Net revenue retention (NRR) is one of the metrics investors scrutinise most closely when they price a subscription business, and it adds the one component pharma structurally lacks: expansion, existing customers paying more. That is why a good NRR figure sits above one hundred per cent, and why the honest pharma twin is the gross measure, not the net one.

Revenue per patient is bounded by the approved regimen, clinical practice, access and duration on therapy. The nearest analogues to expansion, up-titration, an added indication, a combination add-on, are real but rarely commercially discretionary. Anyone who has drawn a persistence curve for patients has already drawn the gross version of that chart; the axis just said patients. The tightest modern mapping is not seat-based software at all but consumption pricing, where revenue is usage intensity multiplied by duration, which is precisely pharma's persistence multiplied by dose intensity.

The habit has a practical edge when a deck or information memorandum leads with a retention number. Ask which base it is struck on, the whole book or a chosen cohort of survivors; which twelve months it covers, trailing or a flattering quarter annualised; what the gross figure is underneath the net one, because expansion from the best accounts can bury churn everywhere else; and whether the number survives being recomputed from the billing ledger rather than the reporting pack. Four questions, an afternoon with the right access: persistence-curve reflexes pointed at an investor deck.

The mapping breaks here, and the break is instructive in both directions. Patients discontinue for reasons that include tolerability and clinical outcomes and, just as often commercially, access and cost: a formulary change, a non-medical switch, a lapsed prior authorisation, a co-pay shock. Customers churn for the price-driven reasons that rhyme with those, voluntarily and also involuntarily through failed payment, account closure or the customer's own business failing.

A subscription cohort curve can bend upwards again through win-backs and expansion; a persistence curve cannot, and the reason is construction rather than behaviour. A conventional persistence curve is drawn as time to first discontinuation, so it can only fall. Patients do restart, in some therapy areas at high rates, and pharma models restarts and second episodes separately rather than letting the curve rise. And pharma's version carries a harder floor: stopping a therapy carries clinical consequences that no commercial churn does, including churn from software a business genuinely depends on.

The stopping event is not always observed, either. Subscription churn is contractual and becomes visible at cancellation, though the disengagement behind it starts earlier; in a marketplace-majority consumer business it is latent, inferred from purchase gaps, which is why my own repeat figures are a range rather than a rate; pharma sits between the two, inferring discontinuation from refill gaps against a permissible gap that the literature expects to be tested rather than assumed.

One practical consequence of the mapping: in both worlds the middle rate is the leading indicator. Implementation predicts persistence and engagement predicts retention, which means the curve reported at the end of the year was largely legible from behaviour visible in the first weeks.

A cohort curve read both ways: persistence in pharma, retention in subscription - gross revenue retention is one point on that curve with a wallet attached, capped at one hundred per cent, while win-backs and expansion are the upward bend only the consumer revenue curve can take

Value per unit: gross-to-net is gross-to-net

The bottom of the pharma cascade converts patients into money through net revenue per patient, and the word doing the work is net: list price is rarely the price anybody realises, and the real number emerges after rebates, chargebacks, statutory and confidential discounts, prompt-pay terms, distributor fees, returns and co-pay assistance. Free goods, patient assistance programmes and compassionate use sit outside that bridge, because unlike co-pay assistance they reduce no realised price: they consume patients without generating a gross sale at all, so they dilute revenue per patient rather than revenue per unit, and a model that nets them off price is double-counting.

The subscription equivalent is ARPU, which is already struck on recognised revenue and is therefore post-discount by construction, so the bridge that matters runs from list to realised price after promotions and coupons, then returns, refunds, chargebacks and service credits. Platform and marketplace fees sit outside that bridge on the usual facts: a seller that controls the goods before transfer is the principal under IFRS 15, recognises the gross consideration and carries the platform's fee as a cost, so netting it off revenue misstates the top line while leaving contribution unchanged. Which side of that test you fall on is a conclusion to reach, not an assumption to inherit. Free tiers belong nowhere in that bridge; they change the denominator, which is why the discipline separates ARPU from average revenue per paying user (ARPPU).

One more line of the mapping runs past contribution to cash, which is where an interim mandate usually starts. Pharma carries channel inventory and gross-to-net accruals that settle long after the sale is recognised; subscription carries bookings that become billings, billings that become deferred revenue, and deferred revenue that becomes cash on its own schedule, with contracted annual recurring revenue and live annual recurring revenue rarely the same number; consumer carries inventory and payment terms. Three vocabularies, one question: how far does the cash sit from the revenue it belongs to, and who owns the gap?

One more discipline travels with the bridge. Net revenue per patient is a revenue measure, so it pairs with ARPU for a period, not with lifetime value. Lifetime value (LTV) needs two further steps the mapping must not skip: duration, which is persistence, and gross margin, which is why the formula usually quoted puts margin in the numerator and churn in the denominator, on the assumptions that churn is constant, ARPU is stable, the periods match, and a pound of contribution in year four is worth a pound today. An LTV quoted on revenue rather than gross margin is the error an investor reader spots first, and it matters more in pharma's direction than in software's, because classic software gross margins sit near the top of the range while pharma carries cost of goods and royalties, though that gap is closing where inference costs have moved into software's cost of goods.

One correction runs the other way, from pharma to subscription. The lifetime-value shortcut, gross margin per customer per period divided by churn, is a geometric series, and a geometric series assumes the churn rate never changes. Cohorts do not behave that way: measured retention rates rise with tenure, because the churn-prone leave first and the survivors are a progressively different mixture, which is the Fader and Hardie result. A pharma modeller does not get to assume a flat hazard on a persistence curve without defending the choice, which is why persistence is fitted rather than averaged. A lifetime value built on one churn rate makes exactly the error an exponential fit makes on a persistence curve.

The break: pharma's payer is usually not the patient, and reimbursement dynamics, formularies and payer mix have no clean consumer equivalent. Nor, in most markets, does the manufacturer control the price it actually realises. It may set or propose a list price, but statutory and negotiated schemes decide what is reimbursed and therefore what is realised: VPAG and NICE in the United Kingdom, a six-month free-pricing window in Germany before the negotiated AMNOG amount applies from month seven, and the maximum fair price on selected drugs under the US Inflation Reduction Act.

That layer is itself moving. As at August 2026, the United States has been importing reference pricing directly, through most-favoured-nation agreements announced with the large manufacturers and a federal direct-to-consumer sales channel alongside them, while separately opening a trade investigation into Germany's pricing of innovative medicines as an unfair practice. A payer-access layer that trade policy can redraw is not a fixed parameter in anybody's model. Consumer pricing is materially more unilateral by comparison, though retailers, marketplaces, app stores and competition law all constrain it. That difference accounts for a substantial part of what does not transfer between the two P&Ls, alongside gross margin structure, revenue recognition, inventory, and the economics of patents and research. And it is why a pharma commercial model carries a whole layer, payer access, for which the consumer version has only rough analogues: procurement, channel partners, app stores and benefit sponsors all gatekeep, but none of them sets the price of the product for an entire market.

One cohort, two answers

The blended-average trap is easier to show than to describe. Here is an illustrative cohort: round numbers, the shape of a real curve.

A business acquires 1,000 customers in January at a customer acquisition cost of £40 each: £40,000 of spend. The product is £10 a month at a 70 per cent gross margin, so each paying customer contributes £7 a month. Company-wide, blended revenue churn runs at 5 per cent a month: one figure struck across every tenure in the book, from last month's joiners to cohorts three years settled. The textbook lifetime value is £7 divided by 5 per cent: £140 a customer, a 3.5 multiple of the acquisition cost. The deck writes itself.

Now follow January's cohort instead of the blend. Real cohort curves fall fastest at the start: 15 per cent of the cohort gone in the first month, the monthly loss rate then easing steadily to under 3 per cent by year end, with roughly half still paying at month twelve. Count the paying months that actually occur and the cohort delivers about 7,700 of them in its first year: roughly £54,000 of contribution against £40,000 of spend. Payback lands in month nine, and the year-one multiple is 1.35, not 3.5.

The £140 lifetime value is not wrong. It is unfunded: everything beyond 1.35 rests on how the surviving half behaves in years two, three and four. On a curve of that shape the £140 is not collected until somewhere near year four, undiscounted, and a discount rate pushes it further out still. That is exactly the part the blended rate assumes and a cohort curve exists to prove. A patient modeller recognises the move at once, because it is how persistence has always been handled: never the portfolio average, always the fitted curve, with the tail priced separately.

One cohort, two answers: the blended read gives £140 of lifetime value and a 3.5 multiple of acquisition cost; following the January cohort gives payback in month nine and a year-one multiple of 1.35, with everything beyond that resting on a tail the blend assumes and the curve must prove. Illustrative numbers.

The trap both sides share: the window

One analytical trap runs identically on both sides, and I have been caught by it on the consumer side. Any rate measured inside a window is only true of that window. In the patient world this is the censoring problem: measure persistence over a quarter and you will miss the slow attrition that shows at a year. In my own business, which sells a considered, durable purchase, the honest repeat-purchase figure inside a twelve-month cohort window runs between ten and twenty per cent depending on the cohort, and for a while I read that as a low repeat rate. It is a correct number that answers a different question.

The product I sell is replaced on a cycle measured in years, so the twelve-month figure is a complete twelve-month rate and an incomplete observation of lifetime repeat behaviour: in a time-to-next-purchase analysis, every customer who has not yet repeated at the data cut-off is right-censored. It is the right number for a payback decision, because you cannot spend against a repeat you will only collect in year four, and the wrong number for a lifetime claim.

That is the same error patient modellers guard against when they set an observation horizon that fits the therapy and use survival methods for the patients whose follow-up has not finished, rather than reading a quarter as a lifetime. The discipline transfers untouched: name the window, match it to the natural cycle of the thing being measured, and never let a censored observation masquerade as a lifetime figure.

The window trap, identical on both sides: a rate measured inside a short window censors a curve that runs for years - persistence measured at a quarter, repeat purchase measured at twelve months on a replacement cycle measured in years, each a correct short-window number that must never be read as a lifetime truth

The unit is a person

Underneath every rate in the cascade sits a discipline no spreadsheet supplies: get close to how the person behind the unit actually behaves. In pharma that means the realities under the adherence data, regimen burden, side effects, refill friction, the gap between how a therapy is prescribed and how it is lived with. On my own P&L it is unglamorous: reviews, return reasons and customer messages, read as behavioural evidence rather than noise.

At O2 it was usage, the behaviour inside a tariff that the value models stood on. The point of getting close is causal: not the churn rate but the churn reasons, ranked, owned and priced. And the tooling for this has just changed profoundly: review text, return reasons, support threads, even a thumbs up or down inside a product, can now be read at scale with AI and turned into a churn-driver taxonomy rather than an anecdote file. I do that reading on my own numbers with Claude and ChatGPT, on data stripped of anything personal.

It matters most at the top of the cascade, in the sizing, because an untested assumption there compounds: get the base wrong and every rate below it is a percentage of an error. Call it empathy used as an audit tool; I flag it as judgement because that is what it is.

Three eras, one skeleton

I did not learn this mapping from a book. The continuity is the point of telling it.

It started for me at BT Cellnet, which became O2 in 2002, before pharma: subscriber-level value modelling, customer lifetime value by tariff, by device and by sales channel, built when the mobile industry was in its land-grab phase. That was my first behavioural-unit model. When I moved into pharmaceuticals at Chiron in 2003 and stood up long-range planning, the discipline translated: the unit became the patient, the rates became diagnosis, treatment and adherence, and at GSK the same cascade fed the forecasting inside transaction valuations, which I tested, reconciled and defended rather than built.

The pharma decade did not stop at GSK. At Parexel I ran finance for the Early Phase unit within Clinical Research Services, forty people across five sites and three continents, and the same cascade sat underneath the turnaround: that Early Phase unit's first positive operating income in its history, a $13m improvement on plan, with receivables cut from sixty-five days to forty.

And for the last eleven years I have run the consumer version on my own company's P&L: acquisition cost against lifetime value, contribution margin per order, and cohorts measured as each channel's data allowed, which on a marketplace-majority business means the platform only began handing brands repeat-purchase data years into the eleven. My own capital sat behind every assumption throughout.

Subscribers, then patients, then customers. The vocabulary changed twice. The skeleton never did, and it is scored on this month's numbers as I write.

Three eras, one skeleton: O2 subscribers by tariff, device and channel in the early 2000s; patients through diagnosis, treatment and adherence from 2003; customers through cohorts, CAC and LTV on my own P&L today

What a finance leader should do with the mapping

Strip the sectors away and the working tool is a set of questions that travel.

First, find the unit, and refuse revenue lines that will not decompose into one. If a plan cannot say what the unit is, how many there are, and what one is worth after the gross-to-net bridge, it is not a plan; it is a target with formatting. And ask the two questions that turn a unit value into a decision: how long the customer acquisition cost (CAC) takes to come back, and what multiple of it the unit is worth over its life. Pharma asks the same pair in its own vocabulary, as time to break even on launch investment against net present value per patient started. The pair maps; the shape of the spend does not. A consumer business incurs its acquisition cost at the margin, per customer, in the month it spends it. Pharma's defining cost sits years upstream in research and development, including the candidates that failed, sunk before the first patient is started: launch spend exists on both sides, but the R&D mountain has no consumer equivalent. It is why consumer finance thinks in payback per cohort and pharma in net present value per asset. The same discipline is owed to the denominator: a lifetime value struck on margin proves nothing against an acquisition cost that quietly excludes salaries, agency fees, discounting or returns. Fully loaded or direct marketing only: say which, and never mix the two inside one comparison.

Second, demand cohort vintages rather than a blended point average, for anything that retains. An average retention or adherence figure across a mixed base is the single most common way both industries flatter themselves. The curve, by cohort, tells the truth the average was built to hide.

Third, interrogate the window on every rate. Ask what natural cycle the thing being measured actually has, and whether the measurement window covers it. Most flattering numbers are truncations.

Fourth, respect the physics that do not transfer. The skeleton lets you move between the worlds; the physics tell you what to re-learn when you arrive, and pretending otherwise is how sector tourists get found out.

In practice those four principles compress into a first week of questions, short enough to ask across a desk: the unit, the cohort vintages, payback by acquisition month, margin-based LTV and the churn that funds its tail, the CAC definition, involuntary churn and its owner, gross-versus-net recognition, and the window on every headline rate. The eight I use are on the card below, and none of them needs sector experience to ask.

The first week, in questions: eight questions that decompose any consumer or subscription P&L - the unit, cohort retention, payback by month, margin-based LTV, the CAC definition, involuntary churn, gross versus net recognition, and the measurement window

What I would not claim is the sector reflex: the systems, the regulatory habits and the local commercial folklore that only come from being in the building. That is the first month's work, and it is specific. Sit with whoever owns each rate. Rebuild one cohort from source data rather than from the deck. Find the number the last forecast missed, and ask why it missed.

Finance leaders move between sectors, and business models evolve faster than core finance disciplines do, which is why the method is the thing that should travel.

In brief

Are SaaS metrics really equivalent to the patient funnel? Structurally, stage by stage: epidemiology maps to addressable market, diagnosis and treatment to activation and conversion, brand share of the patients being started or switched to win rate on new and switching customers, persistence to cohort retention, with initiation, implementation and discontinuation as activation, engagement and cancellation, and net revenue per patient to ARPU after the gross-to-net bridge. The overlap is in the skeleton, not the physics.

What is net revenue retention, in the pharma frame? Gross revenue retention is persistence with a wallet attached: how much of an opening revenue base survives the stated period after churn and downgrades, capped at one hundred per cent. As reported it is a single blended figure across all tenures rather than a curve; the true twin of a persistence curve is revenue retention by acquisition cohort against tenure. Net revenue retention adds expansion, existing customers paying more, which is the component pharma structurally lacks because revenue per patient is bounded by the approved regimen and duration on therapy. Both are misread the same way, by quoting a blended average where only the cohort curve tells the truth.

What is the most common shared mistake? The window trap: quoting a rate measured inside a short window as if it were a lifetime truth. Persistence measured at a quarter and repeat purchase measured at twelve months on a multi-year replacement cycle are the same error.

Why does this matter for hiring a finance leader across sectors? Because the method is the transferable asset. A finance leader who has built the cascade in one industry knows which questions to ask first, and what remains to learn is the physics of the sector: who decides, who pays, what is regulated, and what a lost unit costs to win back. The skeleton is the part that does not need re-learning.

If your revenue plan will not decompose into units, rates and a defensible value per unit, that is usually the first conversation worth having. Those conversations usually sit in a multi-entity group, a post-acquisition integration, an FP&A function that has outgrown its reporting, or a business where cash and working capital have stopped following the plan. The first output is usually a reconciled unit-economics bridge, from source data to revenue, contribution and cash, with an owner against every rate. I take those conversations directly: get in touch.

Discuss this piece on LinkedIn: linkedin.com/in/jatinderpurewal · More Insights

References

Purewal, J. (2026), "The forecast you can argue with", jatinderpurewal.com/insights/patient-based-forecasting - the pharma cascade in full, including the adherence evidence and the censoring discussion this piece extends. A cross-reference to my own earlier work, not independent corroboration of it.
Skok, D., "SaaS Metrics 2.0", forEntrepreneurs - the standard treatment of subscription unit economics: churn, CAC, lifetime value, the LTV to CAC ratio and CAC payback.
McClure, D. (2007), "Startup Metrics for Pirates" - the origin of acquisition, activation, retention, referral and revenue as named funnel stages.
Bessemer Venture Partners, State of the Cloud annual reports - the annual benchmark series behind the good, better, best framing for net revenue retention. Read the sample before the number: published medians differ materially by segment, scale and year.
Fader, P.S. and Hardie, B.G.S. (2007), "How to Project Customer Retention", Journal of Interactive Marketing 21(1), 76-90 - the shifted-beta-geometric model - why cohort retention rates rise with tenure through heterogeneity rather than increasing loyalty.
Cramer, J.A. et al. (2008), "Medication Compliance and Persistence: Terminology and Definitions", Value in Health 11(1), 44-47 - the ISPOR definitions, which treat compliance and adherence as synonyms and record that no overarching term then combined compliance and persistence.
Vrijens, B. et al. (2012), "A new taxonomy for describing and defining adherence to medications", British Journal of Clinical Pharmacology 73(5), 691-705 - the ABC taxonomy used here: adherence as the umbrella process, in the phases of initiation, implementation and discontinuation, with persistence defined as the length of time from initiation to the last dose before discontinuation. Adopted as the conceptual basis of the ESPACOMP Medication Adherence Reporting Guideline (EMERGE, Annals of Internal Medicine, 2018).
Muntner, P. et al. (2018), "Potential US Population Impact of the 2017 ACC/AHA High Blood Pressure Guideline", Circulation 137(2) - the source for the guideline reclassification example.

The views here are my own. They do not represent the position of any current, former or future employer or client, and nothing here draws on confidential information.

Jatinder Purewal is an interim Finance Director, ACMA, CGMA, Cranfield Executive MBA. In commercial finance since 1999: sixteen years in listed and multinational businesses including IMS Health, O2, GSK, Parexel, Novartis/Chiron and Shionogi, and eleven as founder-owner of an international direct-to-consumer business. More at jatinderpurewal.com/about.