Two of the most consequential money decisions in a software company are made with a toggle. One sits on your pricing page and says "Billed annually, save 17%." The other sits in the AWS console and asks whether a Savings Plan should be paid No Upfront, Partial Upfront or All Upfront. Neither looks like borrowing or lending, yet both are, and a sharp essay on cash, timing, and the stories your numbers don't tell explains why that deserves your attention: it counts financing mismatches among the quiet sources of business fragility and argues that optionality beats prediction. Engineers increasingly own both toggles, so this post does something finance teams rarely do for us. It puts an interest rate on each one, with code you can run.
Every early-payment discount is an interest rate in disguise
Finance people have run this calculation for decades. The old invoice term "2/10, net 30" gives a buyer 2% off for paying within 10 days instead of 30. Two percent for twenty days of patience sounds like pocket change until you annualize it: roughly 37% a year before compounding and about 45% after. A supplier offering those terms is paying a fortune for faster cash, and a buyer with idle money who ignores the discount is walking past one of the best risk-free returns in business.
Your pricing page runs exactly the same trade, with you as the supplier. "Two months free," which is what a "save 17%" toggle really means, has a customer hand you ten months of revenue on day one instead of paying monthly. The first month would have arrived anyway, so what you actually receive is nine months of cash early, and you repay it by not collecting the next eleven monthly payments. Solve for the rate that makes those two streams equal and you get about 3.5% a month. Compounded, that is 51% a year, a price most credit card issuers would be embarrassed to print.
A tiny calculator for any "pay now, save later" deal
You don't need a spreadsheet template to check numbers like these. You need one function that finds the rate hidden inside a stream of payments, which finance textbooks call the internal rate of return. This version has no dependencies and uses plain bisection, so it runs the same in a notebook, a Lambda or a CI job:
def monthly_rate(flows, lo=-0.5, hi=1.0): """Monthly rate at which the present value of `flows` is zero. flows[t] is the money that changes hands in month t (+ in, - out).""" def npv(r): return sum(f / (1 + r) ** t for t, f in enumerate(flows)) for _ in range(100): mid = (lo + hi) / 2 lo, hi = (lo, mid) if npv(lo) * npv(mid) <= 0 else (mid, hi) return mid def yearly(r): return (1 + r) ** 12 - 1 def annual_plan(months_charged=10, churn=0.0): """Seller's view: one annual payment vs. monthly billing that some customers quit.""" stay = [(1 - churn) ** t for t in range(12)] # share still paying in month t return yearly(monthly_rate([months_charged - stay[0]] + [-s for s in stay[1:]])) def all_upfront(upfront, monthly): """Buyer's view: one payment today instead of 12 monthly charges.""" return yearly(monthly_rate([-upfront] + [monthly] * 12)) for churn in (0, 0.01, 0.02, 0.03): print(f"Two months free, {churn:.0%} monthly churn: {annual_plan(churn=churn):.0%}") print(f"40% off paid upfront vs 36% off paid monthly: {all_upfront(7_200, 640):.0%}") Running it prints:
Two months free, 0% monthly churn: 51% Two months free, 1% monthly churn: 34% Two months free, 2% monthly churn: 19% Two months free, 3% monthly churn: 5% 40% off paid upfront vs 36% off paid monthly: 13% Every deal becomes a sequence of monthly cash flows, positive when money arrives and negative when it leaves, and the function searches for the rate at which their present value is zero. The two helpers only translate the way pricing pages and vendors describe their offers into those sequences.
Churn decides whether 51% is a rip-off or a bargain
The first line of output assumes something that's rarely true: that every monthly subscriber would have paid for all twelve months. Add churn and the price of the loan collapses. With 1% of monthly subscribers leaving each month, the implied rate falls to 34%. At 2% it's 19%. At 3% it's about 5%, because a monthly plan at that churn only collects around 10.2 months of revenue per starting customer anyway, barely more than the ten you got upfront. Push churn to 4% and the rate goes negative: the annual plan beats monthly billing before anyone even mentions interest.
So the discount is really two products in one box, a loan and a retention contract, and the retention half can be worth a lot. Research by Bain & Company's Frederick Reichheld, cited in Harvard Business Review's piece on the value of keeping the right customers, found that increasing customer retention by 5% raises profits by 25% to 95%. A customer committed for twelve months can't quietly cancel in month four, and that alone can justify a generous discount.
The catch is self-selection. The customers who choose annual billing are disproportionately the ones who had already decided to stay, so the churn rate that matters isn't your blended average. It's the churn of the people who actually take the offer. If your most loyal users rarely leave, the discount you hand them sits much closer to the 51% loan than to the 5% one, and you're paying a premium to borrow from customers who were never going anywhere. A useful experiment is to measure churn among monthly customers who look like your annual ones (same tier, same company size, similar tenure) and plug that number into the function instead of your average. If the answer looks expensive, try one month free instead of two, which works out to about 21% a year at zero churn, and save the bigger discount for segments where monthly churn is genuinely high.
The same math works when you're the customer. If your team is certain to keep a tool for the next twelve months, paying annually at two months free earns roughly 51% on that cash, which is about as good as finance gets. The only thing you're betting on is your own churn: the reorg, the migration or the cheaper competitor that makes you want out in month five.
The All Upfront button is a loan to AWS
Cloud commitments are the cleanest version of this trade, because the commitment doesn't change with the payment option. A Savings Plan locks in an hourly spend for one or three years no matter how you pay. With No Upfront, AWS charges that commitment monthly. With All Upfront, it charges everything at once in exchange for its lowest rates, and Partial Upfront sits in between with at least half paid on day one. Same obligation, different timing, which makes the choice a pure financing decision.
Say your quote shows 36% off on-demand prices with monthly payments and 40% off if you pay upfront. On $1,000 a month of on-demand usage, that's $640 a month versus a single $7,200 payment, and the last line of the script says prepaying earns about 13% a year. That's roughly triple what three-month Treasury bills were yielding at the time of writing, with no market risk attached. If the cash is truly idle, prepaying is a good trade, and the exact gap in your own quote (it varies by plan, term and region) is worth thirty seconds in the function.
Contractually, you owe the money either way. Practically, the two options diverge in exactly one scenario, and it's the one that matters most: the month cash runs short. With No Upfront, the money that would have gone to AWS is still in your account, buying payroll while you raise, cut or pivot. With All Upfront, it's already gone, and AWS's rules for returning a Savings Plan are deliberately narrow: the hourly commitment must be $100 or less, and the plan must have been bought within the last seven days and in the same calendar month. After that, the commitment runs its full term whether or not the company that signed it still needs the servers.
That is what a financing mismatch looks like at engineering scale. A three-year commitment signed by a company with fourteen months of runway is a mismatch however you pay for it. Paying upfront makes it worse, because it turns flexible cash into a discount that pays off only in the futures where the company survives and costs the most in the futures where cash runs out.
Cash is an option, and options have a price
Here's what the interest rate can't tell you. That 13% compares two ways of paying AWS. It says nothing about what the cash is worth to you. For a company with three years of runway and money parked in Treasury bills, the cash really is idle, and 13% is a gift. For a company with nine months of runway, the same cash is the difference between raising from strength and raising in a panic, and no discount makes that trade attractive. The rate is a fact. Whether it's a good deal depends on timing.
Runway flips the pricing page too. A startup that can't get a bank loan at any price may find that 51% from its happiest customers is the cheapest money actually on offer, with no covenants, no warrants and no new board seat attached. Paying a high rate deliberately, for a defined period, because you need the months it buys, is a legitimate strategy. Paying it by accident because nobody computed the number is not.
Four questions to ask before anyone clicks
None of this requires a treasury department. It requires a habit of asking the same few questions whenever a discount is attached to paying, or getting paid, sooner:
- What rate is hiding in the deal? A discount quoted as a share of the price means nothing until you convert it into a rate per year. Run the function.
- What would this cash do otherwise? Compare the rate with what idle money earns today, and with what replacing that money would cost if you suddenly needed it next quarter.
- Does the commitment outlive the runway? If the term is longer than the cash you have, you're making a bet on survival, and paying upfront raises the stakes of that bet.
- Who is on the other side, and what is their churn? For customers, use the churn of the segment that actually picks annual. For yourself as a buyer, be honest about how likely you are to want out before the term ends.
Lock-in has a cash edition
Engineers spend whole sprints avoiding vendor lock-in: abstraction layers, portable containers, an exit plan for every managed service. Then someone clicks All Upfront on a three-year plan and locks the company in financially, in a way no architecture review would ever approve. The answer isn't to stop prepaying or to drop annual plans, since both can be excellent deals. The answer is to read every "save X%" label as an interest rate, compute it, and decide on purpose. It takes one function and five minutes, and it may be the highest-return code your team ships this quarter.