A backlog that grows by ten tickets a week is annoying. A backlog that grows by ten percent a week will bury you, and it will do it later than you expect and then all at once. The difference between those two sentences is the whole topic: how a number gets bigger, and how badly our intuition reads the curve.

The quick version

  • Linear growth adds a fixed amount each step (+5, +5, +5). Steady, predictable, easy to plan against.
  • Polynomial growth speeds up as the input grows (think area, or "n²" work that scales with the square of your headcount). Faster than linear, but still tame.
  • Exponential growth multiplies by a fixed rate each step (×1.1, ×1.1, ×1.1). It looks flat, then explodes, and people consistently underestimate it.
  • The leadership move: name which curve you're on early, because the three demand completely different decisions about capacity, cost and timing.

The idea in depth

Start with the cleanest test you can carry in your head: what stays constant as the thing grows?

If a fixed amount is added each period, the growth is linear: revenue up £5k a month, two new hires a quarter. Plot it and you get a straight line. If the rate of increase itself grows with the size of the input, but in proportion to a fixed power of it, the growth is polynomial. The textbook case is anything that scales like area (n²) or volume (n³): double the side of a square and its area quadruples. In organisational life, the classic n² trap is communication overhead, the number of possible pairwise conversations on a team of n people is n(n−1)/2, which is why a team of 12 feels disproportionately harder to coordinate than a team of 6, not twice as hard.

If a fixed multiplier is applied each period, ×1.1, then ×1.1 again on the new, bigger number, the growth is exponential. This is compounding, and it is the one humans cannot see coming.

flowchart LR
    Q(["How is it growing?"]) --> A("Same amount added
each step? (+5, +5)")
    Q --> B("Scales with a power
of the input? (n², n³)")
    Q --> C("Multiplied by a fixed
rate each step? (×1.1)")
    A --> AL(["Linear, plan in straight lines"])
    B --> BP(["Polynomial, watch coordination
and cost-to-scale"])
    C --> CE(["Exponential, act before
the curve turns up"])
One question sorts most growth into the right family. Leaders Loop

Why the gut gets exponential growth wrong

This is not a soft claim, it is one of the better-replicated findings in the psychology of judgement. In 1969, physicist Albert Bartlett opened his much-repeated lecture Arithmetic, Population and Energy with the line that "the greatest shortcoming of the human race is our inability to understand the exponential function." That was rhetoric; the experiments came later and backed him up. Wagenaar and Sagaria (1975), in Perception & Psychophysics, asked people to extrapolate an exponential series and found they grossly underestimated it, many produced guesses below 10% of the correct value, and mathematical training barely helped. The instinct is to mentally draw a straight line through the early points, because the early points genuinely look almost straight.

So the move is: never eyeball an exponential trend from its early data. The flat-looking start is a trap. If a metric is compounding, users referring users, costs scaling with a growing base, a fault rate that feeds on itself, get the doubling time, not the slope.

The Rule of 72, and why doubling time beats percentages

Percentages are where intuition fails; doubling time is where it recovers. The Rule of 72 is the shortcut: divide 72 by the percent growth rate and you get the rough number of periods to double. Growing 6% a year? It doubles in about twelve years. Growing 12%? About six. A "small" jump from 6% to 12% doesn't make things a bit faster, it halves the time to double. Make that reframe a habit: stop saying "x percent" and start saying "doubles every n." Doubling is something the human mind can actually hold; a percentage just slides past.

The same maths drives the heuristic-versus-first-principles tension worth naming here: the Rule of 72 is a heuristic, fast and good enough for a meeting; when the decision is large or irreversible, a multi-year cost projection, a capacity bet, drop back to the actual formula and a spreadsheet.

flowchart LR
    L(["Linear: +10/step"]) --> L2(["Step 1: 10"]) --> L3(["Step 5: 50"]) --> L4(["Step 10: 100"])
    E(["Exponential: ×2/step"]) --> E2(["Step 1: 2"]) --> E3(["Step 5: 32"]) --> E4(["Step 10: 1,024"])
Same start, ten steps later. Linear adds; exponential multiplies, and the gap is not close. (Illustrative figures.) Leaders Loop

The honest limitation: nothing grows exponentially forever

Here is where naïve "it's exponential!" thinking breaks. In the real world, exponential curves run into limits, market saturation, finite resources, capacity ceilings, and bend into an S-curve (a logistic curve): exponential at first, then slowing, then flattening at a plateau. It's the shape behind Everett Rogers' Diffusion of Innovations (1962): adoption starts slow, picks up as social proof compounds, then tails off once the easy-to-reach segments are gone.

So the move is: treat an exponential model as a description of a phase, not a destiny. The leadership questions become: where is the ceiling, and how close are we? Extrapolating an early exponential trend to infinity is exactly the over-correction that follows from finally taking exponentials seriously, and it's how forecasts get embarrassed when growth quietly turns the corner into its plateau.

The flat-looking start of an exponential curve is not calm. It is the part you were supposed to act on.

A worked example

(Figures below are illustrative.) A support team handles 400 tickets a month. Two trends land on the lead's desk. First, the product team's roadmap will add features that generate a predictable +30 tickets a month, linear. Second, a viral referral loop is growing the active user base by 8% a month, and tickets track users, exponential.

The linear trend is easy: in a year, that's +360 tickets, so plan hiring against a straight line. The exponential trend is the one the lead's gut will misread. Eyeballing it, 8% a month "feels" like maybe 8 × 12 ≈ a 96% rise, roughly double in a year. Wrong. Compounding at 8% monthly is about ×2.5 over twelve months, and the Rule of 72 says the user base (and the ticket load with it) doubles roughly every nine months (72 ÷ 8). Plan for "double in a year" and you are already under-resourced by spring of year two.

The better response isn't to panic-hire. It's to act on the curve while it's cheap to act: invest in deflection (self-serve docs, automation) now, during the deceptively flat early months, because a fix that removes 20% of tickets is worth far more applied before the base doubles than after. And ask the saturation question, referral loops are S-curves; this 8% will not last forever, so size the permanent team for the plateau, not the peak of the climb.

flowchart TB
    S(["Spot a compounding metric"]) --> D(["Estimate doubling time
(Rule of 72)"])
    D --> P(["Act in the flat early phase,
cheapest leverage"])
    P --> C(["Find the ceiling,
where does the S-curve flatten?"])
    C --> R(["Size for the plateau,
not the peak slope"])
A five-step drill for any metric that compounds. Leaders Loop

Frequently asked questions

How do I tell exponential from polynomial growth quickly?

Ask what stays constant. If each step adds a fixed amount, it's linear. If each step multiplies by a fixed rate, it's exponential. Polynomial sits between: the increase grows with the input but by a fixed power (n², n³), not by repeated multiplication. A practical tell: plot the data on a log scale, exponential growth becomes a straight line, polynomial growth does not.

Isn't "exponential" just a buzzword for "fast"?

In casual speech, yes, and that's the problem, it gets used for anything steep. Technically it means one specific thing: a constant multiplicative rate. The distinction matters because exponential and merely-fast-linear growth demand different decisions. If you're using the word loosely, say "fast-growing" instead and save "exponential" for when something is genuinely compounding.

What's the single most useful number to carry?

Doubling time, via the Rule of 72 (72 ÷ growth-rate-percent ≈ periods to double). Percentages slip through human intuition; "this doubles every nine months" sticks and forces the right planning horizon.

If exponential curves always become S-curves, why worry about the exponential part?

Because the damage, or the opportunity, happens during the exponential phase, before the plateau. Costs blow out, demand outruns capacity, or a competitor's adoption compounds past you, all while the curve still looks early and harmless. Knowing the S-curve ending is coming tells you how big to build; it doesn't excuse ignoring the climb.

Does this only matter for growth, or for decline too?

Both. Exponential decay is the same maths in reverse, churn that compounds, technical debt's interest, an audience that halves each cycle. The same intuition failure applies: a slow-looking early decline can be a fast collapse you're reading as a straight line.

Related in the Toolkit

Where to go next