Can exponential growth be slow?

Yes; the proportional rate can be small.

Is exponential decay the same formula?

A decay model uses a factor between zero and one, or a negative continuous rate.

Can a log-linear fitted value be exponentiated to obtain the mean forecast automatically?

Not generally; stochastic retransformation can require an adjustment.

AI Glossary / Time series and signal processing / Intermediate

Exponential growth

Definition

Exponential growth is growth in which a quantity increases by a constant proportion over equal time intervals. As the quantity becomes larger, the absolute increase also becomes larger. A discrete model multiplies by a fixed factor each period, while a continuous model has a rate of change proportional to the current quantity.

What it means

Adding 50 users each month is linear growth. Increasing the current user count by 10% each month is exponential growth: the increase is 50 when there are 500 users, but 100 when there are 1,000.
The word exponential describes a mathematical pattern, not simply rapid or impressive growth. A small positive proportional rate is still exponential, and a very large one-time increase is not enough to establish that pattern.

How it works

For discrete periods, write yt = y0(1+g) to the power t, with positive starting value y0 and proportional growth rate g greater than zero. The factor 1+g applies repeatedly to the current amount, not repeatedly to the original amount.
For continuous time, write y(t) = y0 exp(kt), with k positive. Its derivative is k times y(t). The continuous rate k and a discrete per-period rate g are linked by k = log(1+g) when the time units match.
Specify the interval carefully. A monthly rate is not an annual rate. Twelve months of compounding at rate g produce an annual factor (1+g) to the power 12, which is generally different from 1+12g.
A constant exponential rate implies a constant doubling time. In the continuous form it is log(2)/k; in the discrete-period interpolation it is log(2)/log(1+g). If observations occur only at whole periods, distinguish the interpolated crossing time from the first observed period exceeding double.
Taking logarithms converts a positive deterministic exponential curve into a straight line: log y(t) = log y0 + kt. In data analysis, an approximately linear log plot may suggest exponential behavior, but noise, changing rates, and selected time windows can produce misleading appearances.
Distinguish the deterministic curve from a stochastic forecasting model. Random multiplicative growth means future values have a distribution. The average log growth, mean growth factor, median future level, and mean future level need not coincide.
Check the scope of extrapolation. Capacity limits, market saturation, resource constraints, policy changes, and competition can make a constant proportional rate implausible over long horizons. An exponential model can be useful locally without being a credible description forever.
Evaluate alternatives and uncertainty using time-respecting data. A rate estimated over a short period can be unstable, and small differences in the rate compound into large forecast differences. Report the starting value, rate, units, horizon, and assumptions rather than showing only a dramatic curve.

Three practical examples

A capacity planner explores a scenario in which service requests grow by a fixed percentage each month.
A storage team estimates when a dataset would double if its recent proportional expansion continued.
An analyst compares a straight-line trend with a log-linear growth model, checking later forecast performance and capacity constraints.

Think of it this way

Think of adding branches to a tree in proportion to the branches already present. Each round creates a larger base for the next round. The illustration captures compounding, but a real tree also faces resource limits, just as a business growth process does.

Worked example

Consider an original planning scenario beginning with 500 daily requests at month zero. Assume the typical daily request level grows by exactly 10% each month, with no noise or capacity limit in the calculation.
After one month, the level is 500 x 1.10 = 550. After two months it is 605, and after three it is 500 x 1.10 cubed = 665.5. The monthly additions are 50, 55, and 60.5, so the percentage is constant while the absolute increase grows.
After six months, the level is 500 x 1.10 to the sixth power = 885.7805. Simply adding six times the initial increase would give 800, which is the different assumption of linear growth.
The interpolated doubling time is log(2)/log(1.10), approximately 7.27254 months. At whole monthly observation points, the level is still below 1,000 after month seven and above 1,000 after month eight. These fractional request levels describe a modeled average rate, not a claim that an individual request can be fractional.

Illustrative case study

Fictional case: Meadow Cloud sees strong early growth in an AI service and fits an exponential curve to its first few months. A planning review identifies onboarding bottlenecks and a limited pool of prospective customers.
The team keeps the constant-rate curve as one capacity scenario but compares it with slowing-growth alternatives. It updates the assumptions as new observations arrive and avoids presenting the initial rate as a guaranteed long-run trajectory.

Common misconceptions and mistakes

Exponential does not mean any upward curve. A constant absolute increase is linear, while a constant proportional increase is exponential. A 10% rise followed by a 10% fall does not return to the starting value, because the bases differ. Continuous and discrete growth rates are not numerically identical except approximately at small rates. A good fit over a selected short window does not establish indefinite continuation.

Business use

Exponential growth models can support capacity scenarios, resource planning, and communication about compounding. They are particularly useful for showing when an assumed rate would exceed a limit. Pair the calculation with alternative rates and constraints, and separate a conditional scenario from a statistically validated forecast.

Check your understanding

Why is the month-three increase larger than the month-one increase?

Answer: The same percentage acts on a larger base.

What is the six-month level in the example?

Answer: 885.7805.

Why does the first whole-month doubling occur at month eight?

Answer: The interpolated crossing lies between months seven and eight.

Related terms

Logarithmic transformation
Multiplicative model
Compound growth
growth rate
doubling time
linear trend
logistic growth
extrapolation

Learn next

Compare equal absolute increments with equal percentage increments. Then vary a scenario's growth rate and horizon to see how sensitive long-run projections become.

Sources

Hyndman-hosted Glossary of Forecasting Terms, Exponential growth: https://robjhyndman.com/mwh3/FG4.pdf OpenStax, Exponential Growth and Decay: https://openstax.org/books/calculus-volume-2/pages/2-8-exponential-growth-and-decay OpenStax, Exponential and Logarithmic Models: https://openstax.org/books/precalculus-2e/pages/4-7-exponential-and-logarithmic-models

Previous: Exponential-Golomb coding Next: Exponential-logarithmic distribution