The more times you do something, the better (and cheaper) you do it. This intuitive principle has a precise and surprisingly regular mathematical form: the learning curve. In production, every time the cumulative volume of units produced doubles, the unit cost falls by a constant percentage — it is the famous «Wright's law», first observed in the aircraft industry in the 1930s and since then confirmed in dozens of sectors, from semiconductors to solar panels to batteries. This calculator turns that principle into concrete numbers: given the cost of the first unit and the learning rate (by how much costs fall at each doubling), it calculates how much the next units you produce will cost, their average cost, the cost of the last unit and the overall reduction. It is the tool you need to make realistic quotes, set competitive prices and plan production, avoiding the classic mistake of estimating future costs on the (high) costs of the first units. It is parametric, currency-neutral and universal: the learning curve holds in any sector and country. This is an informative estimate, not management advice.

What is the learning curve

The learning curve (or experience curve) describes an empirical fact observed in countless productive activities: as you accumulate experience producing units, the cost per unit decreases in a predictable way. It is not a vague impression, but a mathematical regularity discovered in 1936 by Theodore Wright studying aircraft production: he noticed that every time cumulative production doubled, the labor cost per aircraft fell by about 20%. Since then, the phenomenon has been documented in a huge variety of contexts, so much so that it has become a standard tool of cost estimation and planning. The underlying idea is that experience produces improvement on many fronts at once: workers become faster and make fewer mistakes, processes are optimized, waste is reduced, suppliers improve, the product is redesigned to be simpler to make. All these effects, summed, generate a fall in costs that follows a remarkably constant power law. The learning curve is therefore much more than a curiosity: it is a nearly universal law of production cost, which allows one to predict future costs and to make strategic decisions on prices, volumes and investments. This is an informative estimate, not management advice.

The learning rate: the heart of the curve

The parameter that governs the entire curve is the learning rate, expressed as a percentage. Its meaning is precise and elegant: it is the percentage to which the unit cost falls at each doubling of the cumulative volume. A learning rate of 80% — the most cited value — means that, every time the units produced double, the cost per unit becomes 80% of the previous one, that is it falls by 20%. If the first unit costs 1,000, the average of the first 2 costs 800, the average of the first 4 costs 640, of the first 8 costs 512, and so on: an implacable geometric descent. Beware of a counterintuitive point: a lower learning rate indicates faster learning. A 70% (costs falling to 70%, that is −30% at each doubling) is a steeper and more advantageous curve than a 90% (−10% at each doubling). At the limit, a learning rate of 100% means no learning: costs do not fall at all. The typical values observed in reality range from 70% (very rapid learning, typical of complex manual processes) to 95% (slow learning, typical of highly automated processes). Choosing the right learning rate for your process is the key to using the curve well: it can be estimated from historical data (observing how costs fell in the past) or by adopting the sector's reference value. This is an informative estimate, not management advice.

The formula: the power law

The learning curve is described by a power law, one of the most elegant and recurring mathematical forms in nature and economics. The average cost of the first N units is:

$$bar{Y}_N = a cdot N^{,b}, quad b = log_2(LR) = dfrac{ln(LR)}{ln 2}, quad C_{tot}(N) = a cdot N^{,b+1}$$

where a is the cost of the first unit, N the cumulative volume, LR the learning rate (as a fraction, e.g. 0.80) and b the curve exponent, always negative or zero. The heart of it all is the exponent b, defined as the base-2 logarithm of the learning rate. Why base 2? Because the fundamental property of the curve concerns doublings (the 2): the exponent translates «by how much it falls at each doubling» (the learning rate) into «how it falls at each single unit» (the power law). For a learning rate of 80%, b = log₂(0.80) ≈ −0.322. The total cost of the first N units is simply the average cost multiplied by N, that is a·N^(b+1). From these formulas everything is derived: the cost of a single unit, the cost of a batch, the average cost, the reduction. The calculator applies them for you, also allowing you to calculate the incremental cost of the next units starting from a number already produced — the typical situation of those who must make a quote with production already underway. This is an informative estimate, not management advice.

How it is calculated: the breakdown

Let us see the numbers in the default case — cost of the first unit 1,000, learning rate 80%, no units already produced, 100 units to produce. The table shows the breakdown:

Follow the numbers, because they reveal the power of the effect. The first unit costs 1,000, but the cost falls rapidly with experience: the 100th unit costs just 154, that is 84.6% less than the first. The total cost of the 100 units is 22,706, with an average cost of 227 per unit — much lower than the 1,000 of the first unit, but higher than the marginal cost of the 100th (154), because the average also includes the first expensive units. Here is the crucial point: if you had estimated the cost of the 100 units by multiplying the cost of the first unit by 100 (1,000 × 100 = 100,000), you would have overestimated the real cost by more than four times. It is the most common and costly mistake in making quotes, and the learning curve serves precisely to avoid it. Change the values with those of your case — initial cost, learning rate of your sector, units to produce — and get a realistic estimate. This is an informative estimate, not management advice.

Wright's law: every doubling reduces the cost

Here is the conceptual heart of the learning curve, the reveal that makes it so powerful and memorable. The fundamental property is this: every time the cumulative volume doubles, the average unit cost is multiplied by the learning rate. The table shows it with the default case (first unit 1,000, learning rate 80%):

Look at the regularity: at each doubling (1→2, 2→4, 4→8, 8→16…) the average cost is multiplied exactly by 0.80. It does not matter where you are on the curve: the percentage fall at each doubling is always the same. It is this constancy that makes Wright's law so useful and predictable. And it has enormous economic implications: it explains why the cost of solar panels has collapsed by about 80% every time the cumulative installed capacity has doubled (the «Swanson's law»), why lithium-ion batteries follow a similar trajectory, why semiconductors become continuously cheaper. It is not magic nor just technology: it is Wright's law at work on a global scale. Every doubling of the cumulative world volume cuts costs by a constant fraction — and this is why sustaining the growth of volumes (of a product, of a technology) is the main engine of cost reduction. This is an informative estimate, not management advice.

How much the learning rate matters

The learning rate is not a detail: small differences in its value produce enormous differences in costs at high volume. The table compares the average cost of the 100th unit (always starting from a first unit of 1,000) at different learning rates:

The table is illuminating. At 100 units produced, a process with a learning rate of 70% has an average cost of just 94 per unit (−91% from the first), while one at 95% is still at 711 (−29%). Seven and a half times the difference, all else being equal except the learning rate! This shows two things. First: correctly estimating the learning rate is crucial — using an 80% when your process is really a 90% would make you seriously underestimate future costs and set prices at a loss. Second: the learning rate is itself a strategic lever. It is not a fixed and immutable given: investing in training, team stability, process standardization and continuous improvement can «improve» the learning rate (make it lower), accelerating the descent of costs. Companies that learn faster than competitors build a lasting cost advantage. Knowing your own learning rate — and working to improve it — is a first-rate management skill. This is an informative estimate, not management advice.

Why costs fall: the sources of learning

Where does the fall in costs described by the curve concretely come from? Understanding it helps to exploit it and not take it for granted. The sources of learning are multiple and act together. The most intuitive is labor learning: by repeating an operation, workers become faster, more precise, waste less material and make fewer mistakes — it is the historical core of the effect. But there is much more. There is process improvement: with experience, better ways of organizing work are discovered, bottlenecks are eliminated, flow is optimized. There is product redesign: often, after producing the first units, one understands how to simplify the design to make it easier and cheaper to manufacture (design for manufacturing). There is the improvement of tools and equipment, the automation of previously manual steps. There is supplier learning, who in turn move down their curves and lower the prices of inputs. And there are the economies of scale that intertwine with learning (although they are conceptually distinct: scale concerns size, learning concerns cumulative experience). It is important to understand that this fall is not automatic: it requires attention, measurement and deliberate improvement effort. Organizations that actively «manage» their learning curve move down faster than those that wait passively. This is an informative estimate, not management advice.

Quotes and pricing: the first-unit mistake

The most immediate and valuable application of the learning curve is in quotes and pricing, where ignoring it causes costly errors. The classic mistake — let us call it «the first-unit mistake» — is to estimate the cost of a large production by multiplying the cost of the first units (high, because you are at the start of the curve) by the total number. As we have seen, this can overestimate the real cost many times over, leading to off-market quotes: you propose a price that is too high and lose the order, or you refuse it because «it is not worth it» when in reality, moving down the curve, it would have been profitable. The opposite mistake is equally dangerous: if you apply a too optimistic learning rate (or assume a learning that your process does not really have), you underestimate costs and end up working at a loss. The learning curve, used well, allows a powerful strategy: forward pricing. Knowing that costs will fall with volume, you can set from the start a price based on the average cost expected over the entire production (or even on the future cost), «betting» on learning to win volume. It is exactly what companies that enter a market with aggressive prices do: they are not selling at a loss forever, they are anticipating the descent down the curve. This calculator gives you precisely the average cost and the cost of the next units, the numbers you need for informed pricing. This is an informative estimate, not management advice.

The learning curve around the world

Wright's law is not an abstract theory: it is one of the most solidly documented phenomena of industrial economics, observed everywhere in the world and in radically different sectors. In the aircraft industry, where it all started, learning rates typically hover around 80-85%, and the curve is a standard tool to quote multi-year production programs. In electronics and semiconductors, learning (together with technological progress) has produced the continuous collapse of costs we know, with curves around 85-90%. The most spectacular case is that of solar photovoltaic energy: the cost per watt has fallen by about 80% at each doubling of the cumulative installed capacity — a phenomenon so regular as to have a name, the «Swanson's law», which has made solar the cheapest energy source in history. Lithium-ion batteries follow a similar trajectory, with enormous implications for electric vehicles and energy storage. The curve emerges also in surprising areas: medicine (surgeons improve outcomes with the volume of operations), services, software. Wherever there is repetition and cumulative experience, the learning curve is at work. This universality is what makes the concept so valuable: whatever you produce, in whatever country, if you do it repeatedly you can expect — and plan for — a predictable fall in costs. This is an informative estimate, not an opinion on specific sectors.

Cumulative average model versus unit model

A useful technical aspect to know is that there are two main formulations of the learning curve, which sometimes generate confusion. This calculator uses the cumulative average model, the original one by Wright: the formula Y̅ = a·N^b describes the average cost of the first N units, and the doubling property applies to this average. It is the most widespread model in cost estimation and the most intuitive to reason about batches and productions. There is however also the unit model (or Crawford model), in which the same formula a·N^b describes the cost of the single Nth unit, not the average. The two models give slightly different results and, at the same learning rate, trace distinct curves; it is therefore important to know which one is being used when comparing data or reading studies. In practice, the difference is often small for high volumes and the choice depends on the sector convention and on how the parameters were estimated. This calculator, besides the average cost, also gives you the cost of the last unit (the marginal cost of the Nth unit in the cumulative average model) and the incremental cost of a batch, so you have both the average and the marginal view. The important thing is to be consistent: use the same model to estimate the parameters and to make the forecasts. This is an informative estimate, not management advice.

Assumptions and limits of the model

The learning curve is powerful but, like every model, rests on assumptions that it is important to know. The first is that the learning rate is constant for the entire production: in reality, learning often slows down as the process matures (the curves can «flatten» at very high volumes, when there is little else to improve). The second is the continuity of production: the curves work best when production is continuous; long interruptions can cause «forgetting» (learning is partly lost and costs rise again when restarting). The third is the stability of the product and the process: important changes to the design, the technology or the team can reset or alter the curve. The fourth is that the model captures only the costs subject to learning (typically labor and some variable costs), while the costs of materials or inputs at market price may not fall in the same way. Moreover, the learning rate must be estimated, ideally from the historical data of your specific process, and a wrong estimate propagates to all the forecasts. Finally, the curve describes an average tendency: individual units can vary. So use the learning curve as a tool of estimation and planning — excellent to understand orders of magnitude, make realistic quotes and take strategic decisions — not as an exact forecast to the cent. This is an informative estimate, not management advice.

How to use the learning curve

To summarize, here is how to best exploit the learning curve in real decisions. First: estimate the learning rate of your process, preferably from historical data (observe how costs fell as past volumes doubled) or by adopting the reference value of your sector. It is the most important parameter: give it attention. Second: use the curve for quotes, calculating the cost of the units to produce instead of multiplying the initial cost — you will avoid grossly overestimating and losing orders. Third: exploit the curve in pricing, setting prices based on the average cost expected over the entire production, not on the high initial costs; if your learning rate is steep, you can afford aggressive prices to win volume. Fourth: plan investments knowing that costs will fall: a technology expensive today can become competitive after a few doublings of volume. Fifth: actively manage learning — the curve is not a passive destiny, but something you can accelerate with training, standardization, continuous improvement and production continuity. Sixth: do not overestimate the curve for mature or highly automated processes, where learning is slow. The learning curve is one of the most powerful conceptual tools of operations management and cost strategy: using it well distinguishes those who plan on the real costs of tomorrow from those who remain anchored to the high costs of today. This is an informative estimate, not management advice.

The scale effect: average cost at different volumes

The learning curve shows all its power when looking at growing volumes. The table shows how the average unit cost falls (starting from a first unit of 1,000, learning rate 80%) as the cumulative volume increases:

The descent is impressive and continuous: from an average cost of 227 at 100 units it reaches 52 at 10,000 units, a reduction of over three quarters. But note an important characteristic of the power law: the descent slows down in absolute value as the volume grows, while remaining constant in terms of doublings. From 100 to 1,000 units (a 10-fold increase) the average cost more than halves; from 1,000 to 10,000 (again 10-fold) it halves again, but starting from a lower base. This has a crucial strategic implication: the greatest savings in absolute value are obtained at the start of the curve, when each doubling starts from high costs. For a new product, the first doublings of volume dramatically cut costs; for a mature product at very high volume, enormous doublings (and thus a lot of time or market) are needed to obtain further significant reductions. It is the reason why emerging technologies see spectacular price collapses in the first years, which then attenuate with maturity. Planning knowing where you are on the curve is essential. This is an informative estimate, not management advice.

The learning rate around the world: typical values by sector

Although the learning rate must always be estimated on one's own data, knowing the typical values observed in various sectors helps as a starting point and for a plausibility check. The table summarizes some indicative values documented in the literature:

The values (indicative and variable from case to case) reveal a useful general pattern. Labor-intensive and complex processes, like the manual assembly of complex products, tend to have lower learning rates (steeper curves), because there is much room for human learning. Highly automated processes tend to have higher learning rates (flatter curves), because the machine «does not learn» and improvement depends on discrete innovations more than on accumulated experience. Cases like solar (with its remarkably constant 80% for decades) and batteries combine process learning, technological innovation and economies of scale, producing some of the most spectacular cost-reduction trajectories in economic history. Use these values as an initial reference, but replace them as soon as you have real data on your process: the learning rate of your specific case may differ, and basing decisions on the right value makes an enormous difference. This is an informative estimate, not management advice.

Learning curve versus economies of scale

Two often-confused but distinct concepts explain why producing more costs less per unit: the learning curve and the economies of scale. It is important to distinguish them because they have different causes and implications. The economies of scale concern the size of production at a given moment: by producing more simultaneously, fixed costs are spread over more units, volume purchase discounts are obtained, more efficient plants are used. They are a «static» effect linked to the rate of production (how many units per year). The learning curve, instead, concerns the cumulative experience over time: costs fall because you have produced many units in total, regardless of speed. It is a «dynamic» effect linked to the cumulative volume (how many units in total, ever). The practical difference is profound: economies of scale are obtained (and lost) by changing the size of plants; learning, once acquired, tends to remain (although «forgetting» exists). In reality the two effects often coexist and reinforce each other: a company that grows increases both the scale and the cumulative experience. Separating them conceptually helps however to understand the real sources of one's cost advantage and not to attribute to learning savings that come from scale, or vice versa. This calculator models the effect of learning (cumulative volume); for economies of scale different analyses are needed. This is an informative estimate, not management advice.

How to estimate the learning rate from your data

The most valuable input you can give this calculator is a learning rate estimated on your real data, not a generic value. The method is simpler than it seems. The basic idea: find two points of your past production where the cumulative volume doubled, and look at how much the average unit cost fell. If, for example, the average cost was 100 when you had produced 500 units, and fell to 85 when you reached 1,000 units (double), your learning rate is 85% (85/100). Repeat the calculation over several available doublings and take an average: you will get a robust estimate. For a more rigorous analysis, one can use regression: putting the data on a logarithmic scale (log of the cost against log of the cumulative volume), the learning curve becomes a straight line, and its slope directly gives the exponent b, from which the learning rate is derived (LR = 2^b). This linearization is the standard way learning curves are estimated from historical data. Some practical tips: use real cost data (adjusted for inflation, to compare homogeneous values); make sure the product and process remained substantially stable in the period; and collect enough points to cover at least a few doublings. A well-estimated learning rate transforms this calculator from a theoretical exercise into a powerful forecasting tool calibrated on your reality. This is an informative estimate, not management advice.

Common mistakes with the learning curve

The learning curve is powerful, but some recurring mistakes compromise its use. The first, the most serious, is the first-unit mistake already seen: estimating costs by multiplying the initial cost by the quantity, ignoring the descent. The second is the opposite: applying a too optimistic learning rate, assuming a learning that the process does not have, and underestimating costs to the point of working at a loss. The third is to use a generic value instead of estimating one's own: the learning rate of your process can differ a lot from the sector average. The fourth is to ignore forgetting: after long production interruptions, part of the learning is lost and costs rise again; assuming that learning is permanent is risky. The fifth is to extrapolate to infinity: real curves tend to flatten at very high volumes (learning runs out), so projecting the same descent forever leads to unrealistic estimates. The sixth is to confuse learning and scale, attributing to the curve savings that actually come from economies of scale (or vice versa), and thus planning badly. The seventh is to apply the curve to costs that do not learn: materials or inputs at market price often do not follow the curve like labor. The defense against all these mistakes is an informed use: estimate the learning rate from one's own data, distinguish the cost components, remember the limits of the model, and use the curve as a guide to estimation, not as an oracle. This is an informative estimate, not management advice.

Frequently asked questions

What is the learning curve?

It is the mathematical relationship by which the unit production cost decreases predictably with experience: every time the cumulative volume doubles, the average cost falls by a constant percentage (the learning rate). Discovered by Wright in 1936 in the aircraft industry, it is today a standard tool of cost estimation.

What is the learning rate?

It is the percentage to which the unit cost falls at each doubling of the cumulative volume. A learning rate of 80% means that at each doubling the cost becomes 80% of the previous one (that is it falls by 20%). Beware: a lower learning rate indicates faster learning (70% is steeper than 90%).

Why is a lower learning rate better?

Because it indicates a faster fall in costs. With a learning rate of 70%, at each doubling the cost falls by 30%; with 90%, only by 10%. So 70% is a steeper and more advantageous curve. At the limit, 100% means no learning (costs do not fall).

What is Wright's law?

It is the principle by which the unit cost decreases by a constant percentage at each doubling of the cumulative production volume. It is named after Theodore Wright, who observed it in 1936. It explains, among other things, the historical collapse of the costs of solar panels, batteries and semiconductors.

What is the formula of the learning curve?

The average cost of the first N units is Y = a·N^b, where a is the cost of the first unit, N the cumulative volume and b = log₂(learning rate). The total cost is a·N^(b+1). The exponent b is always negative or zero, and the more negative it is the steeper the curve.

Why should I not multiply the cost of the first unit by the quantity?

Because it is the most common and costly mistake: it enormously overestimates the real cost. The first unit is the most expensive (you are at the start of the curve); the subsequent ones cost ever less. In the default, multiplying 1,000 × 100 would give 100,000, while the real cost is only 22,706 — over four times less.

How do I estimate the learning rate of my process?

Ideally from your historical data: observe by how much the average cost fell when the volume doubled in the past. If the cost went from 100 to 80 at doubling, your learning rate is 80%. Alternatively, adopt the typical reference value of your sector (e.g. 80-85% for manual assembly).

Does the curve apply only to manufacturing production?

No, it is universal: it emerges wherever there is repetition and cumulative experience. It holds for aircraft, electronics, solar panels, batteries, but also for services, software, and even medicine (surgeons improve with the volume of operations). It is one of the most documented phenomena of industrial economics.

What is the difference between the average cost model and the unit model?

In the average cost model (the one of this calculator, Wright's original), a·N^b is the AVERAGE cost of the first N units. In the unit model (Crawford), the same formula is the cost of the SINGLE Nth unit. They give slightly different results; it is important to know which one is used when comparing data.

Can I calculate the cost of a batch when I have already produced some units?

Yes. Enter the units already produced and the units to produce: the calculator computes the incremental cost of the next units (the difference between the cumulative total cost at the end and at the start of the batch). It is useful for quotes with production already underway, where the next units cost less because you have already moved down the curve.

Is the fall in costs automatic?

No. The curve describes a potential, but the fall requires deliberate effort: training, process improvement, redesign, production continuity. Organizations that actively manage their curve move down faster. Long interruptions can cause «forgetting» and make costs rise again.

What are the limits of the model?

It assumes a constant learning rate (in reality learning slows with maturity), continuous production, stable product and process. It captures mainly labor costs; materials may not fall in the same way. The learning rate must be estimated, and a wrong estimate propagates to the forecasts. It is a tool of estimation, not an exact forecast.

Is this calculator management advice?

No. It is an informative tool of estimation and planning based on a standard model (Wright's law). The quality of the result depends on your estimates, in particular of the learning rate. For important decisions on costs, prices and investments, integrate the result with specific data and the opinion of experts.

Updated July 2026