Every long-term supply agreement I have read in the auto industry carries a version of the same clause: the supplier shall deliver annual productivity of two to three percent for the life of the program. The buyer files that clause as a win. The supplier files it as a number to pre-fund. Both can be right at the same time, and that is the part worth thinking about.

The question underneath the clause is simple to ask and surprisingly hard to answer. When a program saves two percent a year for six years, did the buyer actually take cost out of the part, or did the buyer just rearrange when the money changes hands? The honest answer is that it depends entirely on where the starting price came from. And most of the time, nobody in the room knows.

Where the two percent comes from

The annual cost-down is one of the oldest rituals in automotive procurement. A long-term agreement in this business typically runs the life of a vehicle program, six to twelve years, and the price is not meant to hold flat across that span. Suppliers are expected to give it back on a schedule. Over the past decade OEMs pushed for average annual reductions of roughly three percent, and the contractual language is explicit: in one Ford supply agreement disclosed in SEC filings, the supplier agreed to a 3.5 percent annual price reduction "consistent with the amount of annual productivity that Ford expects from its other major Tier 1 suppliers." Other agreements specify one and a half to two and a half percent. The number varies. The ritual does not.

The logic behind it is not cynical, at least not originally. It rests on the learning curve, which is one of the most durable empirical findings in manufacturing. In 1936 an aeronautical engineer named Theodore Wright noticed that every time cumulative aircraft production doubled, the labor required to build the next airframe fell by about fifteen to twenty percent. The pattern holds across industries, usually somewhere between ten and thirty percent per doubling of cumulative volume. Workers get faster. Yields improve. Tooling gets amortized. Engineers find a cheaper fastener or delete a bracket. This is real, and a buyer who walks away from it is leaving money on the table. The instinct to write down those gains and demand a share of them is sound.

Note the variable, though. The learning curve is driven by cumulative volume doubling, not by the calendar. That distinction is where the trouble starts.

The case for the target

Set the cynicism aside for a moment, because the productivity model has a genuine argument behind it. A target nobody has to hit is a target nobody works toward. The discipline of an annual number forces both sides to keep hunting for cost reduction long after the launch adrenaline wears off, and a lot of real money hides in that long tail: a value-engineering change in year three, a yield improvement once the line settles, a resin substitution that nobody had time to qualify before start of production.

Toyota built an entire cost culture on this premise and called it kaizen, the practice of continuous small improvements. The difference is that Toyota treats the give-back as something the two companies find together rather than something the OEM extracts. That distinction shows up in the data, and I will come back to it. The point for now is that a well-run cost-down program is not theater. It is a forcing function, and the savings can be entirely real.

A target is also a gift to finance. A CFO can put "three percent annual material productivity" in a model and defend it to the board. There is real organizational value in a number that is predictable, even if the predictability is somewhat manufactured. None of this should be dismissed.

The case against it

Here is the part everyone in the industry knows and few say out loud: a supplier who is told on day one that the price must fall two percent a year does not absorb that out of generosity. The supplier prices it in. If I have to surrender roughly ten percent of margin over a program and I have any pricing power at all, I quote the start-of-production price four points high and call the planned erosion a "cost-down." The buyer celebrates a saving that was, in fact, an overpayment being slowly refunded. This is the mechanism the cynical read describes, and it is common enough that experienced buyers assume it is happening and price their own expectations around it. The overpayment also lands at the worst possible time. Start of production and the volume ramp are when a program spends hardest on tooling, inventory, and launch containment, so the buyer ends up funding the illusion of a future saving at the moment it can least afford the waste.

A quoted price is not only a manufacturing cost, either. It is also a record of how the buyer behaves. A supplier that expects late drawing changes, volume forecasts that come in optimistic, slow payment, and the occasional demand to absorb an engineering change prices every bit of that into the opening number. The same supplier will hand two different OEMs very different prices for the same part, and the gap has nothing to do with the part. A hard annual cost-down sits on top of that risk premium, not in place of it.

The deeper problem is the one I flagged earlier. The learning curve runs on cumulative volume, and a fixed annual percentage does not. A part can climb from five thousand cumulative units to ten thousand in the first weeks after launch, then take years to go from one million to two million, so the doublings that drive real savings bunch up at the start. Early in a program, volume doubles quickly and genuine savings are abundant; demanding two percent then is almost too easy. Late in a program, when annual volume is flat and cumulative volume takes years to double, there is no learning-curve money left to find. The mandated two percent in year six cannot come from efficiency, because the efficiency has already been captured. It can only come from margin. So the back half of a long contract quietly converts a productivity target into a margin transfer, and calls it the same thing.

That transfer lands on companies that cannot absorb it. Bain's automotive profitability work put fourth-quarter 2025 supplier EBIT margins at about 6.9 percent and OEM margins at 3.6 percent. Neither number is fat, and suppliers have been absorbing the cost-reduction pressure of OEM performance programs on top of tariffs and soft volume. The result is visible in the bankruptcy docket. Marelli, a Tier 1 with tens of thousands of employees, filed for bankruptcy in June 2025 and blamed tariffs directly. First Brands followed into Chapter 11 in September 2025. AlixPartners, surveyed by Bloomberg, named automotive the global sector most vulnerable to financial distress. When you squeeze a seven-percent-margin supplier for another two points it cannot earn, the savings do not disappear into thin air; they reappear later as a quality escape, a line-down, or a renegotiation with a gun on the table. A healthy supply base is not charity. As we have written before, if your suppliers are unhealthy, your business is unhealthy too.

OEM versus Tier 1 supplier EBIT margins, Q4 2025: about 3.6% for OEMs and 6.9% for suppliers. Neither has much room left to give. Source: Bain & Company.

There is also a relationship cost, and it is measurable. For more than two decades Plante Moran and Planning Perspectives have run the Working Relations Index, scoring how North American OEMs are seen by their own suppliers. Toyota and Honda, the two that press least aggressively on price and involve suppliers earliest, sit at the top almost every year, while the companies that lean hardest on cost-downs sit at the bottom. Planning Perspectives founder John Henke estimated that GM alone could have captured hundreds of millions of dollars a year in additional value if its supplier relations matched Toyota's. Suppliers in the 2023 study reported feeling like a "true partner" to the top-three OEMs roughly twelve times as often as to the bottom three. The price you pay for a hard cost-down regime is not only the padding suppliers build in; it is the engineering idea they decide not to bring you. This is the same dynamic the 2026 Working Relations Index tracked when every OEM finally improved at once.

If you want the cautionary tale in human form, it has a name. In the early 1990s General Motors handed global purchasing to José Ignacio López de Arriortúa, whose PICOS program reopened signed contracts and demanded immediate price cuts, reportedly booking around a billion dollars in savings. For a year it looked like genius. Then it became the template for everything suppliers learned to distrust about Detroit, because once a supplier believes you will reopen a deal whenever you need to make a quarter, it stops bringing you its best cost ideas and starts pricing in the next surprise. López decamped to Volkswagen in 1993 in the middle of an industrial-espionage scandal, but the operating style outlived him by decades. The lesson it left behind is the one that matters here: there is a difference between managing cost and merely extracting it, and it is the difference we tried to stay on the right side of at Tesla.

What we did at Tesla: start at the right price

At Tesla we mostly did not negotiate cost-downs. We negotiated the starting price, and we spent the effort up front instead of stringing it across a decade.

The premise was that a supplier needs to make money, and that the right time to figure out the right price is before start of production, not after. So we did the work to understand what a part should actually cost. Sometimes that meant a full should-cost model, building the price from the bottom up out of raw material, machine time, labor, overhead, and a fair margin, the discipline McKinsey calls cleansheet target costing and the gold standard for serious hardware buying. Sometimes, when a commodity was well understood, it just meant benchmarking the same part across three or four suppliers until the real number revealed itself. Either way the goal was the same: arrive at a fair, true cost at start of production, pay the supplier a margin it could live on, and then largely leave the price alone.

The interesting consequence is that this approach often produces fewer cost-downs later, and that is the point, not a flaw. If you start at true cost, there is no artificial fat to unwind. You forgo the satisfying annual reduction because you never overpaid to create room for one. What you do not forgo is genuine learning-curve savings, and the cleanest way to capture those is to share them deliberately rather than mandate them blindly.

For parts dominated by raw material, we did something more honest than a fixed cost-down: we let the price float against an index. If a stamping is mostly steel and a connector is mostly copper, pretending the price should fall two percent a year while the London Metal Exchange does whatever it wants is a fiction that benefits no one. A commodity adjustment clause ties the price to a published index, adjusts it on a defined cadence, and usually shares the movement with a collar. This is standard practice, common enough that the Defense Department codifies it as the Economic Price Adjustment clause for steel, aluminum, and copper mill products. It removes the single largest source of pricing disputes by admitting up front that neither party controls the commodity. It is also the only sane way to write a contract in an era when a tariff announcement can move a material cost more in a week than a cost-down program moves it in a year. We made a related argument in our inflation white paper: when input costs are volatile, the contract has to flex with them or it breaks.

Here is the same logic I learned building cars at Tesla, laid out against the conventional model.

When each one is right

This is not a morality play with a hero and a villain. The fixed cost-down has a wrong version and a defensible one, and so does true-cost pricing.

The annual target works when the part is a mature, high-volume commodity riding a genuine learning curve, when the OEM frames the give-back as a shared find rather than a tax, and especially when the buyer lacks the cost-engineering muscle to know true cost in the first place. A blunt instrument beats no instrument. The true-cost approach works when you have the capability to build credible cost models, when parts are complex or newly engineered, and when input prices are volatile enough that a fixed schedule is obvious fiction. It demands more, and not every company can staff it, which is the honest case against it: a should-cost model anchored to the wrong number is worse than no model, because it gives false confidence in a negotiation.

The best programs I have seen refuse the binary. They decompose the part. Raw material floats on an index. The conversion cost, where the learning curve and value engineering actually live, carries a genuine improvement target, ideally with a gain-share so the supplier keeps part of what it finds. Margin is set fairly and held. Done this way, the "two percent" stops being a ritual and starts describing something true about the part, which is the only version of a cost-down worth defending. Getting there requires early supplier involvement rather than a clause imposed after the design is frozen.

This is not only an automotive habit

The fixed cost-down is most associated with car companies, but the underlying tension between mandated reductions and true-cost transparency runs through every industry with serious procurement.

Defense sits at the regulated extreme. The Truth in Negotiations Act, on the books since 1962, requires contractors on sole-source work above a threshold to hand the government certified cost or pricing data, and the Defense Contract Audit Agency can claw back the difference for years afterward if that data turns out to have been defective, with no need to prove intent. The threshold rises from 2.5 million to 10 million dollars for contracts entered after June 30, 2026, under the latest defense authorization, but the principle holds: in defense, true cost is not a negotiating style, it is the law. The whole apparatus exists because without a competitive market to set price, the only alternative to transparency is getting fleeced.

Consumer electronics shows the opposite case, where mandated cost-downs are mostly honest because component prices genuinely fall. Apple's position as the world's largest buyer of NAND flash drove the price of the memory in an iPhone from roughly 19 dollars to 10 over a single product cycle, a real decline a teardown can verify. That is what makes the electronics should-cost discipline work: the savings are sitting in the silicon. But even here the curve is not gravity. IHS teardowns pegged the iPhone 16 Pro Max bill of materials at about 485 dollars, 32 dollars higher than the prior generation. Costs fall when physics and volume cooperate, and not on a schedule a contract picks in advance.

Retail runs the buyer-power version, and 2025 made it vivid. Walmart, whose Every Day Low Cost program is the retail cousin of the automotive cost-down, asked its Chinese suppliers in February 2025 to cut prices by as much as ten percent for each round of new tariffs, trying to push the tariff cost back up the chain. Few suppliers agreed, because they were already running on margins too thin to give. That is the cost-down model meeting its limit in real time: you cannot extract a reduction from a supplier that has nothing left, no matter how much volume you control. The same logic governs the transition from automotive to aerospace procurement, where buyers raised on annual givebacks have to relearn cost discipline in a should-cost world.

Cost visibility is the common denominator

Notice that every approach in this piece, the honest ones and the cynical ones alike, depends on the same thing: knowing what the part actually costs. The buyer who wants to verify that a two-percent give-back is real and not just padding being refunded needs a cost model. The buyer who wants to price true cost at start of production needs a cost model. The buyer who wants to index steel and argue about nothing else needs to know how much of the price is steel. Cost visibility is not a strategy; it is the precondition for having one.

This is the unglamorous problem LightSource works on. The challenger manufacturers we build for are trying to launch faster than incumbents many times their size, and they do not have floors full of cost engineers or two years to overpay before a refund schedule kicks in. LightSource builds a cost model for the parts on the bill of materials, normalizes incoming quotes line by line so hidden margin has nowhere to sit, and tracks each price against raw-material indices and tariff exposure across the life of the program. Whether a team holds a supplier to an annual target or sets a fair price once and indexes the rest, the prerequisite is identical: you have to see the cost structure clearly enough to argue from facts instead of from bargaining power. The same conviction shaped what we learned moving from Tesla to Waymo to a software company: the team that sees the system most clearly makes the best decisions fastest.

The tariff era is going to settle this argument the hard way. A contract that assumes cost only ever falls was written for a world that ended somewhere around early 2025. The clause promising two percent a year is now colliding with steel that moves on a headline and a supplier base too thin to absorb the difference. The teams that come through it will not be the ones with the cleverest cost-down language. They will be the ones who can see what their products actually cost, and who started the relationship at a price both sides could live with.

Sources

• PwC -- Automotive Cost Reduction -- OEMs forced average annual supplier cost cuts of roughly 3% over the past decade.

• Visteon Form S-1, Ford-Visteon agreement (SEC EDGAR) -- the 3.5% annual reduction "consistent with the amount of annual productivity that Ford expects from its other major Tier 1 suppliers."

• Plante Moran -- 2025 Working Relations Index Study -- Toyota leads, Detroit and Nissan trail; partnership outscores price pressure.

• Plante Moran -- 2023 WRI insights -- suppliers feel like a "true partner" ~12x more with top-3 vs bottom-3 OEMs.

• WardsAuto -- Toyota rises in OEM-supplier relations -- Henke's estimate of the value GM left on the table.

• Automotive News -- Lopez: from cost-cut hero to defector -- GM's PICOS purchasing program and the archetype of adversarial supplier relations.

• Harvard Business Review -- Building Deep Supplier Relationships (Liker & Choi) -- the Toyota/Honda partnership model versus arm's-length cost pressure.

• McKinsey -- Cleansheet target costing -- building price from the bottom up to a should-cost figure.

• aPriori -- Your should-cost estimate is not accurate, and that's okay -- why imperfect cost models still win negotiations.

• Our World in Data -- Learning curves and Wright's Law -- ~10-30% cost decline per doubling of cumulative production.

• Bain -- Automotive profitability: how OEM and supplier margins are faring -- Q4 2025 supplier EBIT ~6.9%, OEM ~3.6%.

• Crain's Detroit -- Marelli bankruptcy a red flag for suppliers -- Tier 1 bankruptcy blamed on tariffs, June 2025.

• Modern Distribution Management -- First Brands bankruptcy -- the parts supplier entered Chapter 11 in September 2025.

• Bloomberg Law -- Automotive is the sector most vulnerable to distress -- AlixPartners survey, 2025.

• DFARS 252.216-7000 -- Economic Price Adjustment, steel/aluminum/copper -- the government's standard commodity indexation clause.

• U.S. Bureau of Labor Statistics -- PPI price adjustment guide -- how to index contract prices to published indices.

• Truth in Negotiations Act (overview) -- certified cost or pricing data requirements in defense contracting.

• Government Contracts Law -- FY2026 NDAA cost-data threshold change -- threshold rises from $2.5M to $10M after June 30, 2026.

• AppleInsider -- iPhone 16 costs more to make than iPhone 15 -- teardown BOM ~$485, up $32, showing cost-downs are not automatic.

• Investing.com / Bloomberg -- Walmart asks China suppliers for tariff price cuts -- up to 10% per tariff round, February 2025; few accepted.

Frequently Asked Questions

What is an annual cost-down in automotive supplier contracts?

An annual cost-down, also called annual productivity or a price-down, is a contractual commitment that a supplier's price will fall by a set percentage each year over the life of a vehicle program, commonly two to three percent. The logic comes from the manufacturing learning curve, where costs genuinely fall as cumulative volume grows. The risk is that suppliers anticipate the reductions and inflate the starting price to fund them, so the "savings" can amount to refunding an early overpayment.

What is should-cost or target costing, and how is it different?

Should-cost modeling, also called clean-sheet or target costing, builds a part's price from the bottom up out of raw material, machine time, labor, overhead, and a fair margin, to estimate what it ought to cost under efficient production. Instead of negotiating a percentage off a supplier's quoted number, the buyer negotiates toward an independently derived true cost. It requires more capability up front but produces a defensible starting price rather than a discount off an unknown base.

Why did Tesla avoid fixed annual cost-downs?

The approach was to invest in understanding true cost before start of production, through should-cost modeling and multi-supplier benchmarking, then pay a fair price with a healthy supplier margin and largely hold it. Starting at true cost leaves little artificial margin to "give back" later, so fewer cost-downs appear, by design. For raw-material-heavy parts, the price is tied to a commodity index rather than a fixed reduction schedule.

Do other industries use annual cost-down targets?

Yes. Retail runs an aggressive version, as when Walmart asked Chinese suppliers in 2025 to cut prices to offset tariffs. Consumer electronics relies on genuine cost-downs because component prices fall as volume scales. Defense takes the opposite path, legally requiring certified cost data under the Truth in Negotiations Act rather than relying on annual percentage targets. The pattern: mandated cost-downs dominate where buyer power is concentrated, and true-cost transparency dominates where it is regulated or where costs genuinely decline.

How do raw-material index clauses work?

A commodity adjustment or economic price adjustment clause ties a part's price to a published index, such as the London Metal Exchange price for aluminum or copper, and adjusts it on a defined cadence, often with a cap and floor to limit swings. A typical formula resets the price as base price times the ratio of the current index to the base index. It shares commodity risk explicitly so neither side has to guess where steel or resin will be in three years, which removes one of the largest sources of pricing disputes.

How does cost visibility help regardless of negotiation strategy?

Whether a buyer demands an annual reduction or prices true cost up front, both require knowing what the part actually costs. Cost visibility lets a buyer detect padding in a quoted price, verify that a promised give-back is real rather than a refund of early overpayment, and track each part against raw-material indices and tariff exposure over time. Platforms like LightSource build that cost model across the bill of materials and normalize supplier quotes so decisions rest on the cost structure rather than on bargaining power alone.