In short: A retroactive volume tier makes the last order of the period worth far more than its invoice value, which produces a marginal price the commercial team never agreed to and a demand signal the supply team cannot use. Switching the same money to an incremental tier, paid only on volume above the threshold, removes the discontinuity without changing what the customer earns at plan. Rebates are variable consideration under IFRS 15 and ASC 606, so the accrual is a customer level forecast that lands in reported revenue every month. The test for whether a tier is buying anything is bunching: plot final period volume against the threshold and look for a hole just below it.
It is the second week of December and a distributor is sitting at 94 percent of the threshold that pays three percent back on everything they bought this year. Their planner has already worked out that the cheapest six weeks of stock they will ever buy are the ones that close that gap. The order lands on the eighteenth, the warehouse takes it, and January ships almost nothing.
Everyone involved understands what happened. The sales manager books a strong December, the customer books a strong rebate, the plant works overtime, and the demand planner watches a spike arrive that has no consumer behind it. The structure produced the behaviour, and the structure was designed by people who were thinking about the year rather than about the last fortnight of it.
What a rebate is doing that a price cut is not doing
A rebate is a conditional price. It pays after the fact, it conditions on something the customer controls, and it is the instrument you reach for when you want a lower effective price for some buyers without publishing a lower list price for all of them.
The conditions in common use divide into a few families. Volume tiers pay for scale. Growth rebates pay for change against a base period. Mix rebates pay for buying the part of the range you want moved. Compliance rebates pay for behaviour, which might be sharing sell-out data, honouring a range commitment, paying to terms, or ordering in full pallets.
The last family gets used least and defends itself most easily, because it buys something you can verify and something you actually want. A rebate that pays half a point for full pallet ordering and half a point for weekly stock file transmission is buying two operational improvements at a price you can compare with what they are worth.
The cliff, priced properly
The design flaw that causes most of the damage is retroactivity. A tier that pays on the whole year once a threshold is reached makes the marginal unit carry the entire tier.
Take a customer with a threshold of one million and a three percent retroactive rebate. They are tracking to 940,000.
If they stop there, you invoice 940,000 and pay nothing. If they buy the remaining 60,000, you invoice one million and pay 30,000 back, so you keep 970,000. The extra 60,000 of shipment brought in 30,000 of net revenue. The marginal price on that increment is fifty cents in the dollar.
Now suppose your gross margin is 25 percent, so those goods cost 45,000 to make. You sold them for 30,000 net. The order that closed the year at plan lost 15,000, and it will be reported as a success in every review it appears in, because the reporting looks at the customer's annual margin rather than at the increment.
The same money paid incrementally behaves completely differently. Pay three percent only on volume above one million and the customer at plan earns nothing at the line, the customer at 1.2 million earns 6,000, and the marginal unit is priced at 97 cents rather than 50. Nobody has a reason to buy six weeks of stock in the last fortnight of December, because the reward for doing so is proportional rather than lumpy.
The objection is that an incremental tier is worth less to the customer at any given threshold, which is true and is fixable by moving the threshold or the rate. Solve for the same payout at plan volume and you have converted the structure without changing the commercial deal.
Testing whether the tier bought anything
Most rebate programmes are never evaluated, because the counterfactual looks unavailable. There is a way to get at it that needs only your own invoice history.
Plot each customer's final period volume as a percentage of their threshold, in narrow buckets. If the tier is doing nothing, the distribution should be smooth across the threshold. If the tier is changing behaviour, you will see a pile of customers landing between 100 and 104 percent and a hole between 92 and 99 percent, because the customers who would have finished just short bought the gap.
Saez (2010), writing in the American Economic Journal: Economic Policy, formalised this as the bunching estimator: the size of the mass at a kink in a schedule identifies how much the schedule is changing behaviour. It was built for tax brackets and it transfers directly, since a rebate threshold is a kink in a price schedule.
What the pattern tells you is worth the analysis. A large pile just above the line with a matching hole below means the tier is buying timing rather than demand, and you should expect the volume back out in the following period. A smooth distribution means the threshold sits somewhere customers are not managing to, which usually means it is either far above or far below where they naturally land, and in both cases the money is a discount rather than an incentive.
Lee, Padmanabhan and Whang (1997), in Management Science, listed price fluctuation as one of the four causes of the bullwhip effect, alongside demand signal processing, order batching and rationing. A retroactive tier is a price fluctuation on a twelve month cycle, and it produces exactly the amplification their model predicts: orders swinging more than consumption, and the swing growing as it moves up the chain.
The accrual is a forecast, and it lands in revenue
Under IFRS 15 and ASC 606, a rebate is variable consideration. It reduces the transaction price, and it has to be estimated and recognised from the first invoice rather than when it is paid. The estimate uses either an expected value or the most likely amount, and it is constrained: you may only include an amount for which it is highly probable that a significant revenue reversal will not subsequently occur.
Read that as an operating requirement rather than as a disclosure note. Every month, somebody has to say which tier each customer will land in at year end, and that judgement goes straight into reported revenue.
Scale it. On 200 million of trade with an average rebate accrual of four percent, the balance carries eight million. Suppose you accrue each customer at the tier they are currently tracking to. If six customers holding 12 million of volume between them fall a tier at year end, you release accrual into income. If six others jump a tier, you take a charge. An error of ten percent on the total accrual is 800,000 landing in a single quarter's gross margin, with no operational event behind it.
That makes the rebate accrual a demand planning output whether or not the demand planning team has ever been asked for it. The forecast that supports it is not the shipment forecast: it is a probability for each customer of crossing each threshold, which is a different question and one that gets more answerable as the year progresses. Running it monthly, with the distance to threshold divided by remaining periods as the driver, gives finance something better than a straight line extrapolation and gives the sales team a list of accounts where a conversation in October is worth more than one in December.
Designing the structure so it can be planned
A few structural choices do most of the work.
Measure quarterly, true up annually. A quarterly measurement with a smaller cliff produces four small distortions rather than one large one, and the annual true-up preserves the customer's total. The last week of a quarter still lifts, and the amplitude is a quarter of what it was.
Cap the retroactive element. Where a retroactive tier has to stay for commercial reasons, cap the retrospective portion at a base volume and pay incrementally above it. The customer sees continuity with the old deal and the cliff shrinks to whatever the cap allows.
Pay for things you can verify. Volume is verifiable and easy to game. Pallet compliance, data transmission, on-time payment and range compliance are all verifiable and hard to game, and each of them has a cost to you that you can compute, which makes the rate defensible.
Publish the schedule to your own planners. The demand plan needs the tier structure as an input. A planner who knows that eleven accounts have thresholds falling in the same week can build the capacity for it, which is cheaper than discovering it.
Taylor (2002), in Management Science, analysed channel rebates where the retailer's sales effort affects demand and showed that a rebate targeted above a threshold can coordinate the channel where a simple discount cannot, provided the threshold is set with the demand distribution in mind. The theoretical result and the practical one agree: thresholds work when they are placed deliberately, and they misfire when they are inherited from last year's spreadsheet.
Where this stops
Rebate structure cannot rescue a programme that is buying nothing. If your bunching plot is smooth and your growth rebates pay out to customers whose volume fell, the money is a price concession with administration attached, and the honest fix is to move it into list price and stop counting it as an incentive.
There is a legal boundary as well, and it binds hardest on the largest suppliers. In the European Union, loyalty inducing rebates granted by a dominant firm have been examined as potential abuses of dominance, and the Court of Justice's 2017 judgment in Intel v Commission established that where the firm produces evidence that its rebate scheme is not capable of foreclosing an as-efficient competitor, the authority has to engage with that analysis. Anyone designing an exclusivity flavoured rebate at high market share should be taking advice rather than reading a blog.
Administration has a floor too. A rebate programme with 400 accounts, bespoke thresholds and manual reconciliation will consume more analyst time than the smallest hundred accounts are worth. Standard published schedules for the tail and negotiated terms for the top twenty is the usual resolution.
Pull the last two years of monthly volume for your top thirty rebate accounts, compute each account's final period share of its annual total, and compare it with the same figure for accounts on no rebate at all. The gap between those two numbers is what your current structure is costing you in the last month of every period.