Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Programmatic Advertising › Troubleshooting

Programmatic CPM Higher Than Planned

Why programmatic buys cost more per thousand than planned: fees in the figure, stacked controls, reseller hops, priced data, loose bids and deal floors.
Troubleshootingintermediate18 min readUpdated

First, decide which price you are reading and what the plan priced

A price per thousand is a ratio, and a gap between two ratios can sit in either half of it. The DV360 reporting guide separates media cost, partner costs, data fees and the billed figure, and says the cost metric chosen changes each rate derived from it, the price per thousand among them. The impression half moves too. The viewability and attention guide explains that the viewable rate rests on the measurable impressions, and that a viewable price per thousand cannot be set beside an ordinary one until that rate is known. A report can therefore sit above the plan while each platform counts honestly, because the plan and the report priced different things.

As general practice, write down the plan's figure and the basis it was priced on before opening the report: media cost alone, media plus partner costs and data fees, served impressions or viewable ones. Then read the report's own lines for the same line items and the same closed period, and use the CPM calculator to convert each basis into an effective price per thousand from the spend and the impressions. Three readings follow, and each aims the search at a different part of this page. If the gap closes once the plan's figure is restated on the report's basis, it was a difference of basis and the first cause is the whole story. If media cost alone sits above the plan's media price, the price of the inventory itself has moved, which is the territory of the controls, the supply path, the bid and the deals. If media cost matches and only the billed figure is above the plan, partner costs and data fees are the difference, and the data fee line shows whether a priced segment is among them. The split aims the search; it is not a verdict.

The aim is a price that matches the plan's basis, not the lowest price on offer. A goal that rewards the cheapest impressions finds cheap inventory the plan did not choose, and the low-quality sites diagnosis starts from that spend and follows it back to the inclusion list, the bid goal, the verification and the supply path. A price above plan can also be the right price for controls that are doing their job, so keep the cost of an impression that met the control's standard, and cost per outcome, beside the price per thousand until the cause is known.

Diagnosis table

Diagnosis

SymptomCheckLikely causeFix
Realised price per thousand sits above the plan's, with a gap that closes or shrinks when the two are put on one basisMedia cost, partner costs, data fees and the billed figure as separate lines for the same period, set beside the basis the plan priced on and the impressions it priced them overA cost figure that includes fees and data costs the plan priced differently, or a rate read on a different impression basis from the plan'sRestate the plan's figure on the report's basis, or the reverse, and judge the gap on that one basis, keeping cost per viewable impression beside it
The price rose after an inclusion list, a suitability tier, a viewability threshold or pre-bid segments were applied, and reach is below what the plan recordedThe controls on the line item listed one by one, against the reach and price the plan expected under themControls stacked until the pool of inventory is small, each competing for fewer impressionsDecide the controls from the objective, relax the ones it does not need one at a time and judge the result on the cost of an impression that met the control's standard, or on cost per outcome
A trusted publisher's inventory costs more than the plan priced it at, and the publisher is offered through more than one routeDelivery by exchange for that publisher, with each route's seller entry in its ads.txt file and the hops in the supply chain objectA fee stack and reseller hops between the budget and the publisher, so more of each thousand is spent before it reaches inventoryShorten the route to each publisher, drop routes whose extra hop brings no extra inventory and ask each vendor in the chain what it charges
The data fee line is large beside media cost on a line item that carries a priced segmentThe data fee against media cost by line item, then the same audience intent bought through a Google audience in a line item of its ownA priced third-party segment added to media cost on every impression it is applied toIsolate the segment, run a Google audience of similar intent beside it, read the fee with media cost and scale only what the result supports
One line item pays more per thousand than a sibling buying similar inventory, and it runs on a fixed bid or is newThe line item's bids, wins and price paid, set beside the sibling's and beside where the fixed bid came fromA fixed bid above what the inventory clears at in a first-price auction, or a new line item whose shading has too few wins to learn fromLower a fixed bid in steps from a start the sibling's prices support, and give a new line item time to learn before judging it
A deal's line item costs more per thousand than the open-auction line for the same publisher, or two deals with one publisher take the same impressionsThe deal's agreed price or floor in its written terms, beside the open-auction line's price for that publisher in the buyer's own report, and the priority of each dealDeal prices and floors above what the open auction clears at, or several deals with one publisher competing for the same impressionsRead the premium as the price of what the deal adds, move budget away from deals that add nothing the open auction cannot, and retire or merge deals that compete

When none of the six explains the gap. A buyer who made none of these choices may be seeing the market and not a fault in the buy. The auction pricing guide explains that the other side of an auction sits outside the account, and that price per thousand typically climbs around peaks as competitors arrive. This page does not size that effect and gives no market figure. Compare the line only with the buyer's own cost history for the same period, kept on a common basis, and read a gap that the history shows too beside the conditions recorded on each history row before calling it the season's price.

Fix, verify and prevent each cause

Fees and data costs the plan priced differently, or a rate read on a different impression basis from the plan's

The separate lines in the DV360 reporting guide matter here because fees sit outside media cost but are paid from the budget, so they alter how far it stretches. A plan that priced media alone and a budget that pays for everything on the invoice disagree about the number of impressions the money purchases. The second half of the symptom sits in the denominator. A price read per served impression and a price read per viewable impression describe the same money, and the impression calculator walk-through for programmatic deals gives the arithmetic. Neither figure is a fault. They answer different questions, and a plan written for one cannot be checked against a report that shows the other.

Fix: pull media cost, partner costs, data fees and the billed figure for the same line items and the same closed period, and use the calculator to read the price per thousand on media cost alone and on the billed figure. Set each beside the plan's figure restated on the same basis. Where the plan priced media only, restate the plan: add the fees, each restated per thousand impressions, and, for the comparison that matters, divide by the viewable rate on served impressions. This is general practice and changes nothing in the platform.

Verify: the gap, read on one basis, is either gone, which closes the diagnosis, or it is the real gap that the remaining causes have to explain; and the difference between media cost and the billed figure is accounted for by the partner cost and data fee lines, with no remainder nobody can name.

Prevent: write the cost metric each price is read from into the measurement plan, as the reporting guide asks of each key performance indicator, and put the basis beside the plan's price per thousand in the plan itself.

Controls stacked until the pool of inventory is small

The brand safety and viewability guide walks through how the cuts compound, from an inclusion list through a suitability tier and a viewability threshold to pre-bid segments, and why the shortfall gets blamed on the market. The delivery side of the same filters is in the diagnosis for a DV360 line item that spends too little, which reads stacked filters from the bid requests that do not arrive. This page reads them from the price.

Fix: list the controls on the line item one by one and set against each the reason the objective needs it. Relax the ones it does not need, one at a time, and read price and reach after each change. The safety floor stays on; the aim is a pool sized to the objective, not a cheaper pool. Where the inclusion list is the control, the inclusion list tutorial gives the way out when the list leaves too little reach.

Verify: the next report shows what each change did to reach and to the price per thousand, and the comparison that decides it is the cost of an impression that met the control's standard, or cost per outcome, because the raw price looks better without controls. A raw price that falls while cost per outcome rises has fixed nothing.

Prevent: follow the brand safety guide's discipline for controls, and add them in sequence from what monitoring shows rather than all at once, so that a price rise can be matched to a control.

A fee stack and reseller hops between the budget and the publisher

The ecosystem guide traces the fee layers between a budget and a publisher and names the standards that make the path visible, among them the SupplyChain object that the OpenRTB specification places in the bid request, so they are not repeated here. What a longer path does to a price is arithmetic: more of each thousand is spent along the way before it reaches inventory.

Fix: read delivery by exchange for one publisher, set each route's price beside its ads.txt entry and its hops, shorten the route as the ecosystem guide sets out, and ask the DSP, the data vendor, the verification vendor and the ad server what each charges, writing the answers beside the media price.

Verify: the same publisher's impressions arrive through the shorter route in the next report, delivery holds where the same impressions are on offer there, and the report by exchange shows that publisher's price after the change beside its price before.

Prevent: take the look at where the money went that the ecosystem guide asks for each quarter, and write what each fee charges beside the plan's price.

A priced data segment added to media cost

The audiences guide owns how a priced third-party segment is charged and reported, and the Help Centre it cites covers fees where priced segments are combined, so this page states no rule for that. Google audiences carry no data fee, and the guide uses one of similar intent as the comparison. A pre-bid vendor segment is priced per thousand impressions too, and the brand safety guide advises checking the vendor's terms on whether the fee is paid on impressions bid on but not won.

Fix: read the data fee line beside media cost for each line item that carries a priced segment, isolate the segment in a line item of its own, run a Google audience of similar intent beside it, and judge both on cost per outcome with the fee counted, not on media cost alone. Keep the segment only where its result justifies its fee.

Verify: the two line items' results, read with the fee counted, show which audience earns its price, and where the segment is dropped, the data fee line no longer carries its fee on the impressions it no longer touches, and the billed price is read again beside media cost.

Prevent: test a priced segment on its own before it joins a prospecting line item, and record its fee beside the plan's price. The programmatic advertising audit scorecard scores the same habit from outside, asking whether each bought segment earns its fee in results the business can see.

A fixed bid above what the inventory clears at, or a line item still learning

In a first-price auction a bid set too high is also the price. The real-time bidding guide explains bid shading and why a new line item, with few wins to learn from, can cost more per thousand at the start than it does later. This page does not describe how any one platform shades a bid. It reads the outcome in the buyer's own report.

The bidding and pacing guide covers the other half. A fixed bid is a maximum price set for each impression, which makes it a sound start on a new inventory source because it exposes what the inventory clears at, and it carries the risk of overpaying in cheap auctions. The guide's funnel table compares a line item's win rate with a sibling line item's for a bid set too low; reading the same comparison for a bid set too high is general practice. A fixed bid copied from another publisher's flight carries that publisher's clearing prices into inventory that may clear lower.

Fix: set the line item's bids, wins and price paid beside a sibling buying similar inventory. Where the fixed bid sits above what the sibling's prices support, lower it in steps and read wins and price after each step. Change the bid on its own, not in the week the strategy or the target changes. Both are general practice, and the guide's warning about restarted learning concerns an automated strategy. For a new line item, give it time and judge it against its sibling once it has wins to learn from, not in its first days.

Verify: the price paid per thousand moves towards the sibling's while the win rate holds up. A win rate that collapses says the step went too far, and the troubleshooting view's funnel then shows the collapse between bids and wins.

Prevent: start a fixed bid from what a sibling's prices or the buyer's own history on that inventory support, not from another flight, use the first flight on a new source to learn what it clears at, and record the bid and the price paid in the change log beside the date.

Deal prices and floors above the open auction, or deals competing for the same impressions

The deal types guide sets each deal type's price and floor against buying the same impressions in auction, says that several deals with one publisher can compete for the same impressions, and says that only a side-by-side test shows what the premium is worth. This page reads that gap in the buyer's own report.

Fix: set the deal line item's price per thousand beside the open-auction line item's price for the same publisher in the buyer's own report, both on the all-in viewable basis described under the first cause. Read the gap as the premium, then ask what the deal's terms say it buys. Where a deal adds nothing the open auction with an inclusion list does not, move its budget back. Where only access matters, a preferred deal pays its fixed price on impressions the buyer takes up. Where two deals with one publisher compete, retire or merge the one the plan does not need.

Verify: the premium, restated on the all-in viewable basis, is explained by something in the terms or the results, whether access, data, certainty or a better cost per outcome, and each remaining deal's delivery shows impressions it wins itself and not impressions another deal would have taken.

Prevent: agree the terms in writing before launch, as the deal types guide sets out for negotiating, and record the premium each deal carries beside the plan's price.

Prevention: a plan written on the same basis as the report, changes made in turn and a price recorded beside each setting

The habits below move the diagnosis from the day the price looks wrong to the day the plan is written. Each is general practice unless a guide is named.

Write the plan on the report's basis. The media planning framework treats the price per thousand as the planning currency for reach channels and asks the plan to state both the price it expects and the volume that price buys. Add the basis to both: media only or all-in, served or viewable, with fees listed as their own lines, so that the plan and the DV360 report name the same cost metric and two people pulling the figure get the same figure.

Make changes in turn. Change one setting at a time and let learning settle before reading the result. The bidding and pacing guide warns that a change to a strategy restarts learning, and the brand safety guide warns that applying all controls at once makes a shortfall impossible to attribute. A buyer who changes a bid, a list and a segment in the same week learns that the price moved and nothing about why.

Record the price beside each setting. In the change log, write against each change the setting, the date, the price per thousand on the plan's basis, the cost per viewable impression and the cost per outcome, first before the change and then after the learning period. The next rise in price can then be matched to a change or ruled out as one, and the log is the buyer's own history for the line.

Hold the routine. The programmatic campaign management playbook holds the rhythm for this, from the reconciliation to the close-out that feeds the cost history. A price that moves with no change in the log goes through the six causes first and only then to the closing paragraph of the diagnosis table.

A fretwork kit seller reads its price (illustrative)

Illustrative example — not a real result

Where the extra price came from

Dravhuset, an invented maker of fretwork clock cases and kits, plans a four-week open-auction display flight with a budget of £9,000. The plan prices it at £3.00 per thousand impressions on media cost alone, which would buy 3,000,000 impressions. The closed flight's report shows 1,200,000 impressions served, a viewable rate of 75 per cent of served impressions and a billed figure of £9,000, a price of £7.50 per thousand. Media cost is £5,400, or £4.50 per thousand, and the rest is £1,200 of data fees and £2,400 of partner costs. On the plan's own basis, media cost alone, the gap is £1.50 per thousand; the other £3.00 of the difference between £7.50 and £3.00 is data fees and partner costs the plan did not price.

The team then looks for what sits behind the gap. The data fee line is a priced third-party in-market segment, and a pre-bid segment's fee is £900 of the fees in this invented report, so the two together account for £2,100 of the £3,600 in fees and data. The line items also stack an inclusion list of 80 domains with a viewability threshold, which removed much of the pool the plan priced. And a fixed-bid line item set at £6.00, copied from another publisher's flight, pays £6.00 per thousand against £3.80 for a sibling on similar inventory, with a win rate of 62 per cent against the sibling's 35 per cent.

The team changes one setting at a time and writes the price on the plan's basis beside each change in its change log. It gives the segment a line item of its own and runs a Google audience of similar intent beside it, widens the list by 40 domains from the next candidates in its ranking and lowers the £6.00 fixed bid in steps, reading wins and price after each, to £4.00. Before judging each change it reads cost per viewable impression and cost per order beside the price per thousand. Once the last of the three changes has settled, a flight of the same size bills £6.00 per thousand, so the £9,000 buys 1,500,000 impressions, and the viewable rate eases from 75 per cent to 68 per cent. The all-in price per thousand viewable impressions moves from £10.00 to about £8.82, and kit orders rise from 36 to 52, so the cost per order moves from £250 to about £173. The price per thousand fell by a fifth, cost per viewable impression by less and cost per order by nearly a third, and the team judges the changes on the last two. The segment's cost per order, with its fee counted, is read against the Google audience's before the segment is kept in prospecting. Each price, fee and rate here is an assumption for the invented business, not a market figure.

The programmatic troubleshooting section places this page beside the diagnoses for delivery, conversion counts, placements, viewability and frequency, so a price that moves together with any of them can be read from both sides.

Common mistakes

  • Comparing a billed figure with a media-only plan. The report and the plan name different costs, so the gap is a difference of basis until both sit on one basis.
  • Reading the price on served impressions alone. The viewable rate changes what a thousand impressions are worth; read cost per viewable impression and cost per outcome beside the price per thousand.
  • Removing a control to lower the price without reading what it was for. The price falls and the reason the control was there comes back.
  • Chasing the cheapest price. A goal that rewards cheap impressions finds inventory the plan did not choose, and the low price hides where the money went.
  • Accepting a priced segment without a test. It adds its fee to each impression it touches; give it a line item of its own and a Google audience as the comparison before it scales.
  • Copying a fixed bid from another flight. The bid carries another publisher's clearing prices into new inventory; start from a sibling's prices and read the win rate.
  • Reading a new line item's price too early. Its shading has a short win history to learn from, and early prices run high.
  • Treating a deal's premium as a fault or as free. It is the price of what the deal adds; test it against the open auction before keeping it or cutting it.
  • Changing more than one setting at a time. The price moves and nothing says which change moved it.
  • Calling it the market without the history. Read the buyer's own history for the same period, beside its recorded conditions, before calling a gap the market.

Frequently asked questions

The report's price per thousand is above the plan's. Does that mean the buy is overpaying?

Not yet. First check that the two figures sit on the same basis. A plan priced on media cost alone and a report that includes partner costs and data fees will disagree without anything having gone wrong, and a price read per viewable impression looks dearer than the same money read per served one. Restate one figure on the other's basis, then judge the gap, and read cost per viewable impression or cost per outcome beside it before treating a price as a fault.

Should we remove a control to bring the price down?

Only after reading what the control is for. Each control removes inventory, so a stricter list, tier, threshold or segment raises the price per thousand while it narrows the pool. Whether that trade is worth making depends on what the objective needs, and it is judged on what an impression that met the control's standard costs, or on cost per outcome. Relax one control at a time, read price and reach after each change, and keep the safety floor the written policy requires.

Could the market explain the gap even if no cause on this page applies?

It could. A buyer who made none of these choices may be seeing competition for the same impressions, and the auction pricing guide notes that competition and season raise the price of attention. This page sizes no market effect and gives no market figure. Compare the line with the buyer's own cost history for the same period, kept on a common basis, and read a gap that the history shows too beside the conditions recorded on each history row before calling it the season's price.

Terms used on this page

Tools to try

Frameworks

Sources

  1. Display & Video 360 Help — Google
  2. OpenRTB specification — IAB Tech Lab
  3. Google Ad Manager Help — Google

Related