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

Affiliate Marketing › Affiliate Measurement

Setting the Affiliate Lookback Window From Your Own Orders

How to set an affiliate lookback window from the programme's own orders: read click-to-order delay by partner type, price each option and record the choice.
Tutorialintermediate20 min readUpdated

Why a window chosen by habit decides who is paid

The tracking guide describes the window as a separate commercial setting, and the commission structures guide covers how its length follows the kind of buying. Both add that the length should come from how long the business's own customers take, not from what other programmes do. Neither explains the working: what to read, where the two times come from, how the delay differs between partner types, or what another length would cost. This tutorial is that working.

A window copied from the first network, or left at a default nobody read against the business's own orders, still decides something. An order that arrives after the window has closed pays the partner nothing, however long that partner's page had been shaping the decision. An order inside it can earn commission, however little the click mattered. If the window is shorter than a partner type's customers take, that type is paid for fewer of the orders it started. If it is longer than they need, the business pays for orders the click barely touched. The window moves money between partner types, and between partners and the business, which is why it is read as a price.

Neighbouring settings need keeping apart. The lookback window decides whether a click still counts. The credit rule decides which eligible click is paid. The validation period decides when a credited order can be approved. Analytics keeps its own lookback as well, a property setting that this tutorial does not change. The steps below set each in its place, and nothing here states how any platform, network or analytics tool behaves beyond what the linked guides say.

Before you start

  • The order system. Order references, the time each order was placed, its status, the customer's type (new or returning) and whether it carries a stored click identifier. Ask the web or order-system owner whether a click time is stored beside the identifier.
  • The platform's conversion records for the same orders. The partner, the partner's type as the programme classifies it, the status and, where the platform shows them, the click time and the conversion time.
  • Analytics access to the attribution paths report. It is the source for the wider journey in the fifth step.
  • The current terms and the platform's settings. The window, the validation period, the credit rule and the date each last changed, read from the platform itself rather than from memory.
  • The commission schedule and finance's ceilings. Rates by partner type and customer type, and the margin ceiling for each segment. Finance owns the margin figures, and the window is priced against them, not the other way round.
  • The returns window and the time the business needs to match orders. They set the validation period in the eighth step.
  • Owners. Who reads the orders, who decides, who speaks to finance and who sends the partner notice. Write them down before reading begins.

The steps

  1. Say what the window decides, and for whom

    Write the question down before opening any data. The window decides how long after a partner's click an order still counts for that partner, so the question is for which partners, and for which customers, the present window pays too little or too much. Write the present window, who set it and why, and whether it came from a platform default, a template or the business's own reading. If nobody can say, treat it as habit and read the orders before defending it.

    Name the people the decision touches. Partners whose audiences take time to decide gain from a longer window. The business gains from a shorter one wherever the extra orders were barely influenced by the click. Finance sees the effect as commission on orders already placed, not as new sales, and the web or order-system owner decides what click information an order record can hold. Settle who owns each of these before the reading starts. Remember too that the window is one setting among several: the credit rule and the validation period are set in later steps, beside it.

  2. Choose the orders to read

    Read the business's own orders, not only the ones the platform credited. The orders the platform credited fell inside the window it applied, so they cannot show how many fell just outside, while the order system holds each order with its reference and time. Take a closed stretch of orders that carry a stored click identifier.

    End the stretch early enough for the late orders to have arrived. A report for the latest days keeps filling while the conversion lag plays out, so a stretch that runs up to today makes late orders look rarer than they are. Cover ordinary trading and, where the history allows, a busy and a quiet time. A partner whose trade follows the calendar, for example a gift guide, is read within its own seasonal pattern and beside the same period of the previous year, because mixing a busy and a quiet period without saying so blurs the delay.

    Set aside the orders the delay cannot describe: test and staff orders and duplicates, which the terms exclude anyway, and orders whose click time was not recorded. Count those apart so the gap stays visible. Keep returned and cancelled orders in the reading, since the delay describes when customers bought, not whether the order stood, and flag them so they can come out when the cost is priced. Write down how the orders were chosen, with the dates, the filters and the count after each filter. A reading nobody can reproduce cannot be defended when a partner asks.

  3. Take click time and order time for each order

    Each order needs two times. The order time comes from the order system, which holds the time the order was placed. The click time comes from the stored click identifier: the tracking guide has the identifier kept on the advertiser's side and passed to the order record, and the platform ties the identifier to the click it logged. Ask the platform for its conversion records with click times and conversion times for the orders it credited, and ask the web or order-system owner whether the business stores a click time of its own, which is a source for orders the platform did not credit. Where a source does not exist, record that instead of estimating the time.

    Join the two on the order reference and build one table: order reference, partner, partner type, new or returning customer, click time, order time, status, and whether the platform credited the order. Check that both times use the same clock and time zone, and that the order time is when the order was placed, not when it was approved or shipped. Where an order carries more than one click, decide before reading which click starts the clock. Using the click the programme's credit rule would credit keeps the delay and the payment decision on the same footing. Write the choice down.

    Take the delay as the order time less the click time, in whole days. Use a smaller unit only if the range sells goods people buy within hours of a click.

  4. Read the delay for each partner type and customer type

    Group the table by partner type, using the types the programme already works with, and by new and returning customer. Within each group, sort the delays and read the shape, not the average, since a few late orders pull an average up and hide a second cluster of late orders. The page on conversion windows and attribution reads a delay distribution for an advertising platform's windows. The decision differs: an advertising window shapes what a platform learns, while an affiliate window is a price paid to a partner, which is why the next steps price it.

    The delay may differ by partner type, since the commission guide ties longer windows to content partners and considered purchases, and partners at the end of the journey may sit close to the click. Let the data show how far. If the range spans considered and impulse purchases, split by product category as well, as far as the counts allow.

    Show the count of orders beside each group. The number of orders in a group is the sample size behind its distribution, and a group with few orders gives a shape that would move a great deal if the stretch were repeated. The small numbers guide covers ranges, minimum counts taken from the programme's own past figures, and pooling similar groups. Apply it here: set a minimum, mark a group below it as too few to judge alone, and ask for more orders when money moves on the result.

  5. Add what analytics shows of the wider journey

    Credited orders show only the part of a journey that sat inside a partner's window. For the wider picture, open analytics. The attribution paths report shows path length and the time from the first touch to the key event. The attribution settings guide treats those as inputs when a length has to be picked. Restrict the report to the purchase alone, so that other key events do not blur the picture. Read it as context: how long the customers analytics recorded take from their first recorded touch to the purchase, which is wider than one partner's click-to-order delay.

    Analytics keeps its own window and its own model, so its figures can differ from the platform's, and nothing in this step changes them. The diagnosis of sales that do not match analytics asks that each side's window and credit rule be written down with the date each last changed, which the record in the last step does for the platform's side. Where the order system also keeps the first visit and its time against the order, add a first-visit-to-order delay for the same orders. It comes from the business's own records and gives a second reading beside the paths report.

  6. Price each candidate window

    Choose candidates from the delay reading: the present window, a shorter one that drops the thin tail of late orders, and a longer one that reaches the next cluster of late orders in the group you care about. Name each by the day it ends. For each candidate, count from the table the orders it would credit and subtract the orders credited now. The result is the extra orders the candidate credits, or the orders it would stop crediting, for each partner type and customer type.

    Those extra orders have already been placed, so a longer window adds cost without adding sales, which is the substance of the commission guide's reason for shorter windows: a long one credits orders the partner contributed little to. The cost of a candidate is the commission on the extra orders, at the rate and on the order value the terms set for that partner type, plus the platform's cost on each order if it charges per order. This pricing is the tutorial's own method, offered as general practice and not as how any platform behaves.

    Hold the result against the segment ceiling. The commission guide builds the ceiling from the order's contribution margin, less the platform's share and the profit the business requires, and finance supplies the margin. The break-even calculator turns margin and order value into that ceiling. Compare the commission on one extra order with the ceiling for its segment, then compare the total extra commission for the stretch with what the business wants to pay for orders whose click came well before the order. The two comparisons can disagree: an order can fit under the ceiling and still be an order the business would rather not pay for.

    Decide too whether the programme wants one window or several. If the platform can hold a different window for a partner type, the choice has a cost of its own, with more settings to keep identical in the terms and the platform and a notice to write for each. If the platform holds one window only, the delay reading still shows which partner types the single value serves and which it leaves out.

  7. Set the window beside the credit rule

    The window decides which clicks are eligible; the credit rule decides which eligible click, or which channel, is paid. The attribution and deduplication guide shows how they interact. Which pairing the business wants is a choice to make here, not one to inherit.

    Write the pair as one decision. For each candidate, state which channels can override an affiliate click and for how long before the order, and how clashes between affiliates are settled, including any protected partners. The deduplication rule has to exist in writing before the price of a window is final, because the price depends on which extra orders the rule would hand to someone else.

    Then return to the prices from the previous step and rerun the count with the rule in force. For each extra order a candidate would credit, look in the first-party record of recent touchpoints for a later click from an overriding channel, or a later click from another affiliate the rule favours, and take those orders out of the extra orders. The commission on what is left is the price of the window. Orders the rule gives to someone else cost nothing in commission and should not be priced as though they did.

  8. Set the validation period and the payment date

    The validation period has its own job and is set from the returns window plus the time to match orders, not from the lookback window. State which event starts it, so that a partner can read the date commission becomes payable. A change to the window leaves it where it is unless the returns window or the checking time has changed.

    Tell finance what a longer or shorter window does to the figures it sees: more or fewer credited orders for the same clicks, and commission on them that has to be estimated at the close. The month-end close tutorial holds back conversions that are still inside the validation period and asks for an estimate of the commission on them.

    Write the payment date the terms promise beside the other two, so a partner can see the times together: how long a click counts, how long the business has to check, and when approved commission is paid.

  9. Write it into the terms and the platform and send it through the change gate

    State the window in the terms' timing clause in the same words as the platform setting, with the validation period and payment timing beside it. The timing and credit step of the terms tutorial says how to write it. Use the same wording wherever the window is described, so there is one description of it.

    For a running programme, a change to the window alters what partners are paid, so it goes through the affiliate commission change checklist, which names a longer or shorter window among the changes it gates. For a programme that has not opened, the launch checklist asks for the window to be set and documented before the programme opens. The business's legal adviser reviews the wording. Nothing here is legal advice or a statement of what the terms commit the business to.

  10. Check the first period and record the evidence

    Read the first period after the change for credited orders, commission cost and partner questions. Wait until the new window has had time to play out for the clicks being read, as in the second step, or the late orders priced earlier will not have arrived and the period will look cheaper than it is. Compare with what the pricing expected, partner type by partner type, since a change made for one type can move another's figures. A before-and-after reading is weak evidence, because the season and other changes move the same figures.

    A partner who writes that orders arrived after their window closed, or that late orders went to someone else, is making a claim about specific orders. The tracing tutorial takes such a claim from the partner's own evidence through to a cause.

    Record the evidence in one place: the dates, filters and counts used to choose the orders, the table of delays by group with its counts and any group set aside, the candidates with their prices and the credit rule beside them, the window chosen with the reasons and the alternatives declined, the validation period and payment date, the notice date, the version of the terms and the date the platform setting changed, who decided, and what will prompt another reading: a change to the product range, the partner mix, the credit rule or the returns policy. A reading that is recorded can be repeated. One that is not leaves the next owner with the habit this tutorial started from.

The decision at a glance

ChoiceWhat it creditsWhat it costsWho gainsWhat to check beside it
A shorter window than nowFewer late orders; the thin tail of delayed orders stops countingLess commission on orders already placed; partners with slow audiences are paid for fewer of the orders they startedThe business, where the dropped orders were barely influenced; impulse categoriesThat the cut does not fall on the partner types the business wants; the notice
The present window, kept deliberatelyThe same orders as nowNo change in commissionNo one movesThat the reading supports it, and what will prompt the next reading
A longer window than nowMore late orders, from partners whose audiences take time to decideMore commission on orders already placed, which may not be additionalContent and review partners; considered purchasesThe ceiling, the credit rule, the validation period, and what a comparison holding a group back would show
A different window for each partner typeEach type's own delayMore settings to hold identical in the terms and the platform, and notice for eachTypes a single value serves badlyWhether the platform can hold them at all

What the data can and cannot say

The reading comes from orders that carry a stored click identifier and a recorded click time. It therefore says nothing about customers who declined consent where consent is required, for whom the tracking guide's position is that the identifier should not be stored, nor about buyers whose click was not recorded, and it cannot show a delay longer than the life of the stored identifier. Delays beyond the present window cannot be read from the orders the platform credited, because those orders stopped counting when the window closed. Longer delays show only in the business's own order records, where an identifier was kept, and in the analytics paths report, so take any longer-delay evidence from those two sources and say so beside the figure. What a platform retains of clicks that did not lead to an order, or of orders that came after its window, is for the platform to say, and this tutorial states nothing about it or about any platform's default.

A delay reading counts orders that followed a click. It does not show that the click caused the order. The hierarchy of evidence places attributed figures low, because they include buyers who would have purchased regardless, and a longer window pays for more orders of that kind. If the open question is whether a partner type brings in orders the business would not otherwise have had, a better window will not settle it; a comparison that holds a group back answers it best. The same ranking says a before-and-after reading of the first period is weaker than that comparison.

Where the payment the programme pays for arrives long after the click, as with a subscription whose first payment follows a trial, the click-to-order delay measures the wrong interval. The subscription programmes guide shows how to choose the paid event and where to keep the identifier.

Time and size also limit the reading. A stretch that mixes a busy and a quiet period describes neither, a group below the minimum describes itself poorly, and a stretch that ends too close to today has not yet received its late orders. Prices, ceilings and the choice of partner types remain the business's decisions, agreed with finance. Nothing here is accounting or legal advice.

A loft ladder seller sets its window (illustrative)

Illustrative example — not a real result

Mardelwyn Loft Ladders reads its window from its own orders

Mardelwyn Loft Ladders, an invented seller of loft ladders and loft hatches, has a 30-day window it copied from its first network and has not read against its own orders. Buyers measure up the opening first, and some wait for a builder's quote. The programme manager asks the web owner for a closed quarter of orders that carry a stored click identifier: 1,180 orders. Of these, 96 have no click time and are counted apart, and 22 are test, staff or duplicate orders, leaving 1,062 to read. Of those, 340 came from home improvement blogs, 610 from voucher and cashback sites and 112 from a few other partners, which are pooled. The quarter ended long enough ago for late orders to have arrived.

Days from click to orderBlogsVoucher and cashbackOther partners
Within 711658570
8 to 301062124
31 to 6071312
61 to 904716

The voucher and cashback sites' orders sit close to the click, while 118 of the blogs' 340 orders arrive after the present window has closed. Those later orders appear only in Mardelwyn's own order records, not in the platform's credited orders, because the platform stopped counting them. The analytics paths report, filtered to the purchase, shows a long time from first touch to purchase, which fits the blogs' pattern.

Mardelwyn prices the present window and two longer ones. Its credit rule lets a paid search click close to the order override an earlier affiliate click, and that rule stays as written. The average validated order from the blogs is £260 and their rate is 7%, so one order carries £18.20 of commission, and the table prices every extra order that no later paid search click wins at that figure as a simplification, since 71 of the 86 extra orders at 60 days and 118 of the 140 at 90 days are blog orders. Finance's ceiling for the blog segment is £59.60 an order: contribution of £98.80 before commission, less £4.20 of platform cost and £35 of required profit.

WindowOrders inside the windowExtra ordersOf those, won by a later paid search clickExtra commission
30 days (now)922nonenonenone
60 days1,0088619about £1,220
90 days1,06214033about £1,950

Each extra order sits well inside the ceiling. The table counts commission only; the £4.20 platform cost on each extra order would add about £280 at 60 days (67 orders), about £450 at 90 days (107 orders) and about £170 for the 40 orders that separate the two longer windows, which does not change the order of the options. The choice is therefore not whether the money fits but whether the click is worth paying for. Moving from 60 to 90 days adds 54 extra orders, 14 of them won by a later paid search click, so 40 orders and about £730 of commission, all with a click more than 60 days before the order, which the team judges too weak a link to pay for, and records as its own judgement. It sets a 60-day window. The platform holds one window only, so the blogs and the other partners share it.

The validation period comes from a returns window of 28 days plus 7 days to match orders, which gives 35 days, and the terms are changed to state it beside the window and the payment date. The change goes through the change checklist with the 30 days of notice its terms promise, applies to clicks made on or after the effective date, and reaches the two largest blogs by phone before the general message. At the first close finance is told to expect more credited orders. Mardelwyn then watches credited orders, commission cost and partner questions, reads the first period only after the late orders have had time to arrive, and writes the table, the prices and the reasons into one record with a note to read again if its range, its partner mix or its credit rule changes.

For the other routines that keep affiliate figures trustworthy, the measurement hub lists them beside this tutorial.

Common mistakes

  • Copying the window from the first network or a template. A setting nobody read against the programme's own orders still decides who is paid.
  • Reading only the orders the platform credited. They stopped counting when the window closed, so the late orders in question are not among them.
  • Averaging the delay. An average hides a second cluster of late orders. Read the distribution for each partner type.
  • Judging a group on a handful of orders. Show the counts, set a minimum from the programme's own history and pool or set aside the groups below it.
  • Reading a longer window as sales gained. It credits orders the business has already received, and the commission is the cost.
  • Setting the window and the credit rule apart. The two interact, so the pair is written down as one decision.
  • Changing the platform before the notice. The terms and the platform carry the same words, and the notice comes first.
  • Treating analytics' window as the programme's. Analytics keeps its own window and model, and each side's setting is recorded.
  • Keeping no record. Without the table, the prices and the reasons, the next reading starts from habit again.
  • Reading the first period too soon. Late orders have not arrived yet, and the change looks cheaper than it is.

Frequently asked questions

Can different partner types have different windows?

Whether the platform can hold more than one window is a question for the platform, so ask it before designing around the idea. If it can, the terms have to say which window applies to which partner type, and each change needs its own notice. If it holds one window only, read the delay by partner type anyway: the reading shows which partner types a single value serves well and which it leaves out, and the choice is then made knowing who loses.

Does a longer window mean more sales?

No. A longer window credits more of the orders the business has already received to a partner's earlier click, so commission rises without any extra order reaching the business. Whether those orders were added by the partner is a separate question that a comparison holding a group back answers best. Price the longer window as extra commission on existing orders and decide whether the click earlier in the journey is worth paying for.

What if a partner type has too few orders to read its delay?

Do not judge it alone. Read a longer stretch of orders, pool it with the partner type closest to it, show the range instead of a single figure and mark the group as too few to judge on its own. Set the window on the larger groups, record that the small group was not read separately, and read it again once more orders exist.

Terms used on this page

Tools to try

Frameworks

Sources

  1. Google Analytics Help — Google Analytics Help

Related