One search platform credits a conversion for 30 days after a click by default. One professional network credits it for 90 days, and 90 days after a view. Nothing about your tracking causes those two systems to disagree. Their defaults do.

That is one of six documented reasons a consolidated dashboard cannot make platform numbers reconcile. None of them is fixable by better engineering, and knowing which is which is what separates a useful reporting layer from a permanent argument.

This page lists the structural limits, then says what a dashboard is genuinely good for.

The windows differ before you start

The most consequential difference is a default nobody changed.

One search platform. If you do not customise the click-through conversion window when creating a conversion, the default window is 30 days. It can be set from 1 up to 30, 60 or 90 days depending on the conversion source. View-through defaults to 1 day, engaged-view to 3 days, both adjustable from 1 to 30.

One professional network. For most conversions it recommends a 90-day click, 90-day view window, which is the default in its campaign tool. Standard options are 1, 7, 30 or 90 days, with extended windows to 180 or 365 days available for certain conversion categories through its conversions interface.

A short-video platform. Publishes its options, 1, 7, 14 or 28 days for click-through and off, 1 or 7 days for view-through, without stating a default in the documentation I could open.

And one large social platform. Its help centre is served as a client-side application that returns nothing to extractors, so I could not verify its current default myself. I am not going to assert a number I did not read.

What that spread does. A lead that converts three weeks after clicking a professional-network ad is inside that platform’s window and outside a 30-day search window only if it converts after 30 days. Run both channels and the same person can appear in one report, both, or neither, depending purely on timing.

The fix available to you. Align the windows manually across platforms, or accept the difference and write it beside every cross-platform comparison. Do one of the two, not neither.

Default and available conversion attribution windows across advertising platformsTable of the default and available conversion attribution windows published by the major advertising platforms, showing that the same conversion is credited differently by each before any consolidation is attempted. One search platform states that if the advertiser does not customise the click-through conversion window when creating a conversion, the default window is thirty days, adjustable from one day up to thirty, sixty or ninety days depending on the conversion source, with a view-through window defaulting to one day and an engaged-view window defaulting to three days, both adjustable between one and thirty days. One professional network states that for most conversions it recommends a ninety-day click and ninety-day view window, which is the default in its campaign management tool, with standard options of one, seven, thirty or ninety days, and extended windows of one hundred and eighty or three hundred and sixty-five days available for certain conversion categories through its conversions interface. A short-video platform publishes its available options, being one, seven, fourteen or twenty-eight days for click-through attribution and off, one day or seven days for view-through attribution, but does not state a default value in the documentation that could be opened. One large social platform serves its help centre as a client-side application that returns no content to automated extractors, so its current default could not be verified directly and is therefore not asserted here. The practical consequence of this spread is that a lead converting several weeks after clicking an advertisement may fall inside one platform’s window and outside another’s purely as a function of timing, so the same individual can appear in one report, in both, or in neither. The remedies available to an advertiser are to align the windows manually across platforms, or to accept the difference and record it beside every cross-platform comparison.Default windows, before anything elsePlatformClick defaultView defaultOptions publishedSearch platform30 days1 day1 to 30, 60 or 90Engaged-view defaults to 3 days. View and engaged-view adjustable 1 to 30.Professional network90 days90 days1, 7, 30 or 90”we recommend a 90-day click, 90-day view window, which is the default”. Up to 365 for some categories.Short-video platformnot statednot stated1, 7, 14, 28 / off, 1, 7Large social platformnot verifiednot verifiedhelp centre returns nothing30 days against 90 days, on view-through 1 day against 90The same person can appear in one report, both, or neither, purely as a function of when they convert.Align them manually, or write the difference beside every cross-platform chart. One of the two.
A thirtyfold difference in view-through window between two of them, before any question of tracking arises. Source : Platform conversion window documentation (2026)
How differing default attribution windows cause the same conversion to be counted differentlyIllustration of how differing default attribution windows cause the same underlying conversion to be counted differently by two advertising platforms without any fault in tracking. Consider a prospect who clicks an advertisement on a professional network and separately clicks a search advertisement on the same day, then converts at various later intervals. If the conversion occurs at day twenty, it falls inside the search platform’s default thirty-day click-through window and inside the professional network’s default ninety-day click window, so both platforms count it and the consolidated dashboard double counts unless deduplication is applied. If the conversion occurs at day forty-five, it falls outside the search platform’s thirty-day window and inside the professional network’s ninety-day window, so only the professional network counts it, and the search channel appears to have produced nothing despite having been clicked. If the conversion occurs at day one hundred, it falls outside both windows, so neither platform counts it and the conversion appears in the customer relationship management system attributed to no advertising channel at all. A further asymmetry applies to view-through attribution, where the search platform defaults to one day while the professional network defaults to ninety days, a ninetyfold difference meaning that an impression served without a click is credited for three months by one system and for a single day by the other. None of these outcomes indicates a tracking fault: each platform is applying its own documented default correctly, and the divergence is a property of the settings rather than of the measurement.Same lead, different answers, by timing aloneClicks both channels on day zero, converts later.Converts onSearch (30-day)Network (90-day)What your dashboard showsDay 20Counts itCounts itTwo conversions, one leadDay 45OutsideCounts itSearch looks like it did nothingDay 100OutsideOutsideA CRM deal from no channelAnd view-through is worse: 1 day against 90An impression with no click is credited for three months by one and for a single day by the other.Nothing is broken. Each platform is applying its own documented default correctly.
Nothing is broken in either system. The windows simply disagree about how long credit lasts. Source : Platform conversion window defaults (2026)

Some of your conversions were estimated, and you cannot tell which

The second limit is inside a single column, which makes it invisible in any dashboard that reads that column.

What the platform states. In the Conversions column, it reports both modelled and observed conversions.

When it models. Third-party cookie limitations in certain browsers, first-party cookie limitations, consent restrictions in Europe, mobile tracking permission changes, app store policy changes, and cross-device conversions.

The confidence claim. That modelled conversions are only included when the platform is highly confident that conversions actually occurred as a result of ad interactions.

The timing clause that breaks daily reporting. Modelled conversions can take up to 5 days to fully process and stabilise.

What that does to a dashboard. Yesterday’s number is provisional. A dashboard comparing yesterday to the same day last week compares a stabilising figure against a settled one, and will show a decline that is an artefact of the calendar.

The fix, which is simple and almost never applied. Exclude the most recent five days from any trend you act on, or mark them visibly as provisional. One line of logic, and it removes a whole category of false alarms.

What cannot be fixed. Separating modelled from observed. There is no column for it, so any claim your dashboard makes about how many conversions actually happened is a claim about a mixture.

The counting setting that moves totals on its own

A configuration choice, made once, that changes every number downstream.

The two options. Count every conversion that happens after an ad interaction, or count one conversion per ad click.

What the platform says each is for. Every conversion is described as a good choice if you want to track and improve sales, because every sale likely adds value. One conversion is described as a good choice if you are not interested in the number of sales but in whether a certain kind of lead was generated.

What it does to the totals. Every conversion reports higher figures by counting all transactions. One conversion reports lower ones by limiting to a single conversion per click.

Why it matters in B2B specifically. A prospect who downloads three documents in one session is one lead and three conversions. Which number reaches your dashboard depends entirely on this setting, and nobody remembers choosing it.

The dashboard consequence. Two accounts with identical performance can show conversion totals differing by a factor of three. Comparing them, or comparing one to a benchmark, is meaningless without knowing the setting.

The fix. Write the setting next to the metric. If you cannot find out what it is, that is the first thing to establish before anyone acts on the number.

Three separate factors operating inside a single reported conversions columnDiagram identifying three distinct factors operating simultaneously inside a single reported conversions column, none of which is separable by a dashboard reading that column. The first factor is the inclusion of modelled conversions: the platform states that in the conversions column it reports both modelled and observed conversions, and that it models them in response to third-party cookie limitations in certain browsers, first-party cookie limitations, consent restrictions in Europe, mobile tracking permission changes, application store policy changes and cross-device conversions, while asserting that modelled conversions are included only where the platform is highly confident that conversions actually occurred as a result of advertisement interactions. The second factor is timing: the platform states that modelled conversions can take up to five days to fully process and stabilise, which makes recent figures provisional and causes any dashboard comparing yesterday with the equivalent day in a previous week to compare a stabilising figure against a settled one, producing an apparent decline that is an artefact of the reporting calendar rather than a change in performance. The remedy is to exclude the most recent five days from any trend acted upon or to mark them visibly as provisional, which requires a single piece of logic and eliminates an entire category of false alarm. The third factor is the conversion counting setting, offering a choice between counting every conversion occurring after an advertisement interaction, described by the platform as a good choice for tracking and improving sales because every sale likely adds value, and counting one conversion per advertisement click, described as a good choice where the advertiser is not interested in the number of sales but in whether a certain kind of lead was generated. The every-conversion setting reports higher totals by counting all transactions while the one-conversion setting reports lower totals, so two accounts with identical underlying performance can display conversion totals differing severalfold.Three things inside one column1. Modelled and observed, together”In the ‘Conversions’ column, Google reports both modelled and observed conversions.”Modelled for: browser cookie limits, consent restrictions, mobile tracking changes, cross-device.2. Five days before recent figures settle”modeled conversions can take up to 5 days to fully process and stabilize”So yesterday against last week compares a stabilising number with a settled one.3. A counting rule that moves the total on its own”Every conversion”: higher totals, all transactions. “One conversion”: one per ad click.Three downloads in one session: one lead, three conversions. The setting decides which you see.Fixable: drop the last 5 days from trends.One line of logic. Removes a class of false alarms.Not fixable: separating modelled from observed.There is no column for it.
Modelled and observed together, a counting rule that changes totals, and five days before recent figures settle. Source : Google Ads conversion documentation (2026)

And some rows never arrived

The last two limits happen upstream, in the analytics product, before the dashboard queries anything.

Thresholding. Applied to prevent anyone inferring the identity or sensitive information of individual users from demographics, interests or other signals. System defined, not adjustable, with no published numeric threshold.

What that means for a dashboard. If your report includes demographic dimensions, rows may simply not be returned. Your dashboard receives fewer rows and displays them faithfully, with no indication that anything is missing.

Cardinality. Values beyond the table row limit are condensed into an (other) row. Any dimension above 500 values is considered high cardinality, described as guidance rather than a limit.

What that means downstream. Every share and percentage your dashboard computes from the visible rows uses a denominator that excludes whatever sits in the aggregate row. On a high-cardinality dimension that can be most of the traffic.

The attribution regime change. Four models, first click, linear, time decay and position based, ceased to be available in November 2023. Any series crossing that date spans two regimes.

What follows for long trend charts. A two-year channel comparison contains a methodology change in the middle of it, and nothing in the chart says so.

What a dashboard is actually good for

Given all of the above, the useful version is smaller and better labelled than the one most companies build.

Detecting change, not establishing truth. A dashboard is an alarm. It tells you something moved. It does not tell you what is true, and it should not be asked to.

A small number of numbers. Five or six, each with its definition written beside it. Every additional metric is another thing nobody checks and another thing that can break silently.

One named system of record. In B2B, almost always the CRM, because a deal either exists there or does not. When systems disagree, that one wins, and the dashboard says so in writing.

Definitions on the surface, not in a wiki. Attribution window, counting setting, attribution model, and the date range excluded as provisional. Four lines, next to the numbers they qualify.

Trends over totals. Direction is robust to most of the distortions above. Absolute totals are not, because every one of them shifts the level rather than the shape.

And an explicit list of what it cannot see. Thresholded rows, the modelled share of conversions, anything inside the (other) row, and anything before your retention limit. Writing that list once prevents the same discovery being made repeatedly by different people.

Six structural limits of a consolidated marketing dashboard and whether each can be addressedTable listing six structural limits affecting any consolidated marketing dashboard and indicating for each whether the reporting layer can address it. Differing default attribution windows between platforms, with one search platform defaulting to thirty days for click-through and one professional network defaulting to and recommending ninety days for both click and view, can be addressed by aligning the windows manually across platforms or by recording the difference beside every cross-platform comparison. The instability of recent modelled conversion figures, which the platform states can take up to five days to fully process and stabilise, can be addressed by excluding the most recent five days from any trend acted upon or marking them visibly as provisional. The conversion counting setting, offering a choice between counting every conversion after an interaction and counting one conversion per click, can be addressed by recording the chosen setting beside the metric it governs. The inclusion of modelled conversions alongside observed ones within a single column cannot be addressed, because no separate column exists to distinguish them, meaning any dashboard statement about how many conversions actually occurred is a statement about a mixture. Data thresholding cannot be addressed, since it is system defined, publishes no numeric threshold and is not adjustable, so rows may simply not be returned and the dashboard will display the reduced set faithfully with no indication that anything is absent. High cardinality condensation into an aggregate row can only be addressed prospectively by reducing cardinality at collection time, since existing historical data retains the condensed values, meaning every share calculated from visible rows uses an incomplete denominator. A seventh consideration is the withdrawal of four attribution models in November two thousand and twenty-three, which means any time series crossing that date spans two attribution regimes without the chart indicating so.Six limits, three of them fixableFixable in the reporting layerAttribution windows differ, 30 days vs 90Align them, or label the chartRecent modelled figures unstable for 5 daysDrop the last 5 days from trendsCounting setting changes totalsWrite the setting beside the metricNot fixable. Put these on a written list of blind spots.Modelled and observed in one columnNo column exists to split themThresholded rows never returnedSystem defined, no threshold publishedCardinality condensed into (other)Only preventable at collectionAnd one that affects every long trend chartFour attribution models were withdrawn in November 2023. Any series crossing it spans two regimes.
Three are fixable in the reporting layer. Three are not, and belong on a written list of known blind spots. Source : Platform and analytics documentation (2026)
The composition of a marketing dashboard sized to what it can reliably establishDiagram describing the composition of a marketing dashboard sized appropriately to what such an instrument can reliably establish. The dashboard’s purpose is detecting change rather than establishing truth: it functions as an alarm indicating that something has moved, and it does not indicate what is true and should not be asked to. It should contain a small number of metrics, five or six, each accompanied by its definition written beside it, on the principle that every additional metric constitutes another item nobody checks and another item that can break without anyone noticing. It should name a single system of record, which in a business-to-business context is almost always the customer relationship management system because a deal either exists there or does not, and that system’s figure prevails in writing whenever systems disagree. It should carry its definitions on the surface rather than in a separate document, specifically the attribution window, the conversion counting setting, the attribution model and the date range excluded as provisional, amounting to four lines placed next to the numbers they qualify. It should present trends rather than totals, since direction is robust to most of the distortions documented above whereas absolute totals are not, every one of those distortions shifting the level of the numbers rather than their shape. Finally it should be accompanied by an explicit written list of what it cannot see, comprising thresholded rows, the modelled proportion of conversions, anything condensed into the aggregate row, and anything predating the data retention limit, since recording that list once prevents the same discovery being made repeatedly by different people over subsequent months.A dashboard sized to what it can establishIt is an alarm, not an oracleIt tells you something moved. It does not tell you what is true, and should not be asked to.Five or six numbersEach with its definition beside it.One named system of recordIn B2B, the CRM. It wins, in writing.Four definition lines on the surfaceWindow, counting, model, provisional range.Trends, not totalsThe distortions move the level, not the shape.And a written list of what it cannot seeThresholded rows, which never arrivedThe modelled share of conversions, which has no columnAnything inside the (other) row, and anything before your retention limitWriting it once stops four people discovering it separately over the next year.
Five or six numbers, four definition lines, one system of record, and a written list of what it cannot see. Source : Method, applied to documented platform limits (2026)

Where to go next

You want the collection layer underneath. What not to measure.

You want to know what the analytics tool hides. What GA4 does not show.

Your platform and analytics numbers disagree. Why GA4 and Meta conversions do not match.

Your conversion rate is the number in dispute. Conversion rate and its denominator.

You are choosing which metric to steer on. ROAS, MER, CAC or LTV.

You are modelling the mix. Marketing mix modeling for a small company.

In short

  • Default attribution windows differ: 30 days click on one search platform, 90 days click and 90 days view on one professional network, which recommends that setting.
  • A short-video platform publishes options but no default, and one large social platform’s help centre returns nothing to extractors, so I did not assert its figure.
  • Modelled and observed conversions share one column with no label, so any total is a mixture.
  • Recent figures are provisional: modelled conversions can take up to 5 days to fully process and stabilise.
  • The counting setting moves totals independently of performance, between every conversion and one per click.
  • Thresholded rows never reach the dashboard, and no numeric threshold is published for you to design around.
  • The (other) row breaks every percentage computed from the rows that are visible.
  • Four attribution models were withdrawn in November 2023, so long trend charts span two regimes silently.

Name your system of record, put four definition lines on the surface, and write down what the dashboard cannot see. Book a diagnostic, or see how we approach B2B growth.