Multi-Touch Attribution: What It Reveals and Its Limits

Most multi-touch attribution models are assigning credit correctly given what they can see. The problem is what they cannot see: every platform observes only its own events, analytics tools drop whole categories of visit from credit by design, and device breaks, consent choices and closed platforms remove the rest. So the key lever is measurement coverage and incrementality validation, not model choice. Before you debate linear versus Shapley, find out which parts of the path your data can observe.

Paid social teams lose whole Monday mornings to dashboards that are each internally consistent and collectively unhelpful. One platform claimed the conversion. Another claimed it assisted. The order system confirmed the purchase, but none of the reports could establish which exposure created demand and which one merely appeared near the finish line.

That's the uncomfortable role of multi-touch attribution. It can organize the journey, expose recurring paths, and help a creative team understand where an ad tends to appear. It can't turn incomplete tracking into causal truth. The useful question isn't only “Which model should we use?” It's also “How much of the journey can we see, which blind spots distort channel value, and how does the signal change this week's creative decisions?”

Why do attribution dashboards disagree on Monday morning?

At 9 a.m., the growth team opens its weekly review. Meta Ads Manager reports a large attributed conversion value. TikTok's multi-touch view assigns many of those same purchases to upper-funnel reach. The finance sheet contains one revenue number. The two dashboards contain competing stories.

The CMO asks which platform is lying. Nobody answers, because both systems may be technically correct inside their own measurement boundaries. Each platform sees its own impressions, clicks, modeled events, and identity signals. Neither necessarily sees the complete journey, including the exposure that happened on another device, the creator video that wasn't clicked, the organic search that followed, or the offline influence that never entered the event stream.

The team's real decision sits in the disagreement. Should the next budget increase fund retargeting, prospecting, creator-led video, or branded search protection? Should the creative queue produce another urgency-led closer, or more context-setting ads that introduce the product before the shopper shows intent?

A report on paid social reporting workflows can make the review easier to run, but a cleaner report won't resolve missing signal. Reporting organizes evidence. It doesn't prove that the most measurable touchpoint caused the order.

Practical rule: Treat attribution as a coverage and validation problem before treating it as a model-selection problem.

Four questions decide what to do about it:

• Which model helps with which decision?
• How much of the customer journey is tracked?
• When should MTA yield to incrementality testing or MMM?
• How should the resulting signal shape briefs, tests, and creative verdicts?

The answer is less satisfying than a single winning dashboard. Multi-touch attribution is valuable when the team uses it as a structured hypothesis engine, checks its assumptions against controlled evidence, and sends the useful parts into the weekly creative loop.

What does multi-touch attribution actually tell you?

Multi-touch attribution, or MTA, is a measurement approach that allocates conversion credit across more than one touchpoint in a customer journey. The American Marketing Association puts it plainly: MTA models "provide estimates to allocate conversion credit to some or all of the media/channel touchpoints customers encounter in their purchase journey." The word doing the work there is estimates.

That separates MTA from first-touch and last-click attribution. A first-touch model gives the opening interaction all the credit. A last-click model gives the final trackable interaction all the credit. MTA keeps multiple observed events in the path, then applies a rule, probability, or algorithm to divide the outcome among them.

Consider a typical paid social path:

• A TikTok spark ad appears during a lunch break on Tuesday.
• A Meta retargeting impression reaches the shopper on Thursday afternoon.
• The shopper searches the brand organically on Friday morning.
• The order arrives Saturday evening.

An MTA system might assign fractional credit to each event. A platform-native report might identify its own touch as the decisive interaction. A separate analytics property might use a different lookback window or identity rule and produce another allocation. These outputs can disagree without any one system having fabricated the purchase.

Does MTA explain the path or prove what caused the sale?

MTA is usually descriptive rather than causal. It tells you how credit was distributed among observed touchpoints under a model's assumptions. It doesn't automatically tell you whether the exposure created demand, captured demand that already existed, or merely appeared alongside another influence.

That distinction matters when a team reallocates budget. If a retargeting impression appears before many purchases, MTA may give it meaningful credit. But those shoppers may already have been close to buying. A holdout test asks a different question: what happened to a comparable group that didn't receive the exposure?

The two methods can coexist. Use MTA to understand path structure and tactical sequencing. Use experiments to challenge the assumption that correlation equals contribution.

What does MTA miss even when the report looks complete?

Even a polished report can miss key moments in the journey. A shopper may see a creator mention on one device, open the product page later from a text message, and convert after searching the brand on a work laptop. The report may capture only one or two of those touches. The missing pieces can change the story about which channel opened demand and which one closed it.

Paid social buyers should especially watch for invisible exposures that still shape performance. These include unclicked creator content, dark social sharing, internal word of mouth, app browsing that never syncs to the web path, and consent-related event loss. The danger is not simply that the report is incomplete. It is that the missing touchpoints are often missing unevenly. Some channels leave abundant evidence. Others leave only fragments.

That asymmetry matters more than abstract model sophistication. A less advanced model built on balanced, reliable inputs can be more decision-useful than a mathematically elegant model trained on patchy observation.

Which attribution model helps with which paid social decision?

Paid social teams usually encounter four model families. They aren't interchangeable, and none can repair a journey that the underlying data failed to capture.

Rules-based models apply a predetermined allocation. Linear attribution splits credit evenly. Time-decay favors touches closer to conversion. First-touch and last-click isolate the entry or closing interaction. Position-based rules put heavier weight on selected milestones. A common U-shaped rule gives 40% to the first touchpoint, 40% to the last touchpoint, and divides the remaining 20% among intermediate touches, as documented in Google's legacy multi-channel funnel attribution documentation. These models are useful as directional baselines and sanity checks, not as proof that the chosen positions caused the sale.

Data-driven attribution estimates credit from observed conversion and non-conversion paths. Google Analytics 4 currently offers data-driven attribution, paid and organic last click, and Google paid channels last click, according to GA4's attribution model documentation. GA4 also states that direct visits don't receive attribution credit unless the path to the key event consists entirely of direct visits. That rule alone can change how branded and returning traffic appears in a report.

Algorithmic MTA uses approaches such as Shapley values, Markov chains, or Bayesian modeling to estimate each touchpoint's contribution within a stitched path. It becomes more useful when a brand has a diversified channel mix, but the output depends on identity resolution, event quality, and the model's treatment of confounding. A 2024 causal-MTA preprint frames the core difficulty as separating genuine causal features from confounding variables between users and conversions, without stripping out the user signal that carries real causal effect.

Unified attribution models attempt to combine signals across platforms into a common view. The attraction is obvious: one decision layer instead of several walled-garden reports. The risk is equally clear. A unified model can create a polished answer from incomplete inputs.

For a Meta-heavy account, platform-native data-driven reporting can help with tactical campaign and audience decisions. For a broader portfolio, algorithmic or unified analysis may help map paths, but only after the team audits coverage. Rules-based reporting should remain the baseline, not the boardroom verdict.

When does a more advanced model actually help?

More advanced models help when the brand has a genuinely mixed channel environment, enough volume to support stable pattern detection, and a decision problem that simpler views cannot answer. If the same shopper often moves among paid social, search, email, affiliates, and direct traffic, interaction effects matter more. A better model may help the team see that one channel sets up another rather than merely repeating it.

But advanced should not mean unchallengeable. The buyer should still ask basic questions. What paths are eligible? Which channels contribute view-through data? How are missing identities handled? Are non-conversion paths included? What happens to direct visits? An advanced model that cannot answer those questions should be treated as a black box aid, not a final judge.

Is the attribution model really the problem?

The industry likes model debates because they feel controllable. You can switch from linear to time-decay, rerun the report, and produce a new allocation before lunch. The bigger distortion often sits upstream, in the events the model never received.

Nobody can tell you what share of the journey your stack observes, and any vendor quoting you a single percentage is guessing. What is documented is the shape of the loss. Third-party cookie restrictions, iOS tracking limits, declined consent, ad blockers and closed-platform silos each remove a category of touch. So do the reporting rules themselves: GA4 gives direct visits no attribution credit at all unless the entire path is direct. None of that is a criticism of an algorithm. It is an input problem, and it is not measurable from inside the report that suffers from it.

If Meta captures a large share of browser and server events while TikTok, podcast exposure, email interaction, or dark social remains only partly visible, the model has more evidence for Meta. It may then assign Meta more assisted and closing credit, not because Meta created all that demand, but because Meta left a more complete record.

Where are the blind spots before you change the math?

Start with the plumbing:

• Consent loss: Identify where users disappear after declining measurement.
• Device breaks: Compare logged-in and anonymous journeys across mobile and desktop.
• Platform silos: List exposures that never enter the shared event layer.
• Deduplication: Check whether browser and server events represent one action or two.
• Offline influence: Record calls, retail activity, partner traffic, and other untracked inputs.

Then compare the model's verdict with controlled evidence. Hosahally and colleagues, writing in the Journal of Digital & Social Media Marketing in March 2025 on measuring digital advertising in a post-cookie era, argue for triangulating MTA with incrementality testing and MMM rather than picking one, because MTA is tactical and user-level while MMM works from aggregate spend and outcome data and can take in offline and brand activity. They score the methods on ease of use, accuracy, validation, robustness and predictiveness, and land on the incrementality randomised control trial as the method to adopt. Worth knowing that the author list includes a measurement vendor's co-founder, so read the incrementality preference with that in mind.

A better model can make a narrow view more precise. It can't make that view complete. Before you change the allocation rule, ask which channel is easiest to measure and which channel might be influential but difficult to observe.

What does a real Meta plus TikTok DTC coverage audit look like?

Consider a DTC brand selling a replenishable household product through Shopify, Meta, and TikTok. The analyst is reviewing one purchase path that appears in the unified reporting layer and wants to verify whether the path can be trusted for creative and budget decisions. The analyst records the audit field by field, using plain language that another team member can read later without interpretation.

• Event name: the analyst records "The browser sent Purchase on the confirmation page, the server sent Purchase from Shopify, and both records use the same event name without a renamed variant."
• Event ID: the analyst records "The browser Purchase event and the server Purchase event both carry event_id ord_581204_purchase, and no second server event with a new ID appears for the same checkout."
• Consent state: the analyst records "Consent status is granted for analytics_storage and ad_storage at the time of the browser event, and the server copy is marked as consented rather than inferred."
• Order ID reconciliation: the analyst records "Attributed purchase ord_581204 matches Shopify order #581204, the gross value matches the checkout value, and the order is not canceled or refunded in the current export."
• Cross-device link: the analyst records "The TikTok landing session occurred on mobile Safari, the final purchase occurred on desktop Chrome, and both sessions connect to customer ID c_88419 through the logged-in email capture during the quiz step."
• Campaign identifier: the analyst records "UTM campaign paid_social_tiktok_prospecting_creator_problemaware_bundle_q2 and TikTok campaign ID 7204 both map to the same prospecting concept in the reporting dictionary."
• Meta touch validation: the analyst records "The later Meta retargeting impression appears in the ad platform export, but there is no duplicate click event claiming a second checkout start for the same user and order."
• Timestamp consistency: the analyst records "All purchase timestamps are stored in UTC, and the browser event, server event, and Shopify order time differ only within the expected processing delay."
• Currency and value: the analyst records "Currency is USD in browser, server, and Shopify records, and the attributed purchase value excludes shipping tax adjustments that post after checkout."
• Identity confidence note: the analyst records "Cross-device stitching is supported by an authenticated email capture, so this path is eligible for tactical analysis and does not rely only on probabilistic identity."

Taken together, those sentences do more than verify a path. They tell the team what kind of trust the path deserves. If most purchase paths cannot be documented this clearly, the buyer should narrow the claims made from attribution output. If many paths can be documented this way, tactical path analysis becomes more defensible.

What are the most common field-level failures?

The most common failures are boring, repetitive, and expensive. Event names drift. Event IDs change between client and server. Consent fields go missing in one layer and get assumed in another. Order values differ because one source includes tax and another does not. Customer IDs exist but are not passed where they are needed. Campaign names carry inconsistent structure, making concept-level reporting unreliable.

Each failure affects a different decision. A naming problem may block creative analysis. A deduplication problem may inflate conversion counts. A missing consent state may make path loss look random when it is actually patterned. A weak cross-device link may make prospecting seem unproductive simply because the opener and closer never join into one visible journey.

The discipline is to diagnose the failure at the field level before discussing channel merit. Many attribution arguments are really implementation arguments wearing a strategic disguise.

What data and identity setup does paid social MTA need?

A defensible MTA setup starts with event definitions, not a dashboard subscription. Your order record, ad platforms, analytics property, and first-party customer data need to agree on what counts as an impression, click, checkout, purchase, and repeat order.

For a DTC team, the practical input layer usually includes:

1. Platform events: Send browser and server events through the relevant platform APIs, including purchase value, event time, campaign identifiers, and a stable event ID.
2. First-party web measurement: Use a first-party analytics layer to preserve page, session, and campaign context within consent and privacy requirements.
3. Order truth: Reconcile modeled ad conversions against the ecommerce order system, including refunds, cancellations, subscriptions, and post-purchase adjustments.
4. Identity stitching: Connect permitted email, phone, login, and customer IDs so a mobile research session and a desktop purchase can be evaluated as one possible path.
5. Conversion deduplication: Ensure a browser purchase and its server counterpart resolve to one conversion event.

A practical guide to paid social can help clarify the channel context, but MTA still depends on implementation details that sit beneath the campaign interface.

Which event definitions need to be locked before you trust MTA?

The team should agree on a minimum event dictionary before anyone debates performance. At a minimum, define what counts as ViewContent, AddToCart, InitiateCheckout, Purchase, SubscriptionStart, RepeatPurchase, Refund, and Cancelation or the closest equivalents in the stack. The wording matters less than consistency.

A strong event dictionary makes three things easier. It keeps browser and server events aligned. It lets analysts compare path stages without wondering whether one source counts a cart view while another counts an actual add-to-cart action. It also protects the creative loop from false diagnoses. If Purchase includes one-click upsells in one report but not another, a buyer may misread which creative role deserves credit.

The table below shows the difference between entries that support reliable analysis and entries that create ambiguity.

What is the minimum viable attribution stack a small team can maintain?

A small growth team doesn't need every possible signal on day one. It needs one agreed event dictionary, one order source, consistent campaign naming, server-side deduplication, and a documented identity policy. Add complexity only when it answers a budget or creative question that the current stack can't answer.

Watch for iOS aggregation delays, ATT-related signal loss, cookie restrictions, mismatched event IDs, and double counting between browser and server. When a conversion count changes suddenly, inspect the event pipeline before declaring a creative or channel failure.

The useful output isn't a perfect user history. It's a labeled, auditable set of observed paths with clear uncertainty. That gives the analyst enough context to form a creative hypothesis without pretending the path is complete.

How much identity stitching is enough for useful paid social decisions?

Not every team needs a perfect identity graph. The practical question is whether the available stitching is strong enough for the decisions being made. If the team wants to compare creative roles inside one platform over short windows, partial identity may be enough. If the team wants to compare prospecting on TikTok against retargeting on Meta across devices, stronger stitching matters much more.

Useful stitching usually begins with approved first-party identifiers that can connect sessions when a shopper logs in, enters an email, starts a subscription flow, or authenticates during checkout. The buyer should know which links are deterministic and which are only inferred. That distinction belongs in the reporting notes, not buried in engineering documentation.

When identity is weak, be careful with claims about upper-funnel underperformance. It may be undercounted rather than ineffective. When identity is strong, the team can make more confident sequencing decisions and better judge whether a closing ad actually finished a path or simply inherited it.

When should you use MTA, incrementality testing, or MMM?

These three methods answer different questions.

MTA asks which observed touchpoints correlated with a conversion. Incrementality testing asks what additional outcome occurred because a treatment group received an exposure while a comparable control group did not. Media mix modeling asks how broad channel spend relates to outcomes at an aggregate level, including activity that user-level tracking may miss.

The methods should not compete for one master number. They should constrain one another.

• Use MTA for tactical path questions: Which creative concepts tend to open journeys? Which messages appear late? Which audience paths contain repeated retargeting?
• Use lift tests for causal budget questions: Does removing or adding a channel change outcomes beyond the conversions already likely to happen?
• Use MMM for portfolio planning: How should channel-level investment shift when reach, offline activity, and broader brand effects matter?

A practical operating rule is to trust MTA for in-platform creative and audience decisions when the relevant pixel and event signals are intact. Trust holdout or geo-lift evidence for reallocating budget across channels. Treat MMM as the planning tiebreaker when a material share of spend sits in channels your event layer barely sees, or when quarterly investment decisions include broad-reach and offline channels. Set that trigger yourself, in writing, before the quarter starts. It is a governance rule, not a universal law. The point is to stop using a narrow user-level view for a decision it can't support.

The dashboard can tell you where the path was recorded. The test tells you whether changing exposure changed the outcome.

For a creative team, that distinction is operational. MTA can reveal that a hook appears early in converting journeys. An incrementality test can challenge whether the hook creates demand or reaches people who were already likely to buy. Both findings belong in the brief, but they shouldn't be written as the same claim.

When should a platform report guide action, and when should it not?

Platform reports are often most useful when the decision lives inside that platform and the relevant event stream is healthy. If a buyer is deciding which creative angle to rotate next in a Meta prospecting campaign, Meta's own report may contain enough tactical truth to act. The same is often true for audience exclusions, frequency pressure, and message sequencing within one ad system.

A platform report should carry less weight when the decision crosses systems or when the path likely depends on touches the platform cannot see. Budget reallocations between Meta and TikTok, prospecting versus email, or paid social versus branded search are exactly where platform confidence should drop. A platform can report what it observed. It cannot referee the parts of the journey outside its own walls.

The practical skill is not rejecting platform data. It is assigning it the right job.

What should MMM settle that MTA cannot?

MMM is most helpful when the decision extends beyond user-level visibility. If the brand is balancing broad social reach, branded search protection, creator programs, retail spillover, or seasonal demand, MTA will likely miss part of the picture even when it is working as designed.

That does not make MMM a replacement for path analysis. It makes it a planning layer. MTA can still tell the creative team where certain messages appear and which role a given ad is serving. MMM can help decide how much investment each channel family deserves when many outcomes are influenced by forces that no user-level path can fully capture.

For the buyer, this means one model for one job is rarely enough. The system works when the methods limit each other's overconfidence.

How do you turn attribution signals into better creative?

Attribution becomes useful when it changes the next ad brief. Otherwise, the team is only cataloging old conversions.

Start by grouping paths around creative roles rather than campaign labels. One cluster may contain concept-led prospecting ads that introduce the problem. Another may contain proof-heavy assets that help shoppers evaluate. A third may contain offer or urgency frames that appear close to purchase. A fourth may contain the ads that bring back a shopper who saw the brand and did not buy. MTA can't prove that each role caused the outcome, but it can show where each role tends to appear in the observed path.

How do you translate path position into the next brief?

Use the observed position to define the job of the next creative:

• Opening-path concept: Write a hook that names the problem, tension, or desired outcome without assuming prior product knowledge.
• Consideration concept: Give the viewer a reason to believe, such as a demonstration, comparison, customer objection, or product mechanism.
• Closing concept: Make the next action clear through offer framing, urgency, reassurance, or a reduction of purchase risk.
• Re-entry concept: Address the reason a shopper might return after seeing the brand but not converting.

The same product can need all four roles. A closer shouldn't be judged by the same standard as an opener, because its job in the path is different. A top-funnel asset may be valuable as a first or assisted touch even when it rarely receives last-click credit.

How should you judge creative with verdicts instead of vanity readings?

Give every creative a role, a hypothesis, and a decision rule before launch. For example:

• Creative hypothesis: This objection-led hook will open more qualified paths among first-time visitors.
• Decision: Pass, iterate, or kill after the agreed review window, using path position, assisted presence, conversion quality, and controlled evidence where available.

If a top-funnel hook carries no useful assisted signal after two review cycles, retire or reposition it. If a closer stops appearing in conversion paths, refresh the offer frame or test a different objection. Those are operating rules, not universal benchmarks. Your team should define the review window and thresholds based on spend, conversion volume, and consideration cycle.

The weekly loop should end with a queue of actions:

1. Analyze attribution clusters.
2. Write one creative hypothesis per cluster.
3. Produce variations that match the role.
4. Test, validate, and feed the verdict back into the next brief.

A structured creative tracking workflow helps keep the attribution fingerprint attached to the ad itself, rather than leaving the insight buried in a channel report.

How do you avoid over-crediting closers and under-learning from openers?

Paid social teams often over-learn from the assets closest to purchase because those assets are easiest to see in reports. Offer ads, urgency frames, retargeting carousels, and proof-heavy testimonials may absorb attention in the dashboard even if they mostly harvest intent already built elsewhere.

A better creative review asks separate questions for openers and closers. Did the opener increase qualified visits, email captures, or early-path presence among converters? Did the consideration concept reduce drop-off between product view and checkout? Did the closer help uncertain shoppers finish? These are different jobs, and the verdicts should reflect that.

When the team uses one narrow success definition for every ad, it tends to produce too many closers and not enough demand creation. Attribution, used carefully, can restore role clarity.

What should you check before trusting an attribution report?

Before trusting an MTA output, run a short pre-flight. Confirm that platform event coverage is proportionate to spend, identity resolution is documented, browser and server events are deduplicated, and the lookback window reflects the actual consideration cycle.

Track these as standing diagnostic measures:

• Assisted conversions by channel: Look for shifts in contribution, not just totals.
• Path-length distribution: A sudden change can indicate identity or event loss.
• Holdout-adjusted CPA: Use controlled evidence when making budget decisions.
• Model confidence: Flag allocations that depend on thin or heavily modeled data.
• Order reconciliation: Compare reported conversion activity with the ecommerce source of truth.

Three failures recur. A sudden credit collapse usually indicates tracking loss before it indicates a model failure. Heavy brand dominance can signal that last-click lookalikes are receiving too much credit. Rising MTA scores alongside flat incremental lift often means the same conversions are being observed through more touchpoints, not that the channel created new demand.

Document every fix, including the event change, affected dates, impacted channels, and the decision made. The next analyst should inherit an explanation, not a mysterious break in the trend.

A useful MTA program ends with a creative verdict. If the report can't tell you which concept to brief, iterate, or retire, it hasn't yet earned a seat in the weekly growth review.

What are the fastest pre-flight checks before a weekly review?

Before the meeting starts, the buyer can run a short checklist that catches many false alarms. Confirm yesterday's purchase count against the order system. Compare browser and server purchase volumes for unusual gaps. Check whether any major campaign launched without standard naming. Scan path length for abrupt compression. Verify that the same time zone is being used in the key exports. Look at refund and cancellation updates if finance is part of the review.

These checks do not guarantee truth, but they sharply reduce the odds that the meeting revolves around a technical break mistaken for a media insight.

How do you troubleshoot sudden shifts in assisted conversions?

When assisted conversions jump or collapse, do not start with creative blame. Start with instrumentation. Did the lookback setting change? Did a server event stop deduplicating? Did a naming update break grouping rules? Did one platform start modeling more aggressively while another tightened consent handling? Did a site change remove a crucial page event?

After the implementation review, examine path composition. If assisted credit moved from one channel to another, ask whether the shift reflects better capture, changed sequence, or changed spend allocation. The answer determines whether the next action is technical repair, budget adjustment, or creative refresh.

When should you stop trusting a platform-reported conversion surge?

Stop trusting it as a strategic signal when it cannot be reconciled to order truth, when it appears only in one platform with no support elsewhere, or when the surge follows a tracking or site change that altered visibility. A surge can still be operationally useful for in-platform learning, but it should not automatically trigger a cross-channel budget move.

The buyer should ask three questions. Did actual orders rise? Did another reporting layer observe any related lift? Did the path evidence change in a believable way? If the answer to those questions is weak, treat the surge as a measurement event until proven otherwise.

How Selzee runs the attribution-to-creative loop

Selzee turns attribution signal into production steps inside one interface. The process starts when the team reviews observed path patterns and tags the creative roles that appear most often around qualified conversions. Instead of leaving the finding in a report, the strategist converts it into a brief with a defined job: opening-path concept, consideration concept, closing concept, or re-entry concept.

From there, Selzee helps shape the test plan. The team can outline the angle to test, the objection to answer, the offer frame to compare, and the audience context the asset is meant to serve. That keeps the next round of creative tied to a measurable role rather than a vague request for more variations.

The next step is creator matching. If the signal suggests that a founder explanation works early while social proof works later, Selzee can help the team choose the right creator profile for each role. A demonstration-led creator may fit a proof asset. A direct-response creator may fit a closer. The point is not just producing more ads. It is assigning the right messenger to the job the path analysis suggests.

Once the ads run, Selzee keeps the verdict connected to the asset. The team can compare tracked ads against the intended role, review whether the concept opened paths, assisted mid-path evaluation, or appeared late near purchase, and then grade the result against CPA and ROAS goals. That verdict feeds the next brief, so the system compounds learning rather than generating disconnected creative tests.

FAQ

What lookback window should a paid social team use?

The right lookback window is the one that best matches the real consideration cycle for the product and the decision you are making. If the window is too short, early touches disappear and closers absorb too much credit. If it is too long, stale touches stay attached to purchases they likely did not influence. Use one default rule for governance, then review whether that rule still fits the path patterns you observe for prospecting, retargeting, and repeat buyers.

How do deduplication problems show up in attribution reports?

Deduplication issues usually appear as inflated purchase counts, unstable assisted conversion trends, or inconsistent revenue between platform reporting and the order system. They often happen when browser and server events both fire for the same action but do not share a stable event ID. A team may think a campaign improved performance when the real change was technical. If conversion volume shifts suddenly, inspect event IDs and event timestamps before changing budget or creative strategy.

How should we handle cross-device paths in paid social reporting?

Start by documenting which links are deterministic and which are not. If a shopper logs in, enters an email, or authenticates during checkout, those moments can help tie sessions together in an approved way. If no such link exists, treat the path as weaker evidence. Cross-device analysis is most useful when the team is explicit about identity confidence. Do not make strong prospecting judgments from paths that depend mostly on uncertain stitching.

What should we do when two dashboards disagree?

Begin with the mechanics, not the opinions. Check the conversion definition, order status logic, lookback window, time zone, identity rule, and whether view-through conversions are included. Then compare event counts before comparing revenue totals. Once the reporting differences are clear, decide which dashboard is fit for the decision at hand. If the choice involves cross-channel budget movement, use the conflict as a reason to lean more heavily on testing or aggregate planning.

What is the safest way to use MTA in creative decisions?

Use it to define roles, generate hypotheses, and prioritize tests rather than to declare absolute winners. If a concept repeatedly appears early in converting paths, treat that as a sign to build more opening variations and validate the pattern over time. If a closer appears late across many successful journeys, keep testing it, but do not assume it created all the demand it captures near purchase. MTA is safest when paired with disciplined briefs and clear verdict rules.

Selzee is your AI content team for turning customer feedback, market research, competitor creative, and ad results into concepts, briefs, scripts, finished variations, and creator shortlists. It helps paid social teams move from an attribution signal to a concrete creative test, then grades tracked ads against your CPA and ROAS targets so the next round starts with evidence. Visit Selzee to see how your team can ship more ads with less guessing.