Discussion
  • SEO Analytics

Ecommerce revenue attribution for SEO: last non-direct click, data-driven, or modelled? Which does your finance team accept?

Published 1 reply

Discussion starter: an SEO measurement question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    Report SEO revenue to finance on a last non-direct click basis, scaled to booked revenue, with branded organic on its own line. Show data-driven attribution beside it as a sensitivity check rather than the headline. Keep modelled estimates (SEO split tests, holdout tests, a marketing mix model) for investment decisions, where finance needs to know what the channel adds, not how credit was shared out.

    The qualification is that none of the three is “the accurate one”. Last non-direct click and data-driven attribution both divide up revenue that your tracking observed. Neither says how much revenue would disappear if organic search did. What finance needs from any channel figure is more basic: it should reconcile to the ledger, come out the same when someone re-runs it, and not double count with other channels. A fixed rule passes those tests most easily.

    On an ecommerce site, two adjustments deserve attention before the model debate: how much organic revenue is branded demand that other channels created, and how far GA4 revenue sits from booked revenue. Either can move SEO’s number further than switching models does.

    Three approaches answer three different questions

    Think of commission in a showroom. Last non-direct click pays the whole commission to the last salesperson who spoke to the customer and ignores the cashier, because a customer who walks straight to the till (a direct visit) was sold by someone earlier. Data-driven attribution splits the commission according to how often each salesperson shows up in sales that closed compared with sales that didn’t. Incrementality asks a different question: how many sales would the showroom lose if that salesperson went home?

    In GA4 terms, the “Paid and organic last click” model ignores direct traffic and gives 100% of the key event value to the last channel the customer clicked through before converting, falling back to Direct only when the whole path is direct. That is GA4’s version of last non-direct click. The data-driven model evaluates both converting and non-converting paths, and Google describes it as a counterfactual approach that compares what happened with what could have happened. The word “counterfactual” is easy to over-read. The model still divides credit for conversions GA4 recorded among touchpoints GA4 recorded. It can’t tell you what total revenue would be without organic search, because it doesn’t model where those shoppers would have gone instead.

    Why one GA4 property shows three organic revenue figures

    Check which number each person is looking at before debating models. GA4 uses the paid and organic last click model for user-scoped and session-scoped dimensions, and the model you select, data-driven by default, for event-scoped dimensions. The attribution setting only changes the second group.

    Where the figure appears How organic gets credit Follows the attribution setting?
    Traffic acquisition report (session-scoped dimensions such as Session default channel group) Paid and organic last click, per session No
    User acquisition report (first-user dimensions) The channel that first acquired the user No
    Attribution reports, and reports or explorations using event-scoped dimensions such as Default channel group The selected reporting model, data-driven by default Yes, including for past data
    Your ecommerce platform or ERP No channel credit: orders as booked, including those GA4 never recorded Not applicable

    So the SEO team’s Traffic acquisition screenshot and the paid media team’s attribution report can disagree about organic revenue on the same day without either being wrong. Finance sees two figures from one tool and has reason to trust neither.

    Watch for this:

    GA4 applies a change of reporting attribution model to historical data as well as future data, while lookback window changes only apply going forward. Switching models can change organic revenue for quarters that finance has already reported. Snapshot the figures finance receives each month and log every settings change with its date.

    The tests finance applies to any channel figure

    Whichever model produced it, expect a channel figure to be tested against these questions:

    • Does it reconcile? Channel figures should add up to booked revenue for the period on the P&L’s definition: net of refunds and cancellations, with the same treatment of GST or sales tax, shipping and discounts.
    • Can it be reproduced? Someone re-running the report next quarter should get the same figure for the same period. A fixed rule passes. A model-driven figure, or a report restated by a settings change, may not.
    • Is it de-duplicated? Adding up the revenue each ad platform reports for itself can exceed actual revenue, because each platform credits itself for shared orders. One model across all channels, in one tool, avoids that.
    • Does it separate capture from creation? A branded search can be the last step for demand that email, paid social, TV or word of mouth created.
    • Is it in margin terms where margin varies? Organic revenue concentrated in clearance lines is worth less than the same revenue in full-price ranges.

    Last non-direct click passes the first three most easily, which is why it makes the better reporting default. Data-driven attribution is harder to defend on the second: finance can’t rebuild it in a spreadsheet, and its output can shift as the underlying data changes. Neither model handles the last two, so those adjustments are yours to make.

    Reconcile GA4 to booked revenue before splitting by channel

    GA4 revenue and booked revenue will differ. Consent choices, ad blockers, purchase tags that fail, refunds processed after the order, and phone or in-store orders all widen the gap. The finance-friendly approach takes channel shares from GA4 and totals from the ledger:

    1. Match orders. With the BigQuery export, join GA4 transaction IDs to platform order IDs and calculate the match rate by device, country and channel.
    2. Apply shares to booked revenue. Multiply booked net revenue by each channel’s share of GA4-tracked revenue.
    3. State the assumption. This assumes untracked orders share the channel mix of tracked ones. If match rates differ sharply between segments, for example between browsers or between markets where more visitors decline analytics cookies, scale each segment separately.

    Some people mean GA4’s consent-mode modelling when they say “modelled”. That estimates behaviour for visitors who decline analytics cookies. It changes how complete the data is, not the rule that assigns credit between channels, so it doesn’t replace any of the three approaches in the question.

    Branded organic is the line finance will question first

    A shopper sees a paid social ad, searches the brand name the next day, clicks the organic result and buys. Last non-direct click gives organic search the whole order, and data-driven attribution will give it a share. Report that revenue, because SEO protects it: the brand result still has to rank, and a reseller or marketplace listing can take the click. Just don’t present it as revenue SEO created.

    GA4 can’t join search queries to orders, so the split needs a proxy:

    • Landing-page classification. Organic sessions landing on the homepage and brand pages go on one side; category, product and editorial landings on the other. It’s crude, since some brand searches land on product pages, but it’s reproducible.
    • Search Console’s branded queries filter, introduced in November 2025, to check the branded share of organic clicks against the landing-page proxy. Treat it as calibration, not a revenue figure.

    Agree channel boundaries at the same time. Decide whether GA4’s Organic Shopping channel counts as SEO, and what to do with the AI Assistant default channel group GA4 added in May 2026 for referrals from assistants such as ChatGPT, Gemini and Copilot. Traffic without a referrer still lands in Direct, and practitioners report that clicks from Google’s AI Mode arrive as google / organic, so AI Mode revenue is likely already sitting inside Organic Search.

    Hypothetical example:

    A homewares retailer books $4.0m net revenue in a quarter, after refunds and excluding GST. GA4 records $3.1m of it. Organic Search holds 30% of GA4-tracked revenue in the Traffic acquisition report, so the scaled organic figure is $1.2m. Landing-page classification puts 45% of organic revenue on the homepage and brand pages: $540k brand, $660k non-brand. In GA4’s Model comparison report, organic’s share drops from 30% to 27% under data-driven attribution, which would scale to $1.08m.

    The finance report reads: non-brand organic $660k (last non-direct click, scaled to booked revenue); brand organic $540k (reported and protected, not claimed as SEO-created); data-driven view $120k lower in total. In this example, the brand split moves SEO’s figure four and a half times as far as the model choice. Whether that holds on your site is worth checking before anyone argues about models.

    Where modelled numbers belong

    Modelled approaches estimate what organic search adds, which is the question behind any budget request. They are slower and less precise than attribution, so use them for decisions rather than month-end reporting.

    Approach What it estimates Limits for SEO
    SEO split test: change a template on part of a group of similar pages, keep the rest as a control Incremental organic clicks from one change, converted to revenue using those pages’ revenue per click Needs many similar pages and enough traffic; measures a change, not the whole channel
    Paid brand holdout: pause brand search ads in matched regions or weeks How much brand revenue the organic listing catches without ads Measures the overlap between paid and organic brand clicks, not SEO’s total value; costs revenue if organic doesn’t catch the clicks
    Marketing mix model Each channel’s contribution, estimated from weekly aggregate data Needs a long, stable history; organic clicks move with overall demand and have no spend lever to vary, so expect wide ranges for organic
    Forecast versus actual after SEO work Lift against a baseline forecast for the affected pages Weakest control; seasonality, promotions and core updates all confound it

    If the business already runs a mix model for paid media, ask the modelling team to add non-brand organic clicks from Search Console as a variable instead of commissioning a separate SEO model. Finance already trusts that model’s conventions, and an SEO-only model looks like a channel marking its own homework.

    Agree the rules before the first report

    1. Revenue definition: net of refunds and cancellations, with the P&L’s treatment of tax, shipping and discounts.
    2. One rule for every channel: the reported figure uses the same last non-direct click rule for paid and organic alike, with data-driven figures as a labelled sensitivity line.
    3. Lookback window: if you use event-scoped reports, record it. GA4’s default is 90 days for key events other than the acquisition events first_visit and first_open (which default to 30 days), with 30 or 60 days as the alternatives.
    4. Channel boundaries: Organic Search, Organic Shopping, AI Assistant and Direct, written down.
    5. Brand split method: the landing-page rules, and when they are reviewed.
    6. Snapshots and change log: figures frozen when issued, and every GA4 settings change dated.
    7. Incrementality evidence for budget decisions: named in advance and reported as a range, not a point.

    If I were taking this to a CFO, I would bring one reported figure and one range. The figure is non-brand organic revenue on last non-direct click, scaled to booked revenue, with branded organic beside it as a protected line. The range comes from whatever incrementality evidence exists: split tests on the templates SEO changed, or organic’s estimate in an existing mix model. Finance can audit the first and debate the second, which is the right way round.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.