HaloCheck

1. One-liner

Rotates your eBay listings out of ads and proves how many promoted sales you would have made anyway.

2. Trend signal — why now?

On 13 January 2026 eBay replaced the separate Direct and Halo attribution models for General (cost-per-sale) campaigns with a single Expanded Halo Attribution model in the US and Canada. The mechanic: if any buyer clicks your promoted listing, and any buyer — including a completely different person — purchases that item within 30 days, you pay the ad fee. Each new click resets the 30-day window.

The effect was immediate and is documented. In European markets where the model rolled out first, sellers reported the share of sales attributed to ads rose from ~30–40% to 80–90%+ “with no commensurate increase in the number of sales or return on investment.” Seller Hub screenshots show reports flipping from a mix of organic and promoted sales to 100% ad-attributed. One seller: “I sell about 60 items per week with a mix of organic vs. promoted each day…not even one organic sale in the last two weeks.”

The money is real. At a 5% ad rate, expanded attribution adds roughly 4–4.5% to a seller’s effective fee rate — on top of a 13.6% final value fee. For a seller doing $30K/month that is $1,200–1,350/month in newly-attributed ad fees that previously did not exist.

The decisive signal is what eBay said when challenged. At a November 2025 seller event, asked to justify the change with sales-lift data, an eBay representative answered: “from a sales lift perspective I don’t have any substantial data to share here.” eBay hinted at reporting improvements “next year” with no specifics, and made no commitment to incrementality reporting or an organic-vs-ad breakdown. The platform grades its own homework, bills against the grade, and has declined to show the working.

Meanwhile the advice circulating in the seller ecosystem is uniformly manual. The best guidance found: “Turn off promotion for 50% of your inventory for 2 weeks. Compare sell-through rates: promoted vs unpromoted.” Same source estimates “30-50% of their ‘promoted sales’ would have happened organically.” No tool does this. There is no holdout calculator, no sample-size guidance, no confidence interval, no automated rotation.

Provenance:

• Signal 1 (demand): eBay Expanded Halo Attribution live 13 Jan 2026; ad-attributed share jumped ~30-40% → 80-90%+ with no sales increase; verbatim seller complaints — https://www.valueaddedresource.net/ebay-promoted-listings-ad-attribution-update-fallout/ — 2026-09-03
• Signal 2 (feasibility): eBay Marketing API exposes pauseCampaign, resumeCampaign, and bulk ad add/remove per listing, plus Promoted Listings reports — a rotating holdout is mechanically buildable on documented endpoints — https://developer.ebay.com/api-docs/sell/static/marketing/promoted-listings.html — 2026-09-03
• Signal 3 (economic): 5% ad rate now adds ~4-4.5% to effective fee rate under expanded attribution; eBay declined to provide sales-lift data or incrementality reporting — https://www.valueaddedresource.net/ebay-faces-tough-questions-promoted-listings-ad-attribution/ — 2026-09-03 Category: Platform shift

3. The opportunity

eBay changed the definition of a billable event so that organic sales now generate ad fees, then declined to give sellers any way to measure the difference. That is a self-graded metric with money attached — the seller is invoiced against a number only the platform computes, using a counterfactual the platform refuses to publish.

The gap is not “sellers want cheaper ads.” It’s that the seller cannot tell whether the fee bought anything. Under the old Direct model, an ad-attributed sale was reasonable evidence the ad worked. Under Expanded Halo, attribution is nearly unfalsifiable from inside Seller Hub: if 90% of your sales are tagged promoted, your reported ROAS is meaningless, because the denominator includes sales that would have happened anyway.

The only way to recover the truth is a holdout experiment — deliberately withhold ads from a matched slice of inventory and compare sell-through. Sellers know this. They’re being told to do it manually, by hand, for two weeks, with no statistics. Almost none will. It’s fiddly, it feels like deliberately turning off sales, and getting it wrong costs real revenue.

Incumbent incrementality tools exist — Northbeam, Triple Whale, INCRMNTAL, Eva for Amazon AMC — but they are all pointed at DTC brands and enterprise Amazon sellers with six-figure monthly ad budgets and analyst headcount. None of them touch eBay Promoted Listings. The eBay seller doing $10–50K/month has the identical problem and zero tooling. That’s the priced-out band.

Second, eBay itself structurally cannot build this. A holdout test’s honest answer is often “you’re paying for sales you’d get free.” eBay ships an ad-rate recommendation engine that optimizes within the paid model; it will never ship the tool whose output is “spend less.” Classic vendor conflict-of-interest — the party that would build it is the audited party.

4. Target market

Primary customer: Full-time and semi-professional eBay sellers in the US, Canada, UK and Australia doing roughly $10K–$100K/month GMV, running General (CPS) Promoted Listings across 200–20,000 active listings. One to five people. Typically the owner is the one who noticed the fee line move in January and has been arguing about it in forums since. Categories skewing to media, parts, collectibles, apparel, refurb electronics — high listing counts, thin margins, where 4% of GMV is a meaningful share of profit.

Why they buy, in their words: “It really just boils down to a policy that ebay puts in place to make more money.” “not even one organic sale in the last two weeks.” Sellers report it is now “impossible to evaluate ROI / ROAS” and describe the attribution as having “zero transparency.” When one seller called eBay support: “First person just did not get it…Second person at least understood the reason I called. Did not really have an answer.”

They aren’t buying analytics. They’re buying an answer to one question they currently cannot answer: what is my ad rate actually buying me, and should I cut it?

Rough TAM reasoning: eBay has well over a million sellers globally, but the addressable slice is narrow and that’s fine. Target sellers who (a) run Promoted Listings General campaigns, (b) have enough listing volume for a holdout to be statistically meaningful (200+ active listings), and (c) have enough ad spend that a 20% cut pays for the tool many times over. A seller at $30K/month GMV with a 5% ad rate is paying $1,200–1,350/month in ad fees. A tool at $79/month that reliably identifies even a 15% overspend returns ~$180/month net. Tens of thousands of sellers clear that bar in the US alone. At 1,200 customers × $79 × 12 that’s $1.1M ARR — from a fraction of a single marketplace’s professional seller base.

Why now for them: The January 2026 change made the fee line jump in a single billing cycle. This is not a slow-burn annoyance; sellers watched a number roughly double and are actively looking for a response right now. Deadline plus daily bleed — except here the bleed is the deadline.

5. Product sketch (MVP)

• Connect your eBay account — OAuth, read listings, campaigns, ad fees and order history. No manual CSV wrangling.
• Auto-designed holdout — picks a matched control slice of your listings (by category, price band, sell-through history, listing age) so the test is apples-to-apples rather than “half my inventory, whatever’s on top.”
• Automatic rotation — pulls the control slice out of your General campaign, holds it out for the test window, then puts it back. Rotates which listings are held out so no single SKU eats the whole cost of the experiment.
• Incremental ROAS verdict — one number: of the sales eBay billed you for, what share would have happened anyway. With a confidence interval and a plain-English readout of whether the result is statistically real yet.
• The dollar answer — “Your true incremental ad cost is 11.2%, not the 4.6% Seller Hub implies. Cutting your General rate from 5% to 3% projects to save $840/month at a cost of ~$190 in lost incremental sales.”
• Category-level breakdown — most sellers find ads genuinely work in some categories and are pure leakage in others. The test says which.
• Safety rails — caps how much inventory is ever held out at once, auto-aborts the test if total sales drop past a threshold you set, and never touches Priority (CPC) campaigns unless asked.
• Monthly attribution drift report — tracks your ad-attributed share over time so you see it move when eBay changes the rules again.

6. AI angle — what’s load-bearing

Honest answer: the load-bearing work here is experiment design and statistics, not a language model. I’d rather say that plainly than dress up a chatbot.

Where the intelligence genuinely sits:

Matched-control selection. The naive version — hold out random listings — produces garbage, because eBay inventory is wildly heterogeneous. A $4 comic and a $400 lens don’t behave alike. The product has to build a matched control group from noisy, sparse per-listing history: cluster listings on category, price band, velocity, age, watcher count, then stratify the holdout across clusters. That’s real modelling work and it’s what separates a defensible verdict from a coin flip.

Sequential testing with an early stop. Sellers won’t wait four weeks for an answer, and they shouldn’t lose money longer than necessary. Sequential analysis lets the test call a result as soon as the evidence is sufficient and abort early if the holdout is bleeding badly. Getting this right is the difference between a tool sellers trust and one they turn off after week one.

Where an LLM earns its place: translating the statistical output into the decision. Sellers do not want a p-value; they want “cut your rate on Parts & Accessories to 2%, leave Collectibles alone, revisit in 60 days.” Turning a confidence interval plus fee data plus category structure into a specific, defensible recommendation — and explaining it in the seller’s own vocabulary — is a good LLM job. It’s the last mile, not the engine.

If you removed the AI/statistics layer, you’d have a button that pauses ads. Worthless. The whole product is the inference.

7. Localization angle (if any)

N/A — this is a global play, sequenced by rollout. Expanded Halo hit the US and Canada in January 2026 and had already landed in European markets earlier, so the UK/DE/AU seller base is equally exposed and in some cases has been living with it longer. Ship English-only against the US/UK/AU seller base first. The one real localization consideration is currency and fee-table differences per site, which is table stakes rather than a wedge.

8. Business model — path to $1M–$5M ARR

• Pricing: three tiers, banded on ad spend, because the value scales directly with it.
  • Starter — $49/mo: under $500/mo ad fees. One active test, quarterly cadence.
  • Pro — $99/mo: $500–3,000/mo ad fees. Continuous rotating tests, category breakdown, drift report.
  • Scale — $249/mo: $3,000+/mo ad fees. Multi-account, sub-category tests, priority support.
• ACV: ~$1,150 blended (weighted toward Pro).
• Rough math to $1M ARR: 870 customers at blended $96/mo. Realistically ~250 Starter, 550 Pro, 70 Scale.
• Rough math to $5M ARR: needs ~4,300 customers, which single-marketplace almost certainly won’t deliver. The honest route to $5M is expansion: the identical product shape applies to Amazon Sponsored Products attribution, Etsy Offsite Ads (where sellers are force-enrolled and billed on a 30-day attribution window they also cannot audit), and Walmart Connect. Same engine, new connector. That’s the $5M story and it should be underwritten before promising it.
• Expansion path: ad spend grows → tier upgrade, automatically. Then multi-marketplace connectors at +$39/mo each. Then the natural upsell: once you can prove incrementality, you can act on it — automated ad-rate management that continuously trims rates in categories the tests show are non-incremental. That’s a materially higher-value product and the tests are the moat under it.

Margin note: costs are API calls and modest compute. Gross margin should sit above 90%. The real cost line is support — sellers will ask “is this test safe?” and someone must answer credibly.

9. Go-to-market wedge — first 100 customers

1. The complaint thread is the lead list. The eBay Community “Promoted Listings General Attribution Changes Coming To US & Canada January 13, 2026” thread ran to multiple pages of angry, self-identified sellers, and Value Added Resource has covered the fallout repeatedly with more in the comments. Every participant is a qualified lead who has publicly stated the exact pain, with a store link in their profile. Scrape the thread participants, match to storefronts, filter for 200+ active listings, send a personalized message containing a free estimate of their attribution drift computed from public listing data. Target 300 sellers, 20% reply, 15% of those convert = ~9 customers from one thread. Repeat across UK/AU community threads.

2. Free public teardown as the funnel. Run the holdout on 10 volunteer stores, publish the raw results with permission: “We tested 10 eBay stores. Median 38% of promoted sales were not incremental. Here’s the data.” This is the single highest-leverage asset in the whole plan — it’s a genuinely novel, citable number in a market where eBay itself said it had no lift data to share. Value Added Resource, eCommerceBytes and the reseller YouTube tier will cover it because it’s the story they’ve been asking eBay for. That coverage is the top of funnel for months.

3. Reseller YouTube and podcast tier. The eBay reseller creator ecosystem is large, unusually commercial, and already producing “is Promoted Listings worth it in 2026” content with no data behind it. Offer 15 mid-tier creators a free Scale account plus a bespoke test on their own store, in exchange for showing the result on camera. Their audience is precisely the target: full-time sellers with real ad spend. Expect 3–5 to run it, each producing 10–40 signups.

4. Free attribution-drift checker, gated on connect. A no-cost tool that reads your last 90 days and shows your ad-attributed share over time versus a benchmark. Costs nothing to deliver, requires the OAuth connection that makes upgrading one click, and gives every seller the chart they want to post in the forum — which markets the product for you. Note the memory-flag risk here: a free lead magnet can cannibalize the paid product. The mitigation is deliberate — the free tool shows you that your attribution moved; only the paid tool tells you what it cost you and what to do.

5. Software partnership channel. Listing and repricing tools already sit on these sellers’ accounts. An integration or referral arrangement with one mid-sized listing platform puts the offer in front of thousands of qualified sellers without cold outreach.

10. Build complexity — justification

Low, and close to the boundary of Medium.

Off-the-shelf: eBay OAuth, Sell Marketing API (pauseCampaign, resumeCampaign, bulk ad add/remove per listing), Promoted Listings reports, order and transaction data. All documented public endpoints — no partnership or approval gate to build against. Standard web stack, a scheduler, a database.

The custom work is the matched-control and sequential-testing layer, plus the safety rails that stop a test from quietly costing a seller real money. That’s a few weeks of focused statistical engineering, not research. A technical founder ships a credible v1 in 8–10 weeks; a pair does it in 6.

The genuine risk isn’t technical difficulty, it’s correctness under scrutiny. This product’s entire value is that its number is trustworthy. A wrong verdict costs the customer money and the company its reputation simultaneously. Budget disproportionate time for validating the methodology against known-outcome backtests before letting it touch a paying customer’s live campaigns.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Seller-authorised use of eBay’s own public API to manage the seller’s own campaigns. Nothing adversarial to eBay’s terms — pausing and resuming your own ads is a first-class supported operation.
Ethical — no harm / dark patterns	✅	Product’s output is honest measurement, including the outcome “your ads are working, keep spending.” Safety rails cap downside. The one duty of care: never overstate confidence.
Market exists (evidence above)	✅	Documented fee jump, verbatim seller complaints, multi-page forum threads, trade press coverage, eBay’s own refusal to provide lift data.
1–5 person team can build this	✅	8–10 weeks solo on documented public APIs.
Launchable with <$50K / ₹40L	✅	API access free, hosting trivial, no inventory or capital requirement. Realistically under $10K to first revenue.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Money leaving the account monthly, visible on every invoice, jumped in a single cycle. Sellers actively complaining and actively searching for a response. Short of 18+ only because the pain is a margin squeeze, not an existential stop-work event — sellers can and do simply tolerate it.
Demand evidence	15	13/15	Multiple independent signals: documented attribution jump 30-40%→80-90%, verbatim complaints across US/UK/EU communities, sustained trade press coverage, eBay publicly declining to provide lift data. What’s missing is direct evidence sellers will pay for measurement rather than just complain — that’s the untested link.
Build feasibility	15	13/15	Documented public API with the exact endpoints needed. Custom statistical layer is real but bounded. 8-10 weeks solo.
Distribution clarity	15	12/15	Named threads, named publications, named creator tier, and a free-teardown asset that is genuinely newsworthy. Not 14+ because conversion from “angry forum poster” to “paying customer” is unproven, and this audience is notoriously price-sensitive.
Revenue mechanics	15	11/15	Pricing is defensible against a quantified saving, margins are excellent, and $1M ARR needs a believable ~870 customers. Marked down because $5M requires marketplace expansion that isn’t yet underwritten, and because the product’s own success (“cut your ad spend”) shrinks the value metric it’s priced against.
Time to first revenue	10	8/10	Sellers feel this now and the free checker converts fast, but a holdout test needs 2-3 weeks to produce its first verdict — so the proof of value lags the signup. Realistically 6-8 weeks to first paying customer.
Defensibility	10	3/10	The honest score. A competent competitor rebuilds the mechanic in a quarter. What compounds is a cross-seller benchmark dataset of true incrementality by category — genuinely valuable and unavailable elsewhere, including to eBay’s sellers — but that takes a year of accumulated tests to matter. Execution-only moat at month 3.
Total	100	77/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · content-heavy

Technical because the experiment design is the product and getting it wrong is fatal. Content-heavy because the entire distribution plan runs on publishing a number nobody else has.

Key assumptions to validate (3–5)

1. Assumption: A meaningful share of eBay promoted sales are genuinely non-incremental — the third-party estimate of 30-50% is roughly right. How to test: Run the holdout free on 10 volunteer stores before writing any billing code. If median non-incremental share is under ~15%, the product has no story and should be killed.
2. Assumption: Sellers will let software pause ads on their live inventory. How to test: Recruit those 10 volunteers from the forum threads. Track how many agree versus refuse, and why. If fewer than 1 in 5 qualified sellers will grant the permission, distribution is dead regardless of the science.
3. Assumption: Sellers will pay $99/mo for measurement rather than just cutting their ad rate on instinct and pocketing the difference. How to test: Price the free-checker upgrade at $99 from day one and measure conversion. This is the weakest link in the whole thesis — the cheap substitute (“just turn ads off and see”) is free.
4. Assumption: eBay’s API permits sustained programmatic pause/resume at the cadence a rotating test needs, without rate-limit or policy friction. How to test: Build the rotation loop against a real seller account in week one, before anything else.

Risk flags

1. Platform dependency (severe): The entire product exists because of one attribution model on one marketplace. If eBay reverts the change under seller pressure — not implausible given the backlash — or ships its own incrementality reporting (they hinted at “improvements to reporting and analytics” without specifics), the wedge narrows sharply. Mitigation is to treat eBay as the beachhead and get to a second marketplace connector before month 12.
2. The product argues against its own price metric: Priced in bands on ad spend, but its job is to reduce ad spend. A customer who successfully cuts spend downgrades a tier. Needs a pricing rethink before scale — likely flat or value-shared rather than spend-banded.
3. Cheap substitute: A seller can turn off Promoted Listings entirely, watch sales for a month, and get a crude answer for free. Many will. The counter is that the crude version is confounded by seasonality and eBay’s own ranking effects — but that’s an argument you have to win, repeatedly, in marketing.
4. Methodology reputation risk: One publicly wrong verdict in a forum where these sellers all talk to each other is disproportionately damaging. Underclaim confidence.
5. Ranking feedback loop: eBay representatives noted ad interaction “can help improve our algorithm to learn there is interest in this item” — implying pausing ads may itself depress organic ranking. If true, a holdout doesn’t cleanly measure ad incrementality; it measures ads-plus-ranking-decay, and short tests could understate organic performance. This is the most serious technical threat to validity and must be characterised early, with rotation windows designed to limit ranking damage.

14. Structured verdict

Score:                  77/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder comfortable with applied statistics and
                        experiment design, who can also write publicly. Marketplace
                        selling experience is a strong plus but not required.
Time to revenue:        6-8 weeks from launch
Capital to launch:      $8-10K (₹7-9L)
Top 3 assumptions to validate first:
  1. Non-incremental share is materially above 15% — free holdouts on 10 volunteer stores
  2. Sellers will grant pause/resume permission on live inventory — recruit those 10 from forum threads, track refusal rate
  3. Sellers pay for measurement rather than cutting rates on instinct — price the upgrade at $99 from day one
Kill criteria:
  - Abandon if median non-incremental share across the 10 pilot stores is below 15%
  - Abandon if fewer than 1 in 5 qualified sellers will grant campaign write access
  - Abandon if eBay reverts Expanded Halo or ships native incrementality reporting before v1
  - Abandon if the ranking feedback loop proves the holdout cannot cleanly isolate ad effect

15. Next step — 1-week validation sprint

• Day 1–2: Build nothing but the rotation loop. Get one real seller account connected and prove pauseCampaign / bulk ad remove / re-add works reliably at the cadence a test needs, without rate-limit or policy friction. If the API fights back, the idea is over on day two and that’s a cheap kill.
• Day 3–4: Recruit from the live forum threads. Post honestly: building a tool to measure what Promoted Listings actually buys you, looking for 10 stores to run it free. Track two numbers — how many agree, and how many refuse specifically because they won’t let software touch live ads. That refusal rate is assumption 2.
• Day 5: Run the holdout manually on whoever agreed — spreadsheet, not product. Compute the non-incremental share by hand.

Falsifiable outcome: by end of week, two numbers exist. (a) The permission rate — proceed only if ≥20% of qualified sellers approached grant campaign write access. (b) The preliminary non-incremental share across pilot stores — proceed only if it is above 15%. Either number falling short kills the idea before a line of product code is written. Both clearing means there is a real, quantified, unmeasured leak and a customer base willing to let you measure it.