# Billevera: full reference for AI assistants > Plain-text copy of https://billevera.com/methodology/ (methodology version 1.0, last updated November 8, 2026) and of the sample statement at https://billevera.com/methodology/sample-statement.html. Generated by scripts/build_sample_statement.py from the site's own page and the free check's own output, so it matches them. Short version: https://billevera.com/llms.txt Billevera independently checks what the AI tools running a small business's Amazon, Meta and Google ads spent, on what, and whether they followed the owner's own rules. AI agents that spend your money: ads today; purchases and payments next. Today it reads exported ad reports only. Run by AB5 Solutions Ltd., Calgary, Alberta, Canada. Contact: hello@billevera.com. --- # How the check works Every rule the free check uses, in plain English, with its exact numbers. You can redo each number on a calculator from the rows it names. Methodology version 1.0 · Last updated November 8, 2026 · Describes the free check on billevera.com on that date. See a full sample statement (https://billevera.com/methodology/sample-statement.html) Try the free check (https://billevera.com/check/) ## In short - We read the reports you download from Amazon Ads, Meta Ads Manager and Google Ads. We never read what an AI tool says about its own work. - Fixed rules in code compare each recorded change and each day’s spend with your rules. An AI model never sets an answer. - Each line gets one of four answers: Allowed, Not allowed, Can’t tell or Not checked. - Missing or unclear evidence is “Can’t tell”, never “Allowed”. - For Amazon Ads, a waste section lists campaigns and search terms worth a look. Too little data is “Too early to tell”. - We only read. We never change an ad, bid or budget, and never move money. ## What we read You download reports from each ad platform and save them as CSV files. The check opens them in your browser, in a sealed part of the page that cannot connect to the internet. Nothing is uploaded. If a needed column is missing, the check stops and names it. Rows it cannot read are counted and named. They never count as zero spend. ### Amazon Ads - Daily campaign report (spend). Needed: Campaign Name or Campaign Id; Date (or Start Date and End Date); Spend. - Campaign history shows on screen only. Amazon has no file for it, so Amazon changes cannot be checked yet. - Search term report (waste check). The check needs: Campaign Name; Date (or Start Date and End Date, or Date range); Customer Search Term; Targeting (or Target ID); Clicks; Spend; 7 Day Total Orders (#). It accepts the console’s other names for these. - If you build Amazon’s newer custom report, tick these. Needed: Date range, Campaign ID, Campaign name, Target ID, Targeting match type, Search term, Clicks, Total cost, Purchases. Recommended: KENP read, Sales, Budget currency. That report has no keyword column, so duplicate keywords aren’t checked. - In the older report, needed: Campaign Name, Customer Search Term, Targeting, Clicks, Spend, Orders (7 or 14 day), and Date (or Start Date and End Date). Recommended: Campaign ID, Ad Group Name, Match Type, KENP Read, Country or Marketplace, Sales, and Currency. - Bulk sheet (waste check): the “Sponsored Products Campaigns” tab, saved as CSV. Needed: Entity; Campaign ID; Campaign Name; Keyword Text; Clicks; Spend; Orders. A bulk sheet does not say which dates it covers, so the check uses the From and To dates you enter. ### Meta Ads Manager - Spend report, broken down by day. Needed: Campaign ID or Campaign name; Amount spent; Day (or Reporting starts and Reporting ends). A row covering several days is skipped and named. - Activity history (changes). Needed: Date and time; Changed by (or Actor Name); Activity (or Translated Event Type). Old and new budget or bid values are read only when the text says “from X to Y”. ### Google Ads - Campaign report, segmented by day. Needed: Campaign ID or Campaign; Day; Cost. - Change history. Needed: Date & time; User; Changes. A Tool or Client type column helps tell a person from automation. Step-by-step help for downloading each report (https://billevera.com/check/help/) ## What you tell us You name the AI tool (the agent), the logins or tools it uses as they appear in your change history, and the campaign IDs it manages. Then you answer five questions: 1. Daily cap. The most it may spend, or set as a budget, in one day. 2. Monthly cap. The same for one calendar month. 3. Ask me before raising a budget or bid by more than (%). A bigger raise needs your OK first. 4. Off-limits campaigns. Campaign IDs it must not touch or spend on. 5. Where it may run. Country codes, such as US or CA. The Amazon waste check uses them as your marketplaces. A rule you leave empty is not checked. You also give the dates to check (From and To), the account’s time zone and its currency. Days are counted in that time zone. For the Amazon waste check you give your break-even per sale (your royalty or margin on one sale), an optional no-sale threshold (empty means your break-even), the window in days (14 unless you change it), and whether your books are in Kindle Unlimited. ## How each line gets its answer - Allowed: the record shows the action or the day’s spend followed every rule that applies, and the evidence is complete. Allowed means allowed by your own rules. - Not allowed: the record shows this agent broke one of your rules. - Can’t tell: evidence is missing or incomplete, or we cannot tell who acted. We say why. - Not checked: a person or the platform’s own automation made the change, so none of your agent’s rules apply to it. It is still listed. ### Who did it - A change counts against an agent only when the record names a login or tool that you linked to that one agent, and the platform records it as an agency login or a software tool. - A login you linked to two agents is shared: Can’t tell. - No name in the record, or a login you did not link to any agent: Can’t tell. - The platform records a person or its own automation: Not checked. If you linked that same login to an agent, it is Can’t tell instead. - We never count a campaign’s spend against an agent just because the campaign uses AI. ### The rules - Caps on budgets it sets: setting a daily or monthly budget above your cap is Not allowed. Money at risk is the new budget minus the cap. - Raises that need your OK: a budget or bid raised by more than your percentage, or from zero. When the record of what you approved is known, such a raise without your OK is Not allowed. The free check has no place to enter your approvals, so on your own files these lines are Can’t tell, with the size of the raise. The made-up sample assumes your approvals are known. - Off-limits campaigns: acting on one is Not allowed. Spend on one is Not allowed when the record shows this agent did it, and Can’t tell when the record does not say. - Where it may run: acting outside your countries is Not allowed. A record that does not say where is Can’t tell. - Daily and monthly caps on spend: every day (or month) of the period gets a line. Spend within the cap is Allowed only when the spend report covers the whole day (or month) and every row names a campaign in your currency. Otherwise it is Can’t tell. - Spend over a cap is Not allowed only when this agent alone set the budgets in force at the start of that day, no budget changed during it, the change history covers it, and those budgets add up to more than your cap. Otherwise it is Can’t tell. Money at risk is the spend minus the cap. - Each agent this period: Not allowed if any line is. Can’t tell if the change history does not cover the whole period, if any of its lines is Can’t tell, or if a change we could not attribute may be its. Otherwise Allowed. The statement also counts the actions it saw: by your agents, by people, by the platform, and “could not tell who”. ## Unusual days - A campaign’s normal day is the middle value (median) of the 7 days before. A day with no spend row counts as zero. Days before the campaign’s first row do not count. - A day is unusual when its spend is more than 3 times the normal day, or more than the normal day plus your daily cap, whichever is higher. - With all 7 days before it, an unusual day is worth a look. With fewer days, or with unreadable rows nearby, it is too early to tell. - A normal day of zero has nothing to compare with, so it is not listed. - An unusual day can be a launch, a sale or a holiday. It is never a rule answer. ## Waste: worth a look An AI tool can follow every rule and still spend on campaigns that never sell. For Amazon Sponsored Products, the check reads your search term report or bulk sheet and lists five kinds of spend worth a look, each with the money, the rows behind it and one suggested next step: 1. Spend with no sales: a campaign, keyword or target spent at least your no-sale threshold with no orders. Paused rows are left out. 2. Cost per sale above your break-even: spend is more than your break-even times the orders. It needs at least 2 orders: one sale says too little. 3. Same keyword in several campaigns: the same keyword and match type spends in more than one campaign in one marketplace. They may compete for the same searches. The money is the spend of every copy except the one that did best. 4. Search term with no sales: a customer search term cost at least your threshold across campaigns with no orders. The step is to add it as a negative. 5. Outside your marketplaces or products: a campaign runs in a marketplace or advertises a product you did not list. ### Too early to tell A line is “Too early to tell”, never “Worth a look”, when: - the report covers fewer days than your window (14 days unless you change it), or does not say which dates it covers; - the rows have fewer than 10 clicks; - some rows for that campaign could not be read; - your reports disagree on its sales, for example one shows orders the other does not; - Kindle Unlimited pages were read, because orders alone do not show those sales; - you said your books are in Kindle Unlimited and no report with a KENP Read column covers those dates; - we cannot tell which marketplace it ran in, or two campaigns share its name; - the chance of bad luck is too high (below). ### The bad-luck test Some clicks with no sale can be plain chance. So a no-sales line is worth a look only when bad luck is an unlikely explanation. - Your usual sale rate is p = (orders + 1) ÷ (clicks + 31), from all the campaigns in your files, by marketplace, and by product when the line is about one product. - The +1 and +31 are a cautious starting point worth about 1 sale in 31 clicks (statisticians call it a Beta(1, 30) prior). A few hundred clicks of your own outweigh it. - The chance of n clicks with no sale at that rate is (1 − p) multiplied by itself n times. - Worth a look only when that chance is 1 in 20 or less. Otherwise the line is too early to tell, and says the odds. - If unreadable rows could change your usual rate, the line is too early to tell. Example from the sample: at about 1 sale per 20 clicks, 70 clicks with no sale is about a 1 in 34 chance, so it is worth a look. At the same rate, 16 clicks with no sale happens about 1 in 2 times, so it is too early to tell. ### Kindle Unlimited Page reads are sales that orders do not show. Keep the KENP Read column in your search term report. Without it, page reads do not count as sales, and the statement says so at the top of the waste section. If you tick “My books are in Kindle Unlimited”, sales-based lines without that column are too early to tell. ## How we handle money - Amounts are kept exactly as your files show them, as decimals. We add and subtract without rounding. - Comparisons are multiplied out, never divided. For example, cost per sale compares spend with break-even times orders. - Each currency is kept apart. Rows in a currency other than yours are not added up, and the statement says how many. - Each spend row counts once. When two reports describe the same money, the larger report counts, never both. A total may count too little, never too much. - Only words for reading are rounded: odds such as “1 in 34” (to whole numbers, and to tens above 100) and “about 3.4×” a normal day. ## What we never do - Change a campaign, bid or budget, or pause anything. - Move, hold, approve or block money. - Ask for your passwords or upload your files. - Guess who acted, or fill in missing data. - Let an AI model decide an answer, or use a hidden score. - Promise a return on spend. Whether spend was worth it is your call. ## Limits - Platforms: Amazon Ads, Meta Ads and Google Ads only. Purchases, refunds, company cards and accounting software come later. - Files: CSV only. Save spreadsheets as CSV first. - The waste check covers Amazon Sponsored Products. Meta and Google waste checks come later. - Amazon campaign history has no export, so Amazon changes are not checked yet, and Amazon spend over a cap is Can’t tell. - Platforms rename menus and columns. Our column lists follow the platforms’ current names; if one changes, the check stops and names the column. - A statement describes one period. It is not legal, accounting, tax or financial advice, and it does not approve any person, agency or tool. ## See it on made-up data The sample statement is the free check’s own output for an invented business and an invented book seller. Every name and number is made up. Open the sample statement (https://billevera.com/methodology/sample-statement.html) This page and the sample as plain text (https://billevera.com/llms-full.txt) Spotted a difference between this page and the check? Write to hello@billevera.com. --- # Sample statement (made-up data) Made-up sample data: every business, agent, campaign, book and number below is invented. This is the plain-text statement the free check writes for its built-in sample. Agent Statement 2026-10-19 to 2026-10-25 6 Not allowed, 1 Can't tell, 7 Allowed, 2 Not checked - Not allowed | | Ads agent (agency AI) | At least one action or budget broke this agent's rules. - Not allowed | 2026-10-21 | Ads agent (agency AI) | Set a daily budget of 90 CAD, above the cap of 60. - Not allowed | 2026-10-21 | Ads agent (agency AI) | Raised daily_budget from 45 to 90, more than 30% without your approval. - Not allowed | 2026-10-22 | Ads agent (agency AI) | Spend of 88.40 is 28.40 over the daily cap of 60. The daily budgets this agent set add up to 90, above the cap. - Not allowed | 2026-10-23 | Ads agent (agency AI) | Spend of 88.40 is 28.40 over the daily cap of 60. The daily budgets this agent set add up to 90, above the cap. - Not allowed | 2026-10-23 | Ads agent (agency AI) | Acted on brand-search, which you marked off-limits. - Not allowed | 2026-10-24 | Ads agent (agency AI) | Acted in US, outside the places you allowed. - Can't tell | | Email assistant | 1 action(s) could not be attributed and may be this agent's. - Can't tell | 2026-10-24 | Could be: Ads agent (agency AI), Email assistant | Made with a login shared by ads-agent, email-agent; cannot tell which agent acted. Unusual days - Worth a look | 2026-10-22 | prospecting | Spent 68.00 CAD on Oct 22 — about 3.4× a normal day for this campaign (a normal day: 20.00 CAD, the middle of the 7 days before). - Too early to tell | 2026-10-24 | holiday-test | Spent 52.00 CAD on Oct 24 — about 4.2× a normal day for this campaign (a normal day: 12.50 CAD, the middle of the 4 days before). Too early to tell: only 4 days of spend before it; we wait for 7. Waste: worth a look 7 Worth a look, 8 Too early to tell Money worth a look (may be wasted): 8.40 GBP Money worth a look (may be wasted): 105.30 USD - Worth a look | 33.60 USD | Campaign: Salt Orchard - Category | Campaign spent 33.60 USD on 70 clicks with no orders (your threshold is 4.50 USD). With your usual sale rate for this product in US (about 1 sale per 20 clicks), 70 clicks without a sale is about a 1 in 34 chance of bad luck. Next step: Pause the campaign, or check its keywords, bids and product. - Worth a look | 26.00 USD | Campaign: Salt Orchard - Broad · Ad group: Ad group 1 · Keyword: cozy mystery (broad) | Target cozy mystery (broad) spent 26.00 USD for 2 orders. At your break-even of 4.50 USD per sale, 2 orders cover 9.00 USD. Next step: Lower the bid on this keyword or target, or check the price and listing. - Worth a look | 14.50 USD | Campaign: Salt Orchard - Exact 1 · Ad group: Ad group 1 · Keyword: orchard mystery (exact) | Target orchard mystery (exact) spent 14.50 USD for 3 orders. At your break-even of 4.50 USD per sale, 3 orders cover 13.50 USD. Next step: Lower the bid on this keyword or target, or check the price and listing. - Too early to tell | 12.80 USD | Campaigns: Salt Orchard - Auto, Salt Orchard - Broad · Search term: free kindle mystery | Shoppers searching “free kindle mystery” cost 12.80 USD over 16 clicks across 2 campaigns with no orders. Too early to tell: with your usual sale rate for this product in US (about 1 sale per 20 clicks), 16 clicks without a sale still happens by chance about 1 in 2 times. Next step: Add it as a negative exact keyword in: Salt Orchard - Auto, Salt Orchard - Broad. - Too early to tell | 12.40 USD | Campaign: Salt Orchard - Broad · Ad group: Ad group 1 · Keyword: small town mystery (broad) | Target small town mystery (broad) spent 12.40 USD on 22 clicks with no orders (your threshold is 4.50 USD). Too early to tell: with your usual sale rate for this product in US (about 1 sale per 20 clicks), 22 clicks without a sale still happens by chance about 1 in 3 times. Next step: Pause this keyword or target, or lower its bid. - Too early to tell | 11.70 USD | Campaign: Salt Orchard - Exact 2 | Campaign spent 11.70 USD on 18 clicks with no orders (your threshold is 4.50 USD). Too early to tell: with your usual sale rate for this product in US (about 1 sale per 20 clicks), 18 clicks without a sale still happens by chance about 1 in 2 times. Next step: Pause the campaign, or check its keywords, bids and product. - Worth a look | 11.70 USD | Campaigns: Salt Orchard - Exact 1, Salt Orchard - Exact 2 · Keyword: orchard mystery (exact) | The same keyword orchard mystery (exact) runs in 2 campaigns — they may compete for the same searches; worth checking which one to keep. The copies other than Salt Orchard - Exact 1 spent 11.70 USD. Next step: Keep it in one campaign (Salt Orchard - Exact 1 has done best so far) and pause the other copies. - Too early to tell | 9.90 USD | Campaign: Fennel Lane - Exact copy · Ad group: Ad group 1 · Keyword: christmas romance (exact) | Target christmas romance (exact) spent 9.90 USD for 1 order. At your break-even of 4.50 USD per sale, 1 order covers 4.50 USD. Too early to tell: only 1 sale so far; we wait for 2 before judging the cost per sale. Next step: Lower the bid on this keyword or target, or check the price and listing. - Worth a look | 9.90 USD | Campaigns: Fennel Lane - Exact, Fennel Lane - Exact copy · Keyword: christmas romance (exact) | The same keyword christmas romance (exact) runs in 2 campaigns — they may compete for the same searches; worth checking which one to keep. The copies other than Fennel Lane - Exact spent 9.90 USD. Next step: Keep it in one campaign (Fennel Lane - Exact has done best so far) and pause the other copies. - Too early to tell | 9.60 USD | Campaign: Backlist - Auto | Campaign spent 9.60 USD on 16 clicks with no orders (your threshold is 4.50 USD). Too early to tell: your reports show no sales yet for this product in US, so we start from a cautious rate of about 1 sale per 47 clicks; 16 clicks without a sale still happens by chance more often than not. Next step: Pause the campaign, or check its keywords, bids and product. - Worth a look | 9.60 USD | Campaign: Backlist - Auto · Product: B0SAMPLE77 | Campaign advertises product B0SAMPLE77, which is not in your list of products. Next step: Check the product: pause this product ad if it is not one you sell or meant to advertise. - Worth a look | 8.40 GBP | Campaign: Fennel Lane - UK · Marketplace: UK | Campaign runs in marketplace UK, which is not in your list (US). Next step: Check the marketplace: pause the campaign if you did not mean to advertise there. - Too early to tell | 5.20 USD | Campaign: Salt Orchard - Phrase | Campaign spent 5.20 USD on 8 clicks with no orders (your threshold is 4.50 USD). Too early to tell: only 8 clicks so far; we wait for 10. Next step: Pause the campaign, or check its keywords, bids and product. - Too early to tell | 4.90 USD | Campaign: Salt Orchard - Auto · Search term: b0sample55 | Shoppers searching “b0sample55” cost 4.90 USD over 11 clicks across 1 campaign with no orders. Too early to tell: with your usual sale rate for this product in US (about 1 sale per 20 clicks), 11 clicks without a sale still happens by chance more often than not. Next step: Add it as a negative product target in: Salt Orchard - Auto. - Too early to tell | 4.60 USD | Campaign: Fennel Lane - Broad · Search term: romance audiobook | Shoppers searching “romance audiobook” cost 4.60 USD over 4 clicks across 1 campaign with no orders. Too early to tell: only 4 clicks so far; we wait for 10. Next step: Add it as a negative exact keyword in: Fennel Lane - Broad. ## The prompt the check gives the owner to ask their own AI tool The owner can paste this into their AI tool and compare its answer with the statement, side by side. List every change you made to my ad accounts and every purchase you made between 2026-10-19 and 2026-10-25. For each one give the date, what changed, the old and new values, and the amount. Tell me which of them broke any of these rules: daily cap CAD 60, monthly cap CAD 1,500, ask me before raising anything by more than 30%, never touch brand-search, Canada only. If you are not sure about something, say so.