Supply and installation of solar panels, home batteries and EV chargers for homeowners across the East Midlands, from a showroom and warehouse in Newark. 46 staff. 2 ad platforms and 1 analytics tool; £10,500 a month.
Verdict: Over 6 weeks the platforms' figures were out of line with the business's records: ad enquiries 107 of 123 (87%); quotations 38 of 38 (100%); orders 14 of 14 (100%); analytics enquiries 14 of 118 (12%), plus 62 counted conversions that match no business record. £10,500 a month of ad spend is steered by figures that don’t match the records. The audit found 16 failures and 4 partial passes, including tags that run before consent. The specification sets out 13 fixes, each with a test and a pass condition, due by 27 Nov 2026. Once they pass, 3 of the 5 catalogue products named can run on figures that match the records; 2 already can and aren't counted. Once tags wait for consent, the platforms will count only visitors who accept, and lower figures will be correct; each lead's consent choice is recorded in the CRM so that target is counted, not estimated. Measure again on 28 Feb 2027 over a window of the same length, offline rows on 4 Apr 2027.
Reported by the platform Recorded by the business Counted conversions that match no business record (phone-number taps) are on top of these: 62 in the window.
| Row | Platform and record | Reported | Recorded | Accuracy | Finding | How matched; counted by |
|---|---|---|---|---|---|---|
| R1 | Search ads platform Web form enquiry | 120 | 52 | 231% | Over-counting | Source field on the record; tag before consent |
| R2 | Search ads platform Phone enquiry | 22 | 31 | 71% | Under-counting | Asked the customer; calls from ads |
| R3 | Search ads platform Quotation issued | 0 | 38 | 0% | Not tracked | Source field on the record; not tracked |
| R4 | Search ads platform Order signed | 0 | 11 | – | Not tracked few records: judged by the test log | Source field on the record; not tracked |
| R5 | Social ads platform Web form enquiry | 58 | 34 | 171% | Over-counting | Source field on the record; tag before consent |
| R6 | Social ads platform Phone enquiry | 0 | 6 | – | Not tracked few records: judged by the test log | Asked the customer; not tracked |
| R7 | Social ads platform Order signed | 0 | 3 | – | Not tracked few records: judged by the test log | Source field on the record; not tracked |
| R8 | Analytics tool Web form enquiry | 104 | 118 | 88% | Under-counting | All received (no source needed); tag before consent |
Accuracy is reported ÷ what the row can count. The ±10% band is a diagnostic, not a pass mark: look-back windows, cross-device journeys and modelled conversions all move platform figures. Rows on fewer than 30 records are judged by the test log. Every row should count 100% today because the tags fire before consent. Once they wait for it, form rows can count only visitors who accept; the CRM will record each lead's consent choice (fix F7), so that target is counted from the records rather than estimated. Search ads platform: 180 reported v 132 recorded; Social ads platform: 82 reported v 43 recorded; Analytics tool: 104 reported v 118 recorded.
| Ref | Platform and action | Trigger; counts | Counted? | Reported | Test log | Problem |
|---|---|---|---|---|---|---|
| A1 | Search ads platform Contact form submit | Button click; every conversion | counted Web form enquiry | 71 | Fired twice or more 30-day click; view-through no; consent mode not set | Fires twice |
| A2 | Search ads platform Thank you page | Thank-you page; one per click | counted Web form enquiry | 49 | Fired once 30-day click; view-through no; consent mode not set | Counted twice: another counted action records the same thing |
| A3 | Search ads platform Calls from ads | Call from an ad; one per click | counted Phone enquiry | 22 | Fired once 30-day click; view-through no; consent mode not set | OK |
| A4 | Search ads platform Website phone number taps | Button click; every conversion | counted None (not a business result) | 38 | Fired once 30-day click; view-through no; consent mode not set | Counted but not a business result |
| A5 | Search ads platform Brochure download | Page visit; one per click | secondary None (not a business result) | 15 | Not tested 30-day click; view-through no; consent mode not set | OK (not counted) |
| A6 | Social ads platform Lead | Thank-you page; one per click | counted Web form enquiry | 58 | Fired twice or more 7-day click; view-through yes; consent mode not set | Fires twice |
| A7 | Social ads platform Contact | Button click; every conversion | counted None (not a business result) | 24 | Fired once 7-day click; view-through yes; consent mode not set | Counted but not a business result |
| A8 | Analytics tool Form sent | Form-sent event; one per click | counted Web form enquiry | 104 | Fired once view-through no; consent mode not set | OK |
| Ref | What is checked | Result | Evidence |
|---|---|---|---|
| C1 | A tag (or the tag manager) is on every page where a conversion happens: thank-you pages, forms, booking and call pages | Pass | Site crawl of all 64 pages: the tag manager loads on every page, including the four thank-you pages |
| C2 | Each conversion fires once: one test enquiry gives one conversion in each platform's test view (test log) | Fail | One test enquiry from each platform's test mode: the search ads platform showed 2 form conversions (button click and thank-you page), the social ads platform 2 Lead events (browser tag and the plugin's server copy) |
| C3 | Forms are counted by a thank-you page or a form-sent event, not by a button click | Partly | 'Contact form submit' on the search ads platform is a button click; the battery quotation form has no form-sent event (4 test sends, 3 recorded in the analytics tool) |
| C4 | Non-customer form fills (job applications, suppliers, spam) are kept out of the conversion | Fail | Job applications go through the contact form: 9 in the window, all counted by both ad platforms and the analytics tool |
| C5 | Calls from ads are counted, with a minimum call length | Pass | Calls from ads counted at 60 seconds or more; a test call from an ad on 12 Oct was recorded once |
| C6 | Calls from the website are counted: tracking numbers by source, with a minimum length | Fail | The website shows one fixed number; a call from someone who clicked an ad first isn't counted anywhere |
| C7 | Quotations and orders are sent back from the CRM (offline upload) within each platform's look-back limit, matched by click ID where captured, or by hashed email or phone on a lawful basis; orders cancelled in the cooling-off period are sent as retractions or adjustments | Fail | No quotations or orders are sent back to either platform |
| C8 | Each enquiry's source is captured on the CRM record (a hidden form field; asked and logged for calls) | Pass | A hidden source field has been on every form since March 2026 (4 test enquiries arrived with the right source); the sales office asks and logs the source of every phone enquiry |
| C9 | Where a server copy is sent as well as the browser tag (a conversions API), both carry one shared event ID so the platform counts it once | Fail | The website plugin sends a server copy of each social Lead without the browser tag's event ID; the test view showed two events |
| C10 | Enquiry and quotation actions count one per click, not every conversion | Fail | 'Contact form submit' and 'Website phone number taps' are set to count every conversion |
| C11 | Only business results count in the platforms' conversions column; page visits, taps and downloads are kept as secondary | Fail | Phone-number taps count as conversions on both ad platforms (38 and 24 in the window) |
| C12 | Advertising and analytics tags store or read nothing on a visitor's device, and receive nothing the browser sends automatically (cookieless pings), until the visitor consents; refusing is as easy as accepting (PECR regulation 6; ICO) | Fail | Fresh-browser test on 12 Oct: ad and analytics tags load before any choice; the banner offers 'Accept' and 'Settings' only |
| C13 | The consent choice is passed to every tag, so tags hold back when a visitor refuses and resume if they accept | Fail | No consent signals are set in the tag manager; the tags ignore the banner |
| C14 | Data sent to the platforms is minimised: only the identifiers needed, hashed where sent, and only on a lawful basis | Partly | The plugin's server copy sends email, phone, name and postcode with every lead; only email and phone are needed for the match |
| C15 | The privacy notice names the kinds of ad and analytics platforms, what is shared with them, why, the lawful basis and how to object | Fail | The privacy notice (2023) mentions analytics only: no ad platforms, no offline matching, no way to object |
| C16 | Test conversions are kept out of the platforms' reports and marked in the CRM | Pass | Tests use each platform's test mode; the sales office marks test leads 'TEST' and they are excluded from the CRM counts |
| C17 | The analytics tool counts the same enquiries as the business (one event per form) and excludes internal traffic | Partly | One event per form except the battery form (see C3); visits from the showroom and office aren't excluded |
| C18 | The business owns admin access to every ad account, the tag manager and the analytics tool, and changes are logged | Partly | The agency owns the tag manager account and the business has no admin access to it; no change log is kept |
| C19 | Each platform's click ID and the visitor's consent choice are captured in hidden fields on each CRM lead (and passed by the call tracking service for calls) | Fail | No click ID or consent field on the forms or the CRM leads (4 test enquiries checked) |
| C20 | Enhanced conversions (form details hashed and sent with the tag) run only after consent and are covered by the privacy notice | Fail | The website plugin sends email, phone, name and postcode with each social lead before any choice; the privacy notice doesn't mention it |
| C21 | Leads from forms hosted on the ad platforms reach the CRM once, are counted as business records, and are reconciled | Not applicable | All live campaigns on both platforms checked on 12 Oct: no platform-hosted lead forms in use |
| C22 | Click IDs and the source survive redirects and any other domain in the journey (booking, payment or quotation tools) | Fail | The home-survey booking page runs on the booking tool's own domain; a test link's source and click ID were dropped there |
| C23 | Each campaign bids on the counted business results (campaign-level conversion goals match the account's counted actions) | Fail | Two search campaigns bid on 'Website phone number taps' as their only campaign goal |
| C24 | Server copies honour 'Reject all': nothing is sent from the server for a visitor who refused | Fail | The plugin's server copy of each social Lead is sent whatever the visitor chooses (test with 'Settings' then refuse: Lead received) |
| C25 | An analytics tool linked to an ad account (shared audiences, imported conversions) waits for consent: the statistics exception no longer applies | Fail | The analytics tool is linked to the search ads account (audiences shared, conversions imported), so the statistics exception can't apply |
C12 to C15, C20, C24 and C25 cover consent and data under PECR regulation 6 and UK GDPR, with the ICO's guidance cited in the method. Not legal advice.
| Fix | What to change | Test | Done by | Due |
|---|---|---|---|---|
| F1 A1, A2, C10, R1 | Make 'Contact form submit' a secondary action (delete it after 30 days); keep 'Thank you page' as the only counted form action, counting one per click Search ads platform; tag manager | Send three test enquiries from a test ad click, the first with a deliberate error Pass when: The platform's test view shows one form conversion for each enquiry sent and none for the failed send | Agency Marketing manager | 6 Nov |
| F2 A4, A7, C11, C23 | Make phone-number taps secondary (kept for observation, not counted or bid on), and set every campaign's conversion goal to the counted business results Both ad platforms | Tap the number on a phone after a test ad click on each platform; read each campaign's goal settings Pass when: The tap appears as a secondary action outside the conversions column, and no campaign's goal includes it | Agency Marketing manager | 6 Nov |
| F11 C17 | Exclude visits from the showroom and office as internal traffic Analytics tool | Browse the site from the office network Pass when: The visit is marked internal and left out of the reports | In-house Marketing manager | 6 Nov |
| F13 C18 | Transfer admin ownership of the tag manager to the business; give the agency user access; name each published tag version with what changed Tag manager; both ad accounts; analytics tool | List the users and their access on each account Pass when: The business is admin on every account and each tag version has a note | Owner Managing director | 6 Nov |
| F3 A6, C2, C9, C24, R5 | Remove the copy of the social tag hard-coded in the theme; give the browser tag and the plugin's server copy one shared event ID per enquiry; send the server copy only for visitors who accepted Social ads platform; website plugin; tag manager | Submit a test enquiry with the platform's test-events tool open, once accepting and once after 'Reject all' Pass when: Accepting: one Lead, received from browser and server and marked as deduplicated. Refusing: nothing received | Developer Marketing manager | 13 Nov |
| F4 C3, R8 | Add the form-sent event to the battery quotation form, with the form's name as a parameter Website; tag manager; analytics tool | Send one test from each of the site's four forms Pass when: The analytics tool's real-time view shows one form-sent event per form, each with its form name | Developer Marketing manager | 13 Nov |
| F5 C4 | Move job applications to a separate careers form with its own thank-you page that carries no conversion tag Website | Send a test job application Pass when: No conversion appears in any platform or the analytics tool | Developer Marketing manager | 13 Nov |
| F7 C19 | Capture each platform's click ID (kept on the device only after consent) and the visitor's consent choice in hidden form fields, stored on the CRM lead; the call tracking service passes the same for calls Website forms; CRM | Send test enquiries from a tagged test link for each platform, once accepting and once refusing Pass when: Accepting: the CRM lead shows the click ID and 'accepted'. Refusing: no click ID kept, 'refused' recorded | Developer Marketing manager | 13 Nov |
| F9 C12, C13, C25 | Hold every advertising and analytics tag until the visitor accepts, using the basic consent mode (nothing loads or is sent before consent, no cookieless pings); the analytics tool waits too, as it is linked to the search ads account; put 'Reject all' beside 'Accept all' on the first layer with equal prominence; pass the choice to every tag; keep a record of choices Website; consent banner; tag manager | Fresh browser: reject all, then browse to a thank-you page with the network log open; repeat ignoring the banner; repeat accepting Pass when: Refusing or ignoring: no advertising or analytics request of any kind is sent and only strictly necessary cookies are set. Accepting: the tags fire | Developer Marketing manager | 13 Nov |
| F6 C6, R2, R6 | Show tracking numbers on the website by source, loading the tracking script only after consent; send calls of 60 seconds or more as conversions to the platform the visitor came from, and log the source and consent on the CRM lead (calls from visitors who refused keep the source asked) Website; call tracking service; both ad platforms; CRM | From a test ad click on each platform, ring the number shown and stay on for 70 seconds Pass when: One call conversion in the right platform, and a CRM phone enquiry with that source | Developer Marketing manager | 20 Nov |
| F10 C14, C15, C20 | Set the plugin to send only hashed email and phone; update the privacy notice to name the kinds of ad and analytics platforms, what is shared, why, the lawful basis and how to object; record a legitimate interests assessment for the offline matching Website plugin; privacy notice | Read the plugin's payload in the platform's test tool; read the published notice Pass when: Only hashed email and phone in the payload; the notice is published with its date and covers matching and objection | Owner Managing director | 20 Nov |
| F12 C22 | Pass the source and click ID across to the booking tool's domain (cross-domain linking), or embed the booking form on the main site Website; booking tool | Book a test survey from a tagged test link for each platform Pass when: The booking reaches the CRM with the right source and click ID | Developer Marketing manager | 20 Nov |
| F8 C7, R3, R4, R7 | Upload quotations issued and orders signed each week from the CRM, dated by the CRM date, within each platform's upload look-back limit (check it), for enquiries made after click-ID capture started; match by click ID, and on the social ads platform also by hashed email and phone only once the privacy notice covers it (F10); send orders cancelled in the 14-day cooling-off period as retractions or adjustments CRM; both ad platforms | Upload a test file of two quotations, one order and one cancellation Pass when: The platform's upload report shows the rows accepted and matched, dated as in the CRM, and the cancelled order retracted | In-house Sales office manager | 27 Nov |
| ROI product | What its own kit asks for | Runs off | Status |
|---|---|---|---|
| Paid Advertising Wasted Spend Audit | "Check tracking first. Compare the platform's conversions that count ... with the business's own count." If the platform counts fewer, zero-conversion items may be converting. | Platform conversions Search ads platform | Unlocked once the fixes pass their tests |
| Paid Ad Creative Review | "Spend, frequency, click rates and conversions by ad" from each platform, and "which conversions were real enquiries" from the enquiry log | Platform conversions Social ads platform | Unlocked once the fixes pass their tests |
| Ad Spend and Attribution Review | "New customers credited to each channel come from your own order or CRM records, each counted once" | Source recorded in the CRM | Already possible: not counted |
| Seasonal Ad Budget Plan | "New customers from ads by month" from "your shop or CRM" | Source recorded in the CRM | Already possible: not counted |
| Landing Page Conversion Plan | "Visits and enquiries by landing page" from "analytics: landing pages report, with enquiries as a conversion" | Analytics conversions Analytics tool | Unlocked once the fixes pass their tests |
The counted line, before: ad enquiries 107 of 123 (87%); quotations 38 of 38 (100%); orders 14 of 14 (100%); analytics enquiries 14 of 118 (12%); plus 62 counted conversions matching no business record. Re-measure: Not yet re-measured. 30 Nov 2026; measure again 28 Feb 2027, over 11 Jan 2027 to 21 Feb 2027; offline rows 4 Apr 2027, counting only enquiries after click-ID capture (13 Nov 2026). No £ is put on better tracking: the money is made in the products it unlocks, each valued from the business's own records.
Cost in full. £6,850 cash one-off (fee £2,400, developer or agency £3,850, adviser £600); tools £115 a month; team time 24 hours (£672); £8,902 in the first year.
The platform figures are from each platform's report by conversion action for the window on Inputs (by conversion date, click-through only), exported 9 Oct 2026; the business's figures are from the sources named on each row, matched as stated (a click ID is firmest, asking the customer the weakest). The test log and audit results are the consultant's observations on the dates given, not a continuous check. The tolerance (±10%) is a diagnostic, not a pass mark: look-back windows, cross-device journeys and modelled conversions all move platform figures. It and the fewest records (30) are assumed; rows below that are judged by the test log. Rows counted after consent are judged against the records carrying consent where the CRM holds them, otherwise against the share of all visitors who accept, counted from the banner's records (55%; ignoring the banner is not consent). Offline rows count only enquiries after click-ID capture and are re-measured later by the enquiry-to-order days. The value case is Unlocks: the fixes let the named products run on figures that match the records; no £ is put on better tracking, and the counted line is measured again the same way. Costs: developer or agency quoted, adviser estimated until quoted; team hours estimated. Consent and data points are not legal advice.
The report and specification are finished when they are handed over; the fixes are the developer's and agency's next job, each signed off only when its test passes.
Thrushcombe chose the guided route. At the last session the agent was set up in an AI assistant the business already uses, connected read-only to exports of both ad platforms' reports and the analytics tool, the CRM and the consent banner's records (its IT confirmed this before it committed). Every Monday at 7:00 it reruns the reconciliation in a new dated copy of the check, over a six-week window ending eight days before the run, and sends the marketing manager a short note. It uses the rules on Inputs: a tolerance of ±10% as a diagnostic, rows on fewer than 30 records judged by the test log, and the check's window and export delay. It never logs in to change an ad account. Nothing left the business, and no decision changed, without one of Thrushcombe's own people approving it. These events continue the fictional example.
| Date | What the agent found | What people decided |
|---|---|---|
| 2 Nov 2026 Note only | The first run, over 14 September to 25 October: the check's findings still showing while their fixes were due. Nothing new. | Nothing needed. The marketing manager read the weekly note. |
| 9 Nov 2026 Flag | A new counted action matching no business record: 'Chat started', a button click added to the social ads platform's conversions column on 22 October, 17 in the window. | The marketing manager made it secondary the same day and added it to fix F2's test. Changes to counted actions now need the marketing manager's yes, recorded in the change log. |
| 14 Dec 2026 Flag | Analytics web form enquiries newly over-counting at 112% of recorded. 14 of the form-sent events came from the new careers form, which fix F5 had kept clear of the ad platforms but not of the analytics tool. | The developer removed the event on 16 December, and fix F5's test now covers the analytics tool. |
| 4 Jan 2027 Flag | Data problem: the CRM's count of leads carrying consent included test leads, so row R1 failed its check. The note named the check and gave no figures. | The marketing executive filtered out the test leads. The rerun passed the checks the same morning, with the Christmas window noted. |
| 1 Mar 2027 Note only | The run over 11 January to 21 February, the planned re-measure window: every row judged by tolerance within ±10%, and no counted conversions matching no business record. | The marketing manager enters the re-measure on the Value Case from this run's exports, after checking each source. Offline rows follow on 4 April, as planned. |
At the 30-day check-in in mid November, Geoff and the marketing manager read the first two runs' log together and left the rules as they were. After that the agent, its accounts and its log were Thrushcombe's.
What the agent added. The method is unchanged: the same spreadsheet, the same checks and the business's own rules. The agent made sure the reconciliation was rerun every week while the fixes went in, and caught two new problems that came with the changes: a counted chat button and a careers form feeding the analytics tool. The CRM report error was stopped before any wrong figure reached the marketing manager. Each weekly reconciliation took reading a note instead of exporting and matching by hand (estimate). The re-measure itself is still entered and checked by people. The months after add nothing to the figures above.
Fictional and labelled: these events continue the worked example. The same events are the sample rows in the Agent log of the worked-example spreadsheet; the other weekly rows, which recorded nothing new, are left out of the sample. The agent is part of the guided and done-for-you routes; the self-implement kit does not include it.