What change history records, and what it does not
The account management hub describes change history as the record of every change to the account: who made it, when, what it was and in many cases the previous value. Used well, it answers the question every review begins with, which is what changed before performance moved. The hub is equally clear about the conditions. The record is only useful if each person has their own login, if automated rules and scripts are named so their changes are recognisable, and if significant decisions carry a note.
What the record does not hold matters as much. It does not record why an edit was made or what the person expected; that lives in a decision logA running record, one row per decision, of what changed and why, what was expected, who decided and when the result will be checked., if anyone keeps one. It does not record changes outside the account: a website release, a new checkout, a price change, a feed rule in Merchant Center, a consent banner, a competitor's arrival. And it does not tell you which edit caused a shift; it gives dates and values, and the cause has to be argued from them. This tutorial does not describe the history's retention period, its filter names or its export options; read those in the account or in Google Ads Help.
Two troubleshooting pages read change history as one check for one symptom: the impressions dropped diagnosis and the diagnosis of results that dropped after a bid strategy change. This tutorial is the full investigation for any shift, whatever the symptom. For decomposing a metric change in general, across any channel, the guide to explaining a change in a marketing metric sets out the analyst's method; this page applies the Google Ads account's own record to it.
Before you start
- Read-only or standard access to the account, with change history visible. Editing rights are not needed for the investigation and are safer not to use until the cause is known.
- The daily series for the affected campaigns over a period well before and after the shift: cost, clicks, conversions, cost per conversion (CPACost of each conversion.) or return, and Impression ShareShare of eligible impressions you received. split into budget and rank.
- The business's own records for the same days: orders, enquiries or sales from the shop system or CRM, so that a tracking break can be told apart from a real change.
- A list of known events outside the account: site releases, price changes, promotions, feed changes and stock problems, from the web team and the merchandising calendar.
- The decision log, if the account has one, from the paid search account management playbook.
- The typical conversion lag for the account: how long after a click most conversions arrive.
The steps
Describe the shift and its start date
Write one sentence that states the shift in measurable terms: which metric, which campaigns, from what to what, over which days. "Cost per sale on the shopping campaigns rose from its usual level to well above target over two weeks" is a description; "performance got worse" is not. Plot the metric by day and mark the first day the line broke away from its normal range. That date is the key to everything that follows. Note whether the shift is in volume, in cost, in conversion rate or in all three, because each points to different kinds of edit.
Set the investigation window
Start the window well before the break date and end it at the present. The lead time matters because of Conversion LagThe time between an ad interaction and the conversion it leads to, which leaves the most recent reporting days incomplete.: platforms credit a conversion to the date of the click, so an edit that hurt conversion rate shows up in results reported against the days after it, and the most recent days are always incomplete. An edit to a bid strategy can also take effect gradually. A window that starts on the break date itself will miss the edit that caused it. If the account's lag is long, as with considered purchases or leads that qualify after review, extend the window back further.
Filter changes to the affected campaigns
Narrow the history to the campaigns that moved, then widen it once to the account level, because shared budgets, shared negative lists, conversion actions and account settings sit above the campaigns and an edit there touches all of them. Also look at one campaign that did not move: if the same edit appears there with no effect, the edit is less likely to be the cause. Keep a note of which campaigns were included at each stage, so the reading can be repeated.
Classify each change by type
Sort every edit in the window into one of six types, because each type affects results in a different way and on a different timescale.
Type of change Examples What it usually moves first Budgets Daily budget raised or cut; shared budget reassigned Spend, impressions lost to budget, Budget PacingHow fast a campaign spends its budget across the flight compared with the plan. Bid strategies and targets Strategy switched; target cost per acquisition or return tightened or loosened Bids, impressions lost to rank, cost per click, volume Keywords and negatives Keywords added, paused or changed match type; negatives or shared lists added Search terms matched, impressions, conversion rate Ads and assets Ads paused or edited; assets added or removed Click-through rate, conversion rate, policy status Targeting settings Locations, languages, schedules, audiences, networks Who sees the ads and when; volume Conversion settings Conversion actions added, removed or changed between primary and secondary; counting or value changed Reported conversions; what bidding optimises to Mark who made each change. Edits by people with named logins and by automated rules and scripts with clear names should all be identifiable; anything that is not is itself a finding.
Line changes up with the daily results
Put the classified edits on the daily chart as dated markers. Look for an edit on or just before the break date whose type matches the kind of shift: a budget cut before a fall in impressions lost to budget, a tightened target before a rise in impressions lost to rank, a conversion action change before a jump in reported conversions. Discount edits whose type could not plausibly produce the shift, and edits made to campaigns that did not move. Usually one or two candidates remain.
Allow for conversion lag and learning
Two timing effects disguise the true start of a shift. Conversion lag makes the last days look worse than they will once late conversions arrive; compare like with like by leaving the most recent days out of the reading or by reading conversions by conversion time. And Google documents a learning period after a bid strategy is created or significantly changed, during which performance can fluctuate; the Smart Bidding strategies guide notes that changing the strategy type, changing conversion actions and large changes to targets or budgets are listed as triggers, and that the status column shows "Learning" during the period. A shift that starts with such an edit and settles afterwards may be learning rather than lasting harm.
Check for causes outside the account
Before settling on an edit, rule out what the account cannot record. Compare reported conversions with the business's own records: if orders held steady while conversions fell, tracking broke, and the conversions not tracking diagnosis takes over. Check the web team's release log for the landing pages involved. For product campaigns, check the feed and its item statuses in Merchant Center, as the product feeds guide describes. Read auction insights for a new or stronger competitor. And check whether demand itself changed with the season or the market. When the shift is a fall in conversions while the campaigns still run, the diagnosis for Google Ads conversions that dropped suddenly splits it into clicks, conversion settings, open windows, tracking, bidding, search mix and the page.
Test the leading explanation
State the leading explanation as a cause and a prediction: "the target was tightened on the eighth, so bids fell, so impressions lost to rank rose on the campaigns using that strategy and not on the others." Then check the prediction against data it did not come from: the campaigns not affected, the segment that should have moved most, the metric that should have moved first. Asking why in sequence, as the five whys framework does, takes the finding from the edit to the reason the edit was harmful. If the prediction fails, return to the remaining candidates. Only reverse or soften an edit once the test supports it, and change one thing at a time so the effect can be read.
Record the finding
Write the finding in a few lines: the shift and its dates, the edit or outside event found to cause it, the evidence for and against, the action taken and the date the effect will be checked. Add it to the decision log and link it from the edit's entry. A finding that says "probably the target change, could not be separated from the budget change made the same day" is honest and useful; it tells the team to separate those edits next time. If the team has no log yet, the marketing decision log template sets one up with a column for the reason, the expected effect and the outside events that change history never sees.
Set rules that keep history readable
Finish by fixing whatever made this investigation harder than it needed to be. The hub's conditions are the starting point: one login per person, automated rules and scripts named so their edits are recognisable, and a note on significant decisions. Add the habits the investigation exposed: one material change per campaign at a time, the reason for each budget or target edit written in the decision log the same day, and outside events such as releases and price changes logged beside the account's own history. The account audit tutorial checks these habits as part of its account settings review.
A rise in cost per sale traced to one edit (illustrative)
Glazeburrow Ceramics finds the edit behind a fortnight of dearer sales
Glazeburrow Ceramics, an invented pottery-supplies shop, sells clay, glazes and kiln tools through two shopping campaigns and one search campaign. Over a fortnight its cost per sale on the shopping campaigns rises from £14 to £23 while the search campaign holds at £11. The shop system shows orders from search steady, and orders from shopping down by about a third, so tracking is not the cause. Typical conversion lag is two days, so the investigation window starts three weeks before the break on 9 June.
| Date | Change or event | Type | Campaigns | Who |
|---|---|---|---|---|
| 2 June | Shopping budgets raised from £60 to £75 a day | Budget | Both shopping | Marketing manager |
| 6 June | Target return on the glazes campaign raised from 450% to 600% | Bid strategy target | Shopping glazes | Freelancer |
| 8 June | New product page template released for glazes | Outside the account | Glazes pages | Web developer |
| 9 June | Cost per sale on shopping starts to rise | Shift begins | Both shopping | — |
The budget rise predicts more spend and more sales, not dearer sales, and the search campaign shared nothing with it. The target change predicts lower volume at a higher return on glazes only, which is not what happened: cost per sale rose on the kiln tools campaign too. The template release fits: both shopping campaigns send clicks to product pages, and on the new template the add-to-basket button sits below the delivery table on phones. Conversion rate on mobile shopping clicks has fallen from 2.1% to 1.2% since 8 June, while desktop holds at 2.6%. The web developer moves the button back above the table on 21 June. Over the following fortnight, allowing two days for late conversions, mobile conversion rate returns to 2.0% and cost per sale on shopping to £15. The finding goes in the decision log with a new rule: product page releases are logged beside change history and announced to whoever runs the campaigns.
Common mistakes
- Starting the window on the day of the shift. Conversion lag and gradual bidding effects mean the cause usually comes before the first bad day.
- Reading only the campaigns that moved. Account-level edits to shared budgets, shared lists and conversion actions sit above them and are missed.
- Blaming the most recent edit. The latest change is the most visible, not necessarily the cause; match the type of edit to the type of shift.
- Judging a bid strategy edit during learning. Fluctuation in the learning period is expected; read the status before drawing conclusions.
- Ignoring the world outside the account. Site releases, tracking breaks, feed changes and competitors never appear in change history.
- Reversing several edits at once. If the result recovers, nobody knows which reversal worked.
- Shared logins and unnamed automation. An edit nobody can attribute cannot be asked about.
- No written finding. The next investigation starts from nothing, and the same mistake recurs.
