Bid Strategy Troubleshooting: Why Your ROAS Is Tanking

When ROAS drops on a smart bidding campaign, the reflex is to blame the algorithm and change the target. That reflex is usually wrong twice: the cause is more often upstream of the algorithm, and changing the target mid-slump adds a second problem on top of the first. This post is my diagnostic order for a bidding slump - the same fix-nothing-until-diagnosed discipline as my Performance Max troubleshooting guide, applied to the bidding layer that sits under every campaign type in the paid media series.

The order matters because the checks are sorted by how often each cause is guilty and how cheap it is to rule out. Work down the list and stop at the first hit.

Check 1: did measurement break before performance did?

Smart bidding optimizes toward your conversion data, so corrupted data produces corrupted bidding with perfect obedience. Before judging the strategy, I pull up the conversion action itself and look for a change in the week before the slump: a tag edit, a site migration, a consent banner update, a new conversion action added to the “Conversions” column, duplicate counting after a platform integration. The tell is conversion volume moving in a way traffic does not explain.

The subtle version is an attribution change. Google’s own documentation says it plainly: change your attribution model without updating bids and targets, and “the change in conversion attribution may result in over- or under-bidding.” Credit moves between campaigns, each campaign’s apparent CPA and ROAS shift, and the algorithm chases the new numbers with your old targets. If anyone touched attribution settings recently, you have probably found your cause, and the fix is recalibrating targets to the new numbers rather than reverting in a panic.

Check 2: are you reading lag as decline?

The most common false alarm costs nothing to check. Conversions are reported against the click date, and many arrive days or weeks after it, so recent windows always understate results - Google notes that conversion lag “can sometimes cause CPA to look inflated and ROAS to look deflated.” A seven-day ROAS in a business with a two-week consideration cycle is guaranteed to look like a crisis every single week.

My rule: never judge a period younger than your typical conversion delay, which the “days to conversion” segment in your reports will show you. If the “slump” lives entirely inside the lag window, it is arithmetic, and the correct response is to stop looking at dashboards cut to “last 7 days” for lag-heavy accounts at all.

Check 3: did something restart the learning period?

Every significant change - a new bid strategy, a big target move, major budget shifts, merged campaigns - puts the strategy back into learning, and learning is expensive. Google’s guidance says it can take “up to around 50 conversion events or 3 conversion cycles” for a bid strategy to calibrate after a change of objective. During that window performance is legitimately worse and legitimately not informative.

The failure mode I see most is the thrash loop: performance dips, someone adjusts the target, the adjustment restarts learning, learning performs worse, someone adjusts again. The account never leaves the learning state and everyone blames the machine. The change history is the five-minute check - if targets or budgets moved more than once in the past few weeks, the campaign has not had a fair trial since the first move, and the fix is a freeze, not another edit. The same guidance applies to data exclusions, which look like a tidy way to erase a bad tracking week: Google warns against using them “frequently or for prolonged periods,” and notes that excluding a week or more of clicks can keep performance fluctuating for one to two conversion cycles afterward.

Check 4: is the strategy starving?

Smart bidding is a statistics machine, and thin data makes bad statistics. Google’s floor for Target ROAS on Search and Shopping is at least 15 conversions in the past 30 days, but the floor is not the comfort zone. Optmyzr’s study of 14,584 accounts put the threshold for predictably successful automated bidding at 50 or more conversions in 30 days - and found no account under 25 conversions outperforming those above 50. Between Google’s floor and that threshold lives a zone where the strategy runs but wobbles, and week-to-week ROAS swings are variance wearing a trend’s clothing.

If your campaign sits under that line, the fixes I reach for are structural: consolidate campaigns so conversion data pools, optimize to a higher-volume conversion action upstream in the funnel (with the value question that changes, handled honestly), or accept manual-adjacent strategies until volume grows. What does not fix starvation is a tighter target. A starving algorithm with a stricter goal simply starves faster.

Check 5: is the target fighting the market?

Only now, with measurement sound, learning settled and volume adequate, does the target itself go on trial. A Target ROAS is a constraint, and a constraint set above what the auction currently offers does not produce your target at lower volume - it produces spend collapse, as the system finds fewer and fewer auctions it believes can clear the bar. The tell is impressions and spend falling while the ROAS on what remains looks deceptively healthy.

The market moves under fixed targets: competitors enter, CPCs rise seasonally, your conversion rate shifts with your site and offer. My practice is to review targets against a 90-day trailing reality once a quarter, move them in steps of no more than about 20% (large jumps restart learning, which is check 3 again), and then leave them alone. For which strategy should be carrying which campaign in the first place, my smart bidding selection guide is the companion piece - this post assumes the strategy choice was right and something changed.

What this order refuses to do

Two omissions are deliberate, and I hold both. I do not touch creative here, because ad fatigue produces a slow CTR-led decline with its own signature, and that diagnosis has its own system - running both diagnostics at once on one campaign is how you learn nothing from either. And I do not switch bid strategies as a fix. Google promotes strategy upgrades with numbers like its claim that moving from Target CPA to Target ROAS yields 14% more conversion value at similar return - footnoted, read the fine print, to Google internal data from February 2023 - but a strategy switch on a sick campaign resets learning on top of an undiagnosed problem. Switch strategies when the campaign’s goal changed, on a healthy campaign, deliberately. The governance framing for all of this - you set constraints, the platform executes - is the through-line of my PPC manager playbook.

FAQ

How long should I wait before judging a bid strategy change? Until learning completes and a full conversion cycle lands after it - in practice, the roughly 50 conversion events or up to 3 conversion cycles from Google’s guidance, plus your conversion lag on top. In a low-volume account that can honestly mean six weeks, which is itself an argument for making fewer, larger, better-planned changes.

My ROAS dropped but conversions look normal. Where do I look? Value, not volume: a feed price change, a shifting product mix, a discount period changing average order value, or a value-tracking change. ROAS can fall with perfectly healthy conversion counts when the value per conversion moved, and the conversion-value column by product or conversion action will show it.

Should I use portfolio bid strategies to fix low volume? Pooling campaigns into a portfolio genuinely helps the data-starvation problem, and it is the structural fix I reach for first when several small campaigns share one economic goal. It does not help when the campaigns have different margins or goals - pooling those just hides two problems inside one average.

Is manual bidding ever the answer in 2026? Rarely, and almost only in the starvation zone: very low volume, high-value conversions, where a human’s prior beliefs genuinely beat an algorithm’s thin data. Everywhere else the leverage is in what you feed and permit the machine, which is the whole shape of the modern job.

This post is part of my paid media series. The strategy-selection companion is the smart bidding guide, the creative-side diagnosis is creative fatigue detection, and the wider practice starts at the complete guide to AI marketing.