Scalewell / Blog

AI bidding: when to trust platform bidding and when to set your own rules

Count the qualified conversions each bid strategy would learn from, compare them with each platform's documented floor, and pick between platform bidding, constraints and manual bids campaign by campaign.

A manual control dial and an automated steering wheel side by side on one dashboard, representing platform bidding and manual rules.

AI bidding is the automated bid setting Google, Meta, LinkedIn and Microsoft run in every auction, aiming at the conversion and budget you set. Trust it when a campaign sends enough conversions at the stage you care about, constrain it with caps and value rules when it doesn't, and bid manually where volume is too thin.

You probably know the Monday this goes wrong. A campaign moved to Target CPA two weeks ago, cost per lead looks fine, and then the sales lead asks why half the new leads are students and consultants. Or Maximize conversions spent the full daily budget on a campaign that used to underspend, and finance wants to know who approved it. Nobody did, exactly, because the bidding did what it was told.

This guide is for the head of paid media or PPC lead choosing bid strategies across Google Ads, Meta, LinkedIn and Microsoft Advertising for a B2B account. It sets out each platform's documented data floor as of November 2026, a decision rule for picking a strategy campaign by campaign, the controls that constrain the AI without breaking it, and the cases where manual bidding still wins. What to automate across buying in general sits in our guide to AI media buying, and the wider question of where platform optimization helps or hurts is in AI campaign optimization.

Every platform offers the same four levels of control

AI bidding on all four platforms comes in roughly four levels: hand the platform the budget and let it spend, give it a target, put a hard ceiling on what it can bid, or set the bid yourself. The names differ, which is half the confusion in any cross-channel review.

Level of controlGoogle AdsMetaLinkedInMicrosoft Advertising
Spend the budget, no targetMaximize conversions, Maximize conversion valueHighest volume, Highest valueMaximum deliveryMaximize conversions, Maximize clicks
Aim for an average targetTarget CPA, Target ROASCost per result goal, ROAS goalCost capTarget CPA, Target ROAS
Hard ceiling on the bidBid limits on portfolio Target CPABid capManual bid acts as the ceilingOptional maximum CPC on automated strategies
You set the bidManual CPCNone beyond bid cap, which Meta files under manual biddingManual biddingEnhanced CPC on Search, manual CPC only on audience display, video and lodging campaigns

A few details in that table change real decisions. Google's Smart Bidding page defines its strategies as ones that "use Google AI to optimize for conversions or conversion value in every auction," and Google retired Enhanced CPC: its help page says that "effective the week of March 31, 2025" it "is no longer available for Search and Display campaigns." Manual on Google now means plain manual CPC. Microsoft still offers Enhanced CPC, where you set the bids and Microsoft adjusts them per auction. Its bid strategy guide says you "can only use manual CPC for audience display and video campaigns, and for lodging campaigns," and that a manual CPC request on any other campaign type is set to Enhanced CPC instead, so a Microsoft Search campaign that you bid by hand runs Enhanced CPC. The same guide lets you add an optional maximum CPC to Target CPA, Target ROAS and Maximize conversions.

Meta's help page on bid strategies sorts its Ads Manager options into spend-based (highest volume, highest value), goal-based (cost per result goal, ROAS goal) and manual (bid cap), and notes that "adherence to cost per result goal limits is not guaranteed." Its API documentation adds that the bid cap "does not surpass your specified limit" but may not spend your full budget. On LinkedIn, Campaign Manager's bidding page lists maximum delivery and manual bidding for every objective, while cost cap is limited to brand awareness, website visits, engagement, video views and lead generation. Scalewell helps paid teams use AI to keep track of which strategy each campaign runs and what it returns, so bidding decisions come from evidence instead of memory.

Each platform publishes a different data floor

The conversion volume each platform asks for before its bidding works is documented, and the numbers differ more than most teams assume. These are the figures on each platform's own pages as of November 2026.

Platform and strategyDocumented requirementWhat happens below it
Google Target CPACan start "with no conversion history"Nothing stops it; you set the target from business goals
Google Target ROAS (Search)"At least 15 conversions in the past 30 days"Not eligible
Google lead stage as the bid goalAt least 15 conversions in the last 30 days for that actionBelow Google's guidance; choose an earlier stage that clears it
Microsoft Target CPA, Target ROAS, Maximize conversions30 conversions in any 30-day periodThe strategy "will stop optimizing your bids"
Meta cost per result goal or ROAS goal50 to 100 or more conversions a week recommendedResults may not match the goal you set
Meta conversion leads with CRM dataAt least 200 leads a month, stage reached within 28 days, stage conversion rate of 1% to 40%Falls outside Meta's own fit check
LinkedIn qualified leads goalConversions API integration; no volume floor published, but a two-week learning phase that five or more qualified leads help speed upOnly maximum delivery is available

The Google figures come from its Target CPA, Target ROAS and lead quality pages. Microsoft's line is the strictest, and Microsoft recommends switching strategy if a campaign falls below it regularly. Meta's weekly figure is on its bid strategies help page and the CRM figures come from its Conversions API for CRM guide, and LinkedIn's from its qualified leads help page, which says "the only bidding strategy available is maximum delivery," with the learning phase from its Conversions API use cases page.

So the same campaign can be eligible on one platform and starved on another. Take a hypothetical B2B search program producing 20 sales-accepted leads a month: it clears Google's floor for that stage and falls under Microsoft's. If you copy the Google bid strategy across when you bring those campaigns into Microsoft Advertising, you get a strategy that Microsoft's own documentation says will stop optimizing.

Learning periods differ too. Google's learning period page says the "Learning" status "never lasts more than 7 days," with performance taking one to two conversion cycles to stabilize. Microsoft's Performance Max guidance puts its learning period at two to four weeks or two to three conversion cycles. Meta's developer reference describes a learning phase that "significant edits" restart, with dynamic conversion and day thresholds, so plan on Meta telling you per ad set rather than on one fixed number.

Count qualified conversions per strategy before you choose

The decision rule is a count: for each bid strategy you're about to set, count the conversions it would optimize toward in the last 30 days, at the funnel stage you actually want, and compare that count to the platform's floor. Do this per campaign, or per portfolio if campaigns share one.

  1. Pick the stage. Choose the earliest CRM stage sales vouches for, usually sales accepted or qualified. Google's guidance is to "avoid selecting goals from multiple stages" for optimization, so pick one.
  2. Count it per campaign. Pull 30 days of that stage by campaign from the CRM, matched on click ID or hashed email. For Meta's goal-based bidding, count it by week as well, since that floor is weekly.
  3. Place each campaign in a band. Clearly above the platform's floor means a target-based strategy. Near the floor means a hands-off strategy with a loose target or none, plus the controls in the next section. Below it means pooling first (the next step), then manual bids or a hard cap if the pool is still short.
  4. Pool where the goal is shared. A Google portfolio bid strategy groups multiple campaigns under one goal. We'd group thin campaigns that chase the same stage at a similar value into one portfolio so a single strategy has more conversions to work from, rather than five strategies starving separately.
  5. Write the band down with a review date. Volumes move, so recount every campaign on that date and move it between bands when its count crosses a floor.

Here's a hypothetical to show the rule at work. A company runs brand search, competitor search and a category campaign on Google and the same three on Microsoft. Its CRM shows 45 qualified leads a month from brand, 18 from category and 6 from competitor on Google, and roughly a fifth of those numbers on Microsoft. On Google, brand gets Target CPA on qualified leads, category and competitor share a portfolio Maximize conversions on that stage, and nothing on Microsoft clears 30, so Microsoft runs Enhanced CPC or Maximize clicks with a maximum CPC until the volume exists. Same account, three different answers, all from one count.

A railway track that passes a counting station and splits into three routes, standing for target-based, hands-off and pooled or manual bidding
Start by counting the qualified conversions in the last 30 days for the campaign or portfolio. If the count is clearly above the platform floor, use target-based bidding (Target CPA, cost per result goal, cost cap). If it is near the floor, use hands-off bidding with a loose target or none, plus value rules and caps. If it is below the floor, pool campaigns into a portfolio when they share a goal, and if the pool is still below, use manual CPC or a hard bid cap. The floors are 15 conversions at the stage on Google, 30 on Microsoft, 50 to 100 a week on Meta for goal-based bidding and 200 leads a month for CRM leads, and none published on LinkedIn.

Trust platform bidding when the signal is deep and the target is loose

Platform bidding earns your trust when it has volume above the floor, one honest conversion stage, and a target it can actually reach. The common mistake is setting the target you want instead of the one the data supports.

Start from what the platform has seen. Google's Target CPA page says the recommended target "is the average CPA from the last 30 days, adjusted for any conversion delays." Microsoft's Performance Max guidance is to start "with a less restrictive target amount" and wait "two weeks after each change before updating the amount again." Then tighten in steps.

Know which edits cost you learning. Google says changes to budgets, targets or targeting "never remove your historical data or trigger 'Learning' status," while a new strategy, a change to the campaigns inside a portfolio, or a significant change to the conversion goals does. That means you can move a Target CPA in small steps without restarting anything, but swapping the primary conversion action is a reset. Meta's significant edits behave the same way for its ad sets.

Frederick Vallaeys, CEO of Optmyzr, puts the condition behind all of this in a May 2026 post: "Smart Bidding optimizes for the signals it can see. That's both its greatest strength and its most fundamental limitation."

Constrain AI bidding with caps, value rules and exclusions

The controls built into each platform constrain AI bidding without switching it off, and they work better than daily edits to the target. Each one has a documented cost, so use the one that matches the problem.

ControlPlatformUse it forDocumented cost or limit
Bid limitsGoogle, portfolio Target CPA onlyA hard ceiling on cost per clickGoogle says bid limits "isn't recommended" for Target CPA or Target ROAS
Conversion value rulesGoogleRaising value for ICP audiences, regions or devicesOnly one rule applies per conversion
Seasonality adjustmentsGoogleA known 1 to 7 day change in conversion rate, such as an event"May not work as well" over long periods
Data exclusionsGoogleBroken tracking or a missed offline uploadNot to be used "frequently or for prolonged periods"
Bid capMetaA hard auction ceilingMay not spend the full budget
ROAS goal floorMetaA minimum return on value campaigns"Setting the benchmark too high may lead to under-delivery"
Cost cap or manual bidLinkedInCost control on lead gen and trafficLower caps deliver less of the budget
Maximum CPCMicrosoftA ceiling on Target CPA, Target ROAS or Maximize conversionsOptional; Microsoft still respects the budget

Two of these do the most for B2B. Conversion value rules let you raise the value of a conversion from an audience, such as a customer list of target accounts, and Google says the rules are "used in real time by Smart Bidding." That's how you tell Target ROAS that a lead from a named account is worth more than one from a free email domain. Data exclusions are the other: when the CRM upload fails for four days, an exclusion stops Smart Bidding learning from four days that look as if nobody converted.

Use caps with care. Vallaeys notes that bid caps on portfolio strategies help "in specific situations," mainly ROAS strategies, but that "many advertisers find that setting introduces new performance risks." A cap set below the going rate removes you from the auctions where clicks cost more, which in B2B search are often the ones you wanted.

Manual or rules-based bidding still wins on thin B2B campaigns

Manual bidding still wins where conversions are too rare for any platform to learn from, where you need a firm ceiling per keyword, or where a small ABM audience makes automated delivery overspend on the same people. These cases are common in B2B.

Vallaeys describes the first case in the same post: a manufacturer with "form submissions in the single digits most months," where "the algorithm has almost nothing to work with." His view on ceilings is just as direct: "The safest way to enforce a hard ceiling on what you'll pay per click, for a specific keyword, is still manual CPC." On LinkedIn, the maximum delivery page warns that with "a large budget and small audience size, we might show your ad to the same member more than once," which is exactly the shape of a 400-account ABM campaign. Manual bidding or a cost cap keeps that spend in check.

Manual no longer has to mean someone editing bids every morning. Rules-based bidding means you write the rules, such as lowering a keyword's bid when its cost per qualified lead runs 30% over target for 14 days, and let a script or an AI assistant apply them. Google Ads scripts can run that rule on a schedule, and an assistant connected to the account can propose the same changes for someone to approve. Keep a person on any rule that touches bids or budgets, for the reasons in our guide to humans in the loop on ad spend.

Feed better signals before you tighten anything

Better conversion signals improve AI bidding more than any target or cap, because every strategy above optimizes toward whatever you import. Before you tighten a target, check that the bidding is chasing a stage sales cares about.

Google's value-based bidding guidance gives the order: upload stage data "ideally on a daily basis," keep doing it for "at least 4 weeks or 3 conversion cycles" before switching to value-based bidding, and drop zero-value conversions rather than sending them, since "it's generally not recommended to use conversions with 0 values with value based bidding." The same page accepts "lead scores" as a value, which is where AI lead scoring meets bidding: a score turned into a value gives Maximize conversion value a reason to pay more for one lead than another.

On LinkedIn, sending qualified leads through the Conversions API switches the recommended goal to qualified leads, and with it the bidding to maximum delivery only. That trade is worth making when the daily budget is a number you're happy to see fully spent, and worth delaying when it isn't.

Value-based bidding pays for pipeline value instead of lead counts

Value-based bidding in Google Ads tells Smart Bidding what each conversion is worth, so it can pay more for a lead likely to become pipeline and less for one that won't. Target CPA treats every qualified lead as equal. Google's lead quality guidance states the choice directly: "If you can assign different values to your leads, use value-based bidding," which means Maximize conversion value, with or without a target ROAS, and Target CPA on qualified or converted leads when you can't.

The values come from your CRM stage rates. Take a hypothetical team with an average deal of $40,000 and a 25% win rate from opportunity, so an opportunity is worth $10,000. If 40% of its qualified leads become opportunities, a qualified lead is worth $4,000. Bid on that one stage, since Google advises against optimizing to goals from several stages, and let the value vary inside it by what predicts pipeline. Google's value-based bidding page notes that "Particular answers on lead forms may indicate higher quality leads," so a company-size answer inside your ICP can carry more value than one outside it, and the value rules above add the audience, location and device adjustments.

Target ROAS is then the conversion value you want back for each dollar. If that $4,000 qualified lead is worth paying up to $1,000 for, the target is 400%, because $4,000 divided by $1,000 is 4. Google's Target ROAS page asks you to report values across all relevant campaigns "for 4 weeks or 1-2 conversion cycles (whichever is longer)" before you set the target, and after a target change it says the bidder "will need some time to hit the new target" and to give it 1 to 2 conversion cycles. So judge a new target after one to two conversion cycles, and judge it on cost per opportunity in the CRM rather than on platform ROAS.

AI bidding goes wrong in B2B in a few predictable ways

Most failures of AI bidding in B2B accounts come from the setup around it, and each one has a specific check.

  • The hands-off strategy spends the whole budget. Google's Maximize conversions page says the strategy tries to spend your full average daily budget. If the budget was a ceiling you never expected to reach, reset it before switching.
  • The goal changes quietly. Swapping the conversion action a strategy optimizes toward is the learning reset described above, so agree goal changes in advance and make them once.
  • Tracking breaks and nobody excludes the gap. A failed upload looks like a conversion drought to the bidding. Add a data exclusion the day you notice.
  • The verdict comes too early. Judging a change in week one, on a stage that takes three weeks to appear, measures the conversion delay rather than the strategy.

Judge a bidding change on CRM cost per qualified stage

A bidding change worked when cost per qualified stage in the CRM improves over at least two conversion cycles without the volume of that stage falling. Platform-reported CPA is the first signal and the least reliable one, because it measures what the platform can see.

Google advises judging Smart Bidding after one to two conversion cycles, and Microsoft's Performance Max advice to wait two weeks between target changes sets a similar pace. Where you can, test the change in a Google Ads custom experiment, which runs a changed copy of the campaign against the original and splits the budget between them. Then compare cost per opportunity by arm in the CRM, since a lower CPA that produces fewer opportunities is a worse result that looks better in the platform.

Choose bidding campaign by campaign

Take your current campaign list, add one column for qualified conversions in the last 30 days and one for the platform's floor, and the right bid strategy for most rows will be obvious. If you'd like help setting up the counts, the value rules and the review routine around how your team works, Scalewell does this with paid teams.

Frequently asked questions

Can my own AI assistant set bids instead of the platform?

No assistant can bid at auction time, because only the platform sees the signals for each auction. What an assistant connected through an API or an MCP server can do is change targets, caps and manual bids between auctions, following rules you write. That makes it a good tool for rules-based bidding and a poor replacement for Smart Bidding on high-volume campaigns.

What replaced Enhanced CPC in Google Ads?

Google removed Enhanced CPC from Search and Display campaigns the week of March 31, 2025. Campaigns that need control without automation now run manual CPC, and the automated route is Maximize conversions or Target CPA. Microsoft Advertising still offers Enhanced CPC, so imported campaigns can behave differently on the two platforms.

Should B2B lead generation use Target ROAS?

Target ROAS fits B2B lead generation once you send values for lead stages, such as a higher value for a sales-accepted lead than a raw form fill. Google's Search campaigns need at least 15 conversions in the past 30 days to use it, and Google's value-based bidding guidance advises against sending zero-value conversions. Without values, Target CPA on a qualified stage is the simpler choice.

Does LinkedIn have automated bidding?

LinkedIn has automated bidding: its maximum delivery strategy uses machine learning to bid automatically while aiming to spend the full budget, and it is available for every objective. Cost cap gives automated bidding with a cost limit on some objectives, and manual bidding remains available everywhere. The qualified leads goal works only with maximum delivery.