Smart Bidding Only Knows What You Feed It

Smart Bidding prices every auction against the conversion values you report, and nothing else counts as value to it. Report a form fill and it buys form fills. Report closed revenue out of your CRM and it starts buying the searches that produced closed revenue, provided that revenue arrives fast enough to be usable, which for plenty of lead gen businesses it does not. Two accounts running the same bid strategy at the same target end up somewhere completely different depending on what is sitting in the Conversions column, and Google's documentation is unusually plain about the windows and delays that decide whether a true value can get into that column at all.

I am Jared, and I spend most of my week inside enterprise search accounts. What I keep coming back to is which searches the machine has been taught to want, and the answer to that sits in a spreadsheet somebody set up once and never revisited.

What does the bidder actually see when it prices an auction?

Smart Bidding is Google's family of auction time bid strategies that sets an individual bid for every auction. Value based bidding is the subset that optimizes toward conversion value: Maximize conversion value, with or without a Target ROAS. The other half, Maximize conversions and Target CPA, optimizes toward volume and treats a $400 lead and a $40,000 lead as the same event. In June 2026 Google renamed the strategies to make that split more obvious, so "Maximize conversions with a Target CPA" is now just "Target CPA."

What the bidder reads is the Conversions column, and only that column. Google's offline import FAQ puts it in one sentence: Smart Bidding optimizes for whichever conversion types you include there. Everything else in the account is reporting.

It is good at the job it has. The same documentation describes the bidder building conversion rate models at the level of a search query mapped to specific ad copy and landing pages, refining them with Bayesian learning as data accrues. It also applies what Google calls adaptive historical weighting, leaning on recent data but discounting it when your conversion cycle is long, so a car dealership's last two weeks count for less than a food delivery app's.

All of that machinery predicts whether a conversion will happen. Whether it is worth anything is a number you supply.

The seven day line is the part nobody reads

Buried in Google's value based bidding guidance is the sentence that decides whether a CRM feed is viable at all. Shorter conversion delays, under seven days, are recommended. And if it takes seven or more days for all of your conversions to land in Google Ads, the ramp up period for value based bidding "may take several months."

Put that next to how long deals actually take. Vendor compiled benchmarks published in March 2026 put the average B2B sales cycle around six and a half months, up from just under five in 2019, with SaaS deals under $15,000 in annual contract value closing inside a month and anything over $100,000 running ninety to a hundred eighty days. Those come from consultancies aggregating other consultancies, so take them as shape rather than measurement.

Google's hard windows make the collision concrete. The Google Click ID, the GCLID, is retained for ninety days and a conversion older than that cannot be imported against it. Matching on hashed customer data through enhanced conversions for leads gives you sixty three days. You can restate the value of a conversion you already reported for fifty five days, after which the number you sent is the number the bidder believes. Uploads should run daily, and Google warns specifically against backfilling a pile of historical values at once.

Google's own answer to the delay problem is the interesting part, because it tells you to move earlier. The final purchase might carry the most accurate values, the documentation says, and if the delay is too long to manage, pick a goal earlier in the journey, like a qualified lead. Reasonable advice, and also an admission: the recommended configuration for a long cycle business is a proxy.

Volume pushes the same direction. Target ROAS wants at least fifteen conversions in the last thirty days to be eligible, practitioners generally want thirty to fifty before trusting it, and the closed won action is by construction the thinnest action in a lead gen account. Split that account into narrow campaigns by program and region and you end up with a conversion goal that is true and statistically invisible.

You can adjust value by device and by state, and not by the search that started it

Conversion value rules are the feature Google gives you for expressing value the platform cannot infer. They adjust reported conversion value and Smart Bidding optimization in real time at auction, and Target ROAS and Maximize conversion value read them directly.

Here is the complete list of conditions a value rule can key on, from Google's own table: audiences, geographic locations, device, and a no condition option that exists only for store visit and store sales actions. Each rule takes a primary and a secondary condition. That is it.

The search query is not on that list. You can tell Google that a lead from Colorado is worth double, that a returning customer list is worth twenty percent more, that desktop converts at half the value of mobile. You cannot tell it that the query "emergency" plus your product closes at four times the rate of the query "cheap" plus your product, which in most accounts I have worked in is the largest single source of value variance in the system. Google's documentation even undercuts the dimensions it does offer, with a note right under that table saying Smart Bidding already uses geography, device and first party audience lists, and already accounts for a segment that converts better when your reporting shows it.

The size of this hole is easy to overstate, so let me be exact. The bidder does model at query level, Google says so, and what it models there is conversion rate. What it has no mechanism to receive is a query level statement of worth. Two queries can sit at an identical conversion rate while producing customers whose lifetime value differs by an order of magnitude.

What Google shipped in 2026, and what it leaves alone

Journey aware bidding was announced at Google Marketing Live in May and is in beta for Search campaigns on Target CPA. It lets the bidder learn from every conversion goal in the lead journey, biddable or not, so form fills and phone calls inform a strategy optimizing toward qualified leads. That is a real improvement in how much of your funnel the model can see, and it is bolted to Target CPA, which is volume optimization, so it puts no dollar figure from your CRM near the bid.

Demand led budget pacing came out of the same event, still described in May as rolling out over the following months. Google AI shifts spend toward days it predicts higher demand and pulls back on slower ones, inside your daily and monthly limits. Demand prediction and value are independent of each other, though, and a peak day of cheap leads spends the month faster.

The change with a date attached is the plumbing. From June 15, 2026, offline conversion imports and enhanced conversions for leads uploads migrate to the Data Manager API and are blocked in the Google Ads API, and developer tokens that sent no request between January and June 2026 do not get allowlisted for legacy access. If your value pipeline is a script somebody wrote in 2023, this is the year it stops.

The honest case for leaving all of this alone

The argument against CRM fed values is stronger than most agency content admits, so here it is straight. Google is pooling query behavior across the entire auction while your CRM holds a few thousand rows. The delay math favors the proxy by Google's own thresholds, the closed won action is usually too thin to bid on, and the dimensions you would adjust on are already in the model. Consent mode modeling fills part of the observation gap for free and flows straight into bidding, with Google's worked example showing a fifty percent consent rate producing only a nineteen percent drop in observed conversions. A value pipeline is also a liability that breaks silently, because the bidder keeps optimizing confidently against stale numbers.

Most of that holds. Every one of those points is about predicting whether a conversion happens, though, and the question is what it is worth. The case concedes the argument in passing too, since Google recommends the earlier funnel stage precisely because the true value cannot arrive in time.

What I look at before touching any of it

Which conversion actions are marked primary, first, because stacking a lead at $10, a sales qualified lead at $20 and a closed sale at $50 as three primary actions teaches the bidder that one sale is worth $80. Then the arrival pattern rather than the arrival: daily uploads, no backfilling, nothing sitting in a monthly batch. Then whether the values are two or more genuinely different numbers, since Google requires differentiated values and accepts a lead score as a legitimate proxy when revenue is out of reach. Then the part belonging to the query map rather than the feed, which is whether the queries producing your best closed deals are the queries the bidder pays most for. A search terms export and a CRM pull answer that in an afternoon, and it will not agree with the account.

If a proxy is all you can get to Google inside the window, use the proxy, and spend the effort you saved on making that proxy correlate with revenue. That is a measurement design problem more than a bidding problem, which is how most of our conversion tracking and measurement work starts, and usually the first thing a paid search account audit turns up.

FAQ

Can Smart Bidding use conversion values from my CRM? Yes, through offline conversion imports or the upgraded enhanced conversions for leads, uploaded via Google Ads Data Manager. Match on GCLID within ninety days of the click, or on hashed customer data within sixty three days. From June 15, 2026 those uploads run through the Data Manager API rather than the Google Ads API.

What is the best conversion value to send for lead gen? Whichever true or proxy value you can deliver inside roughly seven days of the click, consistently, at enough volume to be eligible. Target ROAS asks for fifteen conversions in thirty days, and a qualified lead score arriving in two days usually beats closed won revenue arriving in ninety.

Do conversion value rules work at the keyword or query level? No. Value rules key on audiences, geographic location and device, with a primary and a secondary condition per rule. Query level value has to be handled through account structure, negatives and the values themselves.

Are modeled conversions part of what the bidder optimizes toward? Yes. Modeled conversions from consent mode land in the Conversions column and the Conversion value column and flow into bidding. Eligibility needs roughly 700 ad clicks over seven days per country and domain grouping.