AI bidding does not remove management from law firm advertising. It moves the most consequential management decisions upstream.
When software can evaluate each auction and adjust bids automatically, the human team spends less time choosing individual bids. It becomes more responsible for deciding which business outcome the system sees, how values are assigned, which constraints apply, how delayed intake outcomes return, and whether the resulting matters make sense for the firm.
That is why “we use AI” is a weak agency claim. The important question is: What behavior is the system being rewarded for, and what evidence shows that behavior helps the firm?
If every form submission counts as a valuable conversion, automation can find more form submissions. It cannot know that half are duplicate, outside the practice, or never reached by intake unless those distinctions enter the feedback loop.
Separate Smart Bidding, Performance Max, and AI Max
These names describe related but different parts of Google Ads.
- Smart Bidding is Google’s set of auction-time automated bid strategies that optimize for conversions or conversion value. It can be used in campaign types beyond Performance Max.
- Performance Max is a goal-based campaign type that can serve across Google inventory including Search, YouTube, Display, Discover, Gmail, and Maps. It uses automated bidding along with assets, audience signals, URLs, search themes, exclusions, and other controls.
- AI Max for Search campaigns is an optimization layer with targeting and creative features inside Search campaigns. It is not another name for all automated bidding and it is not Performance Max.
The distinctions matter because the controls, inventory, creative requirements, and diagnostic questions differ. “The algorithm handled it” does not identify which system ran, what settings were active, or what data influenced the decision.
Google’s current Smart Bidding overview defines the conversion and conversion-value objective. Its Performance Max overview describes the broader campaign inventory and controls. Account features change, so the manager should show the actual campaign rather than rely on a label from a proposal.
The optimization target is a policy decision
Before choosing a bid strategy, write the outcome ladder:
raw contact → valid inquiry → qualified inquiry → attorney review or consultation → retained client → collected fee/contribution
Then decide which stage is:
- available quickly enough to guide bidding;
- recorded consistently enough to trust;
- frequent enough to provide useful feedback;
- connected strongly enough to the work the firm wants; and
- appropriate to send or represent in the platform.
No stage wins every tradeoff.
Scroll sideways to review every column.Each row is shown as a labeled card.
| Signal | Strength | Failure risk |
|---|---|---|
| Form or call | Fast and relatively frequent | Spam, duplicates, wrong matters, tracking noise |
| Valid inquiry | Removes obvious noise | Still may not fit the practice |
| Qualified inquiry | Reflects matter and market fit | Depends on timely, consistent intake judgment |
| Retained client | Reaches the business outcome | Lower volume, longer delay, definition may differ across teams |
| Conversion value | Can distinguish expected economic value | Bad assumptions become machine-readable confidence |
| Collected contribution | Closest to realized economics | Often arrives too late and too sparsely for daily bidding |
The best signal is not always the deepest. A reliably defined qualified inquiry may guide a campaign better than a retained-client event that arrives months late and is updated inconsistently. Use later outcomes to audit and recalibrate the earlier proxy.
Audit the conversion portfolio campaign by campaign
The paid-media metric chain defines stronger outcome stages, and the attribution framework keeps one matter from becoming several machine-reported successes.
In Google Ads, conversion actions can be primary or secondary. Google’s conversion-goal documentation explains that primary actions are used for bidding when their goal is selected, while secondary actions are generally observational; custom-goal use is a relevant exception.
That means the audit needs more than a screenshot of the “primary” label. For each campaign, record:
Scroll sideways to review every column.Each row is shown as a labeled card.
| Control | What to inspect |
|---|---|
| Campaign objective | Conversion count, conversion value, traffic, or another configured objective |
| Selected goals | Account-default or campaign-specific goals actually applied |
| Conversion actions | Name, source, category, counting method, primary/secondary role, value, and window |
| Duplicate paths | The same form or call recorded through several tags, analytics imports, or call systems |
| Stage meaning | Whether “lead,” “qualified,” or “sale” matches the firm’s documented definition |
| Data return | Which intake or CRM changes are imported, how often, with what diagnostics |
| Bid strategy | Maximize conversions/value, target CPA/ROAS, or another strategy and its actual target |
| Major changes | Date, owner, reason, and expected effect on delivery or learning |
One form might fire a native Google Ads tag and arrive again through an analytics import. If both are bidding goals, the system can treat one prospective matter as two successes. One call action might count every call over a duration threshold even though intake later finds many unsuitable. A “secondary” qualified event may never influence bidding while the raw contact remains primary.
This is why AI bidding starts with data governance.
Assign values from expected economics, not enthusiasm
Ground assigned values in retained-client cost and expected fee value, then audit whether ROAS is hiding forecast, cost, or timing assumptions.
Value-based bidding gives the system a way to prefer some outcomes over others. The value needs a defensible relationship to the firm’s economics.
Suppose a qualified inquiry in a particular matter group has:
- a 20% probability of becoming a retained client;
- $7,500 in expected collected fees if retained; and
- $3,000 in expected variable delivery cost if retained.
Expected contribution before acquisition is $7,500 − $3,000 = $4,500 per retained client.
At the qualified-inquiry stage, expected contribution before acquisition is $4,500 × 20% = $900.
That $900 is a hypothetical estimate, not money collected and not a recommended conversion value. It depends on a comparable mature cohort, stable definitions, and the assumption that future matters resemble the historical group.
Now compare a second matter group with $12,000 expected fees, $8,000 delivery cost, and a 10% retention probability. Expected contribution per qualified inquiry is ($12,000 − $8,000) × 10% = $400. The larger expected fee does not make the inquiry more valuable after cost and probability.
Google’s conversion-value guidance explains how values affect optimization. The firm still owns the meaning. Keep the model version, inputs, sample, effective date, and any caps or ranges used to avoid a single speculative matter dominating the signal.
Treat outcome delay as part of the control system
There are two delays:
- the time from the ad interaction to qualification, retention, or collection; and
- the time from that event to its import and availability in the platform.
A campaign optimized toward retained clients can appear to deteriorate if the newest weeks contain many unresolved inquiries. The same campaign can look unusually strong when a backlog of older outcomes arrives at once.
Use inquiry cohorts and report their age. Keep operational monitoring separate from outcome judgment:
- Broken forms, rejected ads, incorrect locations, overspend risk, and lost call routing require immediate attention.
- Bid performance against retained outcomes requires enough maturity to interpret.
Google’s guidance on how bidding algorithms learn says a conversion cycle includes both the time from click to conversion and, for imported outcomes, the time until the conversion is reported. It also says changing a target does not itself trigger a learning status or erase what Smart Bidding has learned. Avoid universal claims that every edit “resets learning” for a fixed number of days. Record the actual change, the account status, the observed delay, and current platform guidance.
Design a test around a business failure
“Turn on Smart Bidding” is not a test. Name the failure the change is supposed to address.
Consider a hypothetical search campaign that currently bids toward all calls over a duration threshold. Intake later classifies 45% of those calls as qualified. The firm proposes using reliably returned qualified-inquiry outcomes instead.
The test question is:
Can bidding toward qualified inquiries increase the share and number of qualified opportunities without exceeding the firm’s fully loaded cost per retained client?
The evaluation plan needs:
- the campaigns and geography in scope;
- the old and new goal configuration;
- baseline validity, qualification, retention, and cost definitions;
- data-return timing and rejected-import checks;
- media and loss limits;
- major concurrent changes in creative, page, schedule, intake, or market conditions;
- the earliest date when the relevant cohort can be judged; and
- scale, hold, repair, or reversal conditions.
A lower platform CPA is not sufficient if the conversion action changed from “qualified inquiry” back to “all calls.” A higher platform CPA may be acceptable if the campaign acquires more retained matters with better contribution. Preserve both platform and firm outcomes.
Where volume and platform tools support a formal experiment, use it. A before-and-after comparison can still help manage the account, but it cannot isolate bidding as the cause when seasonality, competition, pages, intake staffing, or matter mix also changed.
Keep four human controls in place
1. Business control
The owner or practice leader decides which work the firm wants, what it can serve, the economic ceiling, and the cash risk it will accept.
2. Campaign control
The paid media team configures goals, bidding, budgets, locations, queries or inventory controls, ads, assets, and change logs. It also monitors whether automated expansion is still aligned with the charter.
3. Intake control
The firm records validity, qualification, outcomes, and loss reasons promptly and consistently. Nick Cohen described this dependency in a transcript-verified Non-Billable Hour discussion: “I need you to tell me who’s qualified, who’s not.” He was speaking as a guest; the audio and episode publication date were not independently verified.
4. Governance control
Someone reconciles conversion actions with source records, tests the public journey, reviews permissions and data use, and confirms that the values and definitions remain current.
AI can process the signals. It cannot take responsibility for those choices.
Watch for failure modes that look like success
- Cheaper conversions, worse matters: raw contacts improve while qualification and retention decline.
- More reported value, no new cash: forecast or assigned values rise without mature collections.
- Stable dashboard, broken public journey: account settings are correct but the displayed number, form, or route fails.
- Fast learning, wrong lesson: a duplicate or easy micro-conversion dominates the bidding goal.
- Better average, lower useful volume: the campaign narrows to a small set of easy wins and total contribution falls.
- Automation masks capacity: inquiries rise beyond intake or attorney-review capacity, reducing downstream performance.
- One high-value matter distorts the model: a speculative value dominates a small cohort and redirects spend.
Each failure has a different response. The useful manager traces it to the goal, data, campaign, page, intake, economics, or capacity before changing bids again.
Evaluate the manager by the decisions around the algorithm
Juris Digital’s law firm PPC and paid media service includes campaign management, landing-page and intake feedback, retained-client measurement, and budget economics. If AI bidding is the issue, bring the campaign’s goal configuration, conversion-action list, several months of outcome timing, qualification definitions, and an anonymized mature cohort.
Ask us to show which event and value currently influence bidding, where the signal loses its relationship to retained work, and which human control is missing. The first useful deliverable may be a corrected goal map or outcome-feedback loop rather than a new automated strategy.
The algorithm should be the fastest participant in a system the firm understands. It should not be the only participant anyone can blame.
