Broad Match With Smart Bidding When You Get 15 Conversions a Month: A Clinic Test Design

Google advises pairing broad match with Smart Bidding. Most clinics do not have the conversion volume that advice assumes. A single-location clinic spending a modest budget might record 15 appointment requests a month, split across several treatments. Is broad match with target CPA a sensible experiment at that volume, or an expensive way to buy irrelevant clicks?

This article is for clinic owners and the marketers who manage their Google Ads. It does not argue that broad match is good or bad. It sets out a staged test with stop rules, uses micro-conversions as bridging signals, and explains how to read the results honestly. We have no result data to report here; the value is in the design. If your website converts poorly, fix that first with our website conversion guide and landing page tips, because no bidding strategy compensates for a page that does not convert.

What Google says, and what it leaves open

Google's keyword matching options documentation describes broad match as the most comprehensive match type, able to show ads on related searches that do not contain the keyword's literal wording, and says Google recommends pairing it with Smart Bidding so that bids reflect auction-time signals. Its broad match setting page states that the setting requires a campaign to use a conversion-based or conversion-value-based Smart Bidding strategy.

Google's guidance for value-based bidding gives one volume reference: for Target ROAS on Search, at least 15 conversions in the past 30 days at the conversion-tracking level (Google Ads Help). That is a minimum for one strategy, not a guarantee that broad match will behave well. The documentation does not promise results at the minimum, and neither can we.

So the question is empirical. With thin data the model has less to learn from, and broad match widens the set of queries it must judge. The risk is that spend drifts to irrelevant queries before the model has enough signal. The test below limits that risk.

The principle: protect the control, cap the exposure

A responsible test for a clinic with 15 conversions a month follows four rules.

1. Do not change the core campaign. The campaign that pays the bills continues as is.
2. Cap the test budget at an amount the clinic can lose without a problem, for example a fixed slice of monthly spend.
3. Run the same offer on both arms so the comparison is of match type and bidding, not of landing page.
4. Define stop rules before starting, so nobody argues for continuing a bad test.

Google Ads campaign experiments allow a traffic split between a base campaign and a trial. Use that rather than comparing before and after, which confuses the test with seasonality.

Step 1: Audit the conversion signal

Broad match amplifies whatever you reward. If the conversion is a form fill that includes spam, wrong-number calls and job seekers, the model will find more of them.

Check:

• Is the primary conversion a genuine appointment request or qualified call?
• Are call conversions set with a sensible minimum duration, and are calls tied to the ad click?
• Are duplicates removed? A thank-you page that reloads will count twice.
• Is Google's tag firing correctly across forms, WhatsApp clicks and call buttons?

Review last month's conversions manually. If a third of them are junk, fix the tracking and filtering before testing anything. Our offline conversion tracking guide explains how to bring the quality of the lead back into Google Ads.

Step 2: Build a bridge with micro-conversions

At 15 conversions a month, the model sees a conversion about every two days. That is sparse. Micro-conversions add intermediate signals that occur more often and correlate with eventual appointments.

Candidates for a clinic:

• click-to-call taps on the call button;
• WhatsApp button clicks;
• booking-widget starts (the first step of the form);
• clicks to the directions or location page;
• a view of the doctor profile combined with scroll depth;
• a form field completed, if the form is multi-step.

Do not turn all of them into primary conversions. A page view or a scroll is a weak signal that can pull the model toward curious browsers. Use the following rule: a micro-conversion becomes primary only if you have evidence it predicts appointments.

How to test whether a micro-conversion predicts appointments

For the last 90 days, join ad-click records to your appointment data. For each candidate micro-conversion, calculate how often a click that produced it later became an appointment, compared with a click that did not. A signal is worth using if clicks that include it become appointments clearly more often than those that don't. If you cannot join the data, start with the fewest candidates (call taps and booking-form starts) and verify on the later results.

Mark the genuine appointment as the primary conversion. Mark validated micro-conversions as primary only if the bidding must use them, and keep the weaker ones secondary for observation. Google lets you control which conversion actions count toward bidding.

If you assign values, keep micro-conversions much lower than the appointment, and keep them on a much smaller scale than the appointment itself, because a model that sees inflated micro-values will chase clicks rather than appointments.

Step 3: The staged test design

Stage 0: Baseline, two to four weeks

Record in the existing campaign: spend, clicks, search terms, primary conversions, qualified appointments, cost per qualified appointment, and the share of spend on irrelevant search terms from the search terms report. This is your control.

Stage 1: Tighten negatives and structure

Before enabling broad match, load shared negative lists: job-related terms, free-treatment terms, competitor names you don't want, and unrelated procedures. Separate brand terms into their own campaign so broad match does not feed on your name.

Stage 2: Smart Bidding on exact and phrase, two to four weeks

If the base campaign uses manual bidding, move it to a conversion-based strategy first, such as maximise conversions with a cap, keeping match types unchanged. This isolates the effect of the bidding change before adding the broad match variable. Many accounts skip this step and then cannot say which change caused the result.

Stage 3: Broad match trial arm

Create an experiment arm with a small share of traffic. Add a limited set of broad match keywords drawn from your strongest themes, such as your core treatment and location. Use the same ads, landing pages and negative lists. Keep the daily budget capped.

Use target CPA only if the account meets enough conversion history. Otherwise, use maximise conversions with a cost-per-click cap and move to target CPA when volume allows. Set any target near current performance, not below it. An aggressive target on a thin-data campaign can throttle delivery to nothing.

Stage 4: Read the results

Run the trial for at least four to six weeks, or two full conversion cycles for treatments with long decision times. Judge it on the metric the clinic cares about, which is cost per qualified appointment, and not on cost per raw conversion. Review the search terms report weekly during the test.

Stop rules, defined in advance

Write these down before launch.

• Hard cap: if spend in the trial arm reaches the agreed cap without a qualified appointment, stop.
• Query quality: if, in a weekly review of the trial arm's search terms, a clear majority are irrelevant after negatives are added, pause and tighten before continuing.
• Lead quality: if the front desk reports a rise in unqualified calls or enquiries from the trial arm, stop and review.
• Delivery collapse: if impressions fall sharply because the target is too tight, loosen the target once; if it repeats, end the test.
• Compliance: if the trial surfaces ads against queries that suggest the wrong intent for a sensitive treatment, add negatives and review ad copy before continuing. Google's health-related advertising policies and your local advertising rules still apply.

How to read the outcome fairly

With 15 conversions a month, the difference between arms in a four-week test is likely to be small in absolute terms. A swing of two or three appointments is within normal variation. So treat the test as a decision aid, not proof.

Ask these questions in order:

1. Did the trial arm produce qualified appointments at a cost the clinic can accept?
2. Did it find relevant queries the control did not? Export the new search terms and list the useful ones. These can be added as phrase or exact keywords to the control campaign, which is often the main value of the test.
3. Did the extra volume come with proportionally poorer quality? Compare the front desk's view with the data.
4. Would the result survive another month?

If the answer to the first is no, return to the control. If the answer to the second is yes, harvest the terms. Neither outcome is a failure. Both give the clinic evidence about its own account that no benchmark can supply.

Failure modes

Testing broad match on junk conversions. The model will optimise efficiently toward the wrong thing.

Changing too many variables. Landing page, bid strategy and match type should not all move at once.

Judging by platform conversions only. Always reconcile with appointments recorded in the clinic's system.

Letting the test run unattended. Check search terms weekly.

Over-reading a short window. Treatment decisions have lead times. A dental whitening enquiry converts quickly, an orthopaedic consultation takes longer.

For help with audit and test design, the Google Ads audit service describes our approach.

Where to start

• Reconcile last month's Google Ads conversions with the clinic's appointment records.
• List candidate micro-conversions and validate two against 90 days of data.
• Build negative lists and separate brand from non-brand campaigns.
• Set a trial cap, a run length and written stop rules.
• Launch a small experiment arm with matching ads and landing pages.
• Review search terms weekly, and review qualified appointments at the end.
• Move useful new queries into the control as exact or phrase keywords, whatever the overall result.