How a Sydney Business Law Firm Stopped Training Google to Find Bad Leads

About Our Client

A Sydney-based commercial law firm serving business clients across Australia. Their work spans trademarks and intellectual property, commercial leasing and property, franchising, and general business and commercial law.

Problem/Goal

The firm came to us in mid 2024 with a Google Ads account that had stopped working. Their previous Google Ads manager had made a series of changes that dropped their ad rank sharply, and lead flow had fallen to almost nothing. Before proposing anything, we asked for account access and recorded a full diagnostic review.

The audit found problems beneath the immediate one:

• Conversion tracking that could not be trusted - duplicate tags and unverified phone call tracking distorted every downstream number.
• An automated bidding strategy running on that unreliable data - Google was optimising toward signals the firm could not verify.
• No negative keyword discipline - budget leaked to irrelevant searches with nothing to filter them out.
• Spend spread across a wide range of practice areas - no evidence about which produced work worth having.

Underneath all of it sat the real problem, and it is the one almost every underperforming account shares. The firm was telling Google that a form submission was success. Nobody could say which practice areas were profitable, because the tracking had never been good enough to tell them.

What We Did

1. Rebuilt the measurement layer before touching the campaigns

We cleaned up duplicate tags in Google Tag Manager, verified phone call conversions were recording correctly, and moved the account off automated bidding onto manual control until there was reliable data to optimise against. We added click fraud prevention and began systematic negative keyword work.

Optimising against numbers you cannot trust simply scales the error.

2. Changed the signal Google was optimising toward

This was the turning point, and it is the part most accounts never address.

Google's bidding algorithms optimise toward whatever you define as a conversion. If you tell Google that a form fill is a win, it will go and find you more form fills, and it will find them progressively cheaper, because cheap is what optimisation rewards. Cheap form fills are rarely good ones. It does not take many junk conversions before the algorithm concludes it is succeeding and doubles down on the traffic producing them.

That is exactly what had happened here. The account was reporting healthy conversion numbers while the firm's intake staff were fielding enquiries from people who had reached the wrong sort of firm entirely.

So we changed what counted as success:

• We installed dedicated lead tracking connecting every enquiry to what happened next, and connected it to the ad platforms.
• We trained the firm's intake staff to classify every single enquiry as it came in, and we drew a hard distinction between genuine spam and a real person who simply was not a fit. Those are different problems with different fixes, and collapsing them into one bucket hides both.
• We captured quote values, not just lead counts. Once the system recorded that a single franchise enquiry had produced a $12,200 quote, the difference between a valuable enquiry and a cheap one stopped being a matter of opinion.

The goal throughout was simple to state and hard to do: give Google a definition of success that matches the firm's definition of success.

3. Pulled back Performance Max when it scaled the wrong signal

Performance Max was running when we inherited the account, and it demonstrated the problem at full volume. Monthly spend had roughly tripled, from around $6,000 to $16,000, while lead quality fell sharply over the same period.

Because the campaign was optimising toward form fills, and because Performance Max has far more freedom than search to decide where ads appear, it had wandered a long way from the firm's actual market. Enquiries arrived from people who thought they had reached legal aid, Aboriginal Legal Services, a tribunal, a migration agent, and several rival firms whose brand names Google had decided were close enough.

We wound Performance Max back to a single campaign covering the firm's highest value work, and returned the budget to search campaigns where targeting could be controlled directly. Performance Max is not the villain here. Running it on an unreliable success signal is.

4. Tested cheaply outside the core, and committed inside it

The firm's commercial work was where the evidence pointed early, so that is where the budget went. The practice areas outside commercial law were trialled rather than committed to.

Criminal law and wills and estates together took under 3% of total account spend across their entire life. Each ran long enough to give a directional signal and no longer. When the early numbers were poor and the enquiries were not turning into the work the firm wanted, they were paused rather than propped up.

That is a deliberate trade, and worth being honest about. A small, brief test will not tell you whether a practice area could work with a serious budget and a purpose-built funnel behind it. It tells you whether it is obviously working now. When the alternative is putting the same money into a line that is already producing, that is usually the more useful question.

5. Cut what the evidence did not support, in stages

With trustworthy signal in place, the picture emerged over the following year.

Wills and estates ran for nine months before we paused it in April 2025 on consistently low conversions. In New South Wales it was costing $508 per enquiry. That budget moved to business and commercial work the same month.

Criminal law was tested and wound back, ending at $665 per enquiry in the general campaign, more than three times the cost of the firm's best performing line.

Here is the whole account laid out, which is the clearest way to see how the decisions were made:

Cost per enquiry alone did not drive these decisions, and the table shows why. Two of the campaigns we cut were technically cheaper per enquiry than one we kept. What separated them was volume and what happened after the enquiry arrived. The criminal and wills lines never reached meaningful scale, and the enquiries they did produce were not converting into matters the firm wanted. The commercial lines produced real volume and real work.

This is only visible if you are tracking what happens after the form is submitted. On lead count alone, several of these decisions would have gone the other way.

None of these were day one decisions. Each followed a live test and data good enough to be trusted.

6. Cut by matter type, not just by practice area

Franchising was profitable overall, but not uniformly. Franchise dispute enquiries converted poorly and cost more, while franchise establishment enquiries were exactly the work the firm wanted.

Rather than pause franchising, we added qualification questions to the landing page forms so dispute enquiries were filtered before they ever reached a solicitor. We applied the same thinking to intellectual property, prioritising trademark work over lower value copyright enquiries.

Enquiry quality is decided before the phone rings.

Overall Impact

The account generated 1,183 enquiries from $220,907 in managed ad spend, at an average cost per enquiry of $186.74 and a conversion rate of 8.92%.

More useful than the total is where it ended up coming from. The four scaled commercial practice areas, plus brand, produced roughly 90% of all campaign enquiries. Trademarks became the strongest single line at 307 enquiries and $204 each, followed by commercial leasing and property at 265 enquiries and $232 each.

The two worst performing lines, general criminal at $665 per enquiry and New South Wales wills and estates at $508, were running at two to three times the cost of the firm's strongest commercial work while producing almost no volume.

The lesson generalises well beyond this firm. Most agencies optimise toward the lead, because the lead is the last thing they can see. That is precisely why so many accounts drift: the algorithm is being trained, every single day, on a definition of success that nobody in the business would recognise. Fix the signal first. Then the decisions about what to cut make themselves.