Arellano Construction builds decks in Chattanooga — custom, composite, covered, screened porches, pergolas. A typical job runs from the low twenty thousands into the forties, and a big composite build goes higher. We have handled their website, search, and Google Ads for about two and a half years, which is long enough to have been wrong about the account in several different ways and to have the data to prove it.
This is a piece about what a small-market Google Ads account actually tells you if you read it carefully, and why the two obvious moves — cut the expensive stuff, or spend more — were both the wrong lesson.
What the account looked like
For most of this year the campaign was dominated by one broad-match keyword, a phrase about deck remodeling contractors. It took roughly two-thirds of the spend and produced leads at around $800 apiece. Alongside it sat a handful of exact-match keywords — "deck builders chattanooga" and "deck builder near me" chief among them — that spent a tenth of the budget and converted at about $130.
Put the whole year's keywords side by side and the picture is stark. Exact match turned 101 clicks into 12 leads, a conversion rate just under 12 percent. Everything else — broad and phrase together — turned 1,305 clicks into 18 leads, about 1.4 percent. That is not a small difference. It is an order of magnitude, on enough clicks to be real, and it is the one finding in this account that has survived every re-examination.
The owner's frustration, when it surfaced, was never about arithmetic. It was about lead quality and wasted time. Fair enough: the broad keyword was buying clicks from people who wanted a deck stained, a railing replaced, or a how-to video.
The obvious fix, and what it cost
So we did the obvious thing. In August we paused the broad keyword, rebuilt the negative list to a few hundred entries, added county-level targeting, and cleaned up the account's assets. Impression share on the serious terms climbed from the mid-thirties to over seventy percent. Search terms went from a dozen junk-laden queries a day to a few clean geographic ones. On cost-per-lead, the account was immediately better.
And volume fell by two-thirds. Account impressions went from about a hundred a day to about thirty, because the broad keyword had been the engine and nothing replaced it. The exact-match keywords that convert so well sit in a tiny auction: the best of them has a pool of roughly four and a half impressions a day in Chattanooga. You cannot bid your way to more of a query that only a few people type.
A second round of pruning a few weeks later made it worse. Two phrase keywords were paused for having zero conversions — on 28 clicks combined. At the account's blended conversion rate, 28 clicks should produce about half a lead. Zero was the expected outcome. The decision contained no information at all, and it is the kind of decision small accounts make constantly: treating a handful of clicks as a verdict. Our standing rule on this account is now that nothing gets paused on fewer than 150 clicks, because that is roughly what it takes to tell a dead keyword from an unlucky one.
Reading it correctly
Three things reframed the account once we stopped optimizing for the metric that was easiest to see.
Nothing in it was expensive. At a $20,000 to $40,000 job value, even the worst keyword in the account — $800 a lead — returns well at a modest close rate. The $130 exact-match leads are wonderful, but there are only a dozen of them a year. The question was never cost. It was volume, and tightening is the wrong instinct when volume is the constraint.
The measurement was missing half the leads. Every cost-per-lead figure in the account counts form submissions only. Phone calls go through a third-party tracking number that swaps out the number on the page, which means Google's own call tracking can't see them. For a contractor, phone leads are usually the majority. So every CPA we'd been agonizing over was overstated by an unknown amount, and the real fix is an integration that pushes calls from the tracking platform into Google Ads — not another keyword change.
The ads were never the main channel. This is the part that reorders everything. Arellano gets ten to fifteen leads a month from organic search — the website, the service pages, the local pages — against roughly four a month from ads in a good month. Organic is three to five times the paid channel. The retainer that had been mentally filed under "ads management" was, in substance, buying the organic presence, and the anxiety about ad spend was aimed at the smaller number.
That organic presence didn't happen by itself. In August the site got fifteen new pages: six neighborhood pages for Signal Mountain, Lookout Mountain, Ooltewah, Hixson, Soddy-Daisy, and Red Bank, each written around what actually differs about building a deck there — grade, rock, lot size, tree cover, HOA rules — rather than the same page with the town name swapped; two hub pages; and six guides on cost drivers, composite versus wood, build timelines, financing, covered decks versus screened porches, and permits. The ad landing page was rebuilt and its performance score went from 63 to 91 — the culprit was hardcoded image URLs the caching layer wasn't converting — and it has held in the nineties for two months.
What the data settled
Two more decisions came out of the full-account review, both with enough data behind them to be final.
Search partners — Google's network of non-Google sites — had been switched back on in the hope of volume. The all-time numbers across 800 partner clicks: a 16 percent click-through rate, which looks great, and a 0.25 percent conversion rate against 2 percent on Google Search proper. Triple the click rate at an eighth the conversion rate is the signature of accidental clicks on partner sites. It is off, permanently.
Bidding moved from Maximize Clicks to manual bids. Maximize Clicks had done exactly what it was told: bought the lowest-cost clicks available, which meant piling spend into the broad traffic that converts at one percent while the exact terms that convert at twelve went underfunded. Automated conversion bidding wasn't an option either — thirty-odd conversions a year is far below what the algorithm needs to learn from, and it would be learning from a signal that misses every phone call. Manual bids state the thesis directly: the two converters bid high, everything else bid low.
The lesson
A small-market ads account will always look like a cost problem, because cost-per-lead is the number on the screen. For a contractor selling five-figure jobs, it almost never is. The real questions are how many people in your area actually search for what you do, whether you're measuring all the leads you get, and whether the ads are even the channel doing the work. On this account, the answers were: not many, no, and no — and the right moves were to stop pruning, fix the measurement, and keep building the organic presence that was quietly producing most of the business.