Two people type nearly the same thing into Google. One searches "investment property loan". The other searches "home loan for investment". One of them is a customer. The other is shopping for a house to live in, will never qualify for the product, and costs exactly the same per click.
In lending the words overlap and the products do not. That one fact decided the architecture of an entire account before a single ad ran.
I built the paid search program for Brick City Capital, a private lender that finances property for real estate investors, named a Top Private Lender by Scotsman Guide in 2025. Its core product is a DSCR loan.
DSCR stands for debt service coverage ratio. In plain terms it is a loan underwritten on the rent the property collects rather than on the borrower's paycheck. No W-2s, no tax returns. The building qualifies itself.
That product has an unusual problem in search, and the problem is not volume. Almost every phrase an investor types sits one word away from a phrase a consumer types about their own home.
Two searches, one word apart
Look at what the keyword tools hand you when you research this category. "Investment property loan" is the core term. "Home loan for investment" reads like a synonym and carries owner-occupied intent, which is a different regulatory world and a different lender entirely.
"Equity loan on rental property" looks like a perfect fit until you notice it points at a home equity line, which this lender does not make. "Investment mortgage rates" is generic enough to catch people researching bank stocks.
Every one of those is an expensive click. Pull the forecast on the non-QM cluster, meaning loans written outside the standard qualified-mortgage rules, and Google's own top-of-page estimates run into the high teens and low twenties per click.
This is how a lending account burns money invisibly. The monthly report shows spend, clicks and a respectable number of form fills. Nothing in it shows that a slice of those leads were homeowners who were never eligible for the product they enquired about.
So the first structural asset in the account was the exclusion list.
Google's own guidance on negative keywords puts the job precisely: they "let you exclude search terms from your campaigns and help you focus on only the keywords that matter to your customers", and the terms to hunt for are ones "similar to your keywords, but might cater to customers searching for a different product".
Home equity, HELOC, owner occupied, primary residence, first-time buyer, FHA. Those went in before the first campaign was enabled, and they are the boundary line of the whole account.
There is a second reason to draw that line hard. Consumer home lending and business-purpose investor lending are advertised under different rules. Compliance owns where that line sits.
In a regulated vertical the account structure is the compliance system. Tight themes keep every ad answerable for the query it bought.
What the campaign layer is actually for
Campaigns hold budget, geography, bidding strategy and network settings. Ad groups hold which keywords sit next to which ad. Get that division wrong at the start and every later improvement is a workaround.
So campaigns were split by intent theme, each theme being a distinct thing a searcher wants. Core DSCR product terms. Investor property loans, covering investment property, rental and portfolio phrasing. Research terms like rates, calculators and requirements. Non-QM and short-term rental financing. Competitor names. Brand.
Each theme earned its own campaign because each one deserves its own money, its own geography and its own answer. A research searcher checking today's rates and an investor with a property under contract should never be competing for the same daily spend.
Inside those campaigns the ad groups stayed small enough that one ad could name the exact thing the person searched for. That is the whole rule, no acronym required: if covering the keywords in a group forced the ad to make two different promises, it was two groups.
The keyword I filed in the wrong place
Here is how fine that discipline gets. Reading back my own build, I found "buy rental property loan" sitting in the investment property loans group alongside "investment property financing" and "loan for investment home".
It does not belong there. "Buy" is acquisition intent. Someone with a property under contract, working a deadline, needing to know how fast money can close. The rest of that group is people researching how this kind of financing is structured.
Purchase and refinance are two different conversations, and one ad cannot hold both without going vague. So the phrase moved into its own purchase group with "loan to buy a rental property" and "rental property mortgage", and got an ad that opens on speed to close.
One keyword, one ad group, thirty seconds to fix. Small enough to feel beneath attention, and it decides whether the account is still legible in three months, because a misfiled term teaches the bidding algorithm the wrong lesson about what its group is for.
Forecast before you fund
Before any of it shipped I asked the question that most account builds skip: does each of these keywords actually carry volume, and is it the best term available for getting this lender traffic?
The honest answer at that point was no, because the list came out of a brainstorm. A structure built on plausible-sounding keywords looks identical to one built on evidence until real money moves through it.
Keyword Planner closes that gap for free. Google describes how the numbers are produced plainly enough: forecasted impressions "take your bid, budget, seasonality, and historical ad quality into account to estimate future performance".
What came back reshaped the plan. Investment property phrasing was far bigger than assumed, with the core terms returning roughly 9,900 estimated monthly searches each. Non-QM clustered around 4,400. Rental property financing sat near 2,900.
Other ideas died on contact. "DSCR bank" was my own hypothesis about how people search, and the data would not support giving it a group of its own. Better to learn that from a free tool than from an invoice.
The forecast then drove match type directly. Terms with validated volume and unmistakable intent went to exact match, where the ad shows only on that phrase.
Long-tail and comparison phrasing went to phrase match to catch the variants. Nothing ran on broad match at launch, because broad match spends its way to an education nobody planned for.
Year-over-year direction mattered as much as raw volume. A refinance term growing steeply earns a group built out around it. A term down heavily on the year is one keyword you inherit and never invest behind, whatever its absolute search count says.
Forecast before you fund. Keyword Planner is free due diligence most accounts skip on the way to spending real money.
Bidding on rival names, in a box
Conquesting means bidding on a competitor's brand name so your ad appears when somebody searches for them. In DSCR lending that means Kiavi, New Silver, Lima One, Constitution Lending, Griffin Funding.
It is the most tempting line in any lending plan and the easiest place to quietly lose. Half the people searching a competitor by name have already made their decision, and you are paying premium prices to interrupt a conversation that is effectively over.
So it ran as a contained experiment. Its own campaign, so its cost could never blend into the product themes and flatter them. One ad group per rival, so each could be judged and cut on its own evidence rather than as a bloc.
One more rule, and it is where the compliance instinct and the marketing instinct agree. The keywords name the competitor. The ads do not. Ad copy speaks to the advantage, which for this lender was the number of loan buyers behind each deal and the speed to close, and lets the searcher draw the conclusion.
Competitor terms are rented interest. They pay only when the landing page answers the comparison the searcher was already making.
Which is why that traffic never went to the general product page.
Choosing geography on purpose
Geography carries two separate decisions in a lender's account, and most builds only make one of them.
The first is where you can lawfully lend. Target by presence rather than interest, so ads reach people physically in the states served instead of people idly reading about them, and exclude every state outside the license footprint. An impression in a state you cannot serve buys a conversation that ends in an apology.
The second is the geo-modified search itself. People type "dscr loan florida" and "dscr lender texas", and matching that phrasing lifts relevance sharply. Building fifty ad groups to cover it would have been unmanageable, so a small set of phrase templates covered the states instead.
Then I hit a wall worth passing on. I wrote a headline reading "DSCR loans in {State}" and Google rejected it as invalid syntax. Ads supports keyword insertion, and location insertion where a proper feed exists. There is no state placeholder you can invent.
The fix moved the personalization one level down. Keep the ad general enough to be true anywhere, pass the state through to the destination, and let the page headline carry it. Ad copy handles the intent, and the page handles the geography.
The account is only half the machine
Every structural decision above is an argument about who should arrive. What happens after they arrive is a different build, and it is the half that decides whether the architecture ever pays.
The traffic landed on a branching quiz rather than a static contact form, so the questions that qualify a borrower got asked while the visitor still had momentum. Then the clock that matters started, and the first five minutes after submission did more for closed loans than any bid adjustment I made.
The last loop is what you report back to Google. Send raw form fills and the algorithm optimizes toward whatever produces the most forms, which in this vertical means the cheapest and least qualified traffic it can find.
Send back the leads that actually qualified and it starts hunting for those instead. That is the paid-account lens from the growth audit working as intended, and it is the only way spend here can be graded against gross profit per acquisition dollar rather than against a platform's own homework.
The architecture checklist
This order of operations transfers to any product where clicks are expensive, consideration is long, or claims are regulated. Insurance, healthcare, legal, wealth management, education, B2B software with a compliance surface. The sequence:
- Write the exclusion list first. Name the neighboring audience whose words look like yours, and refuse their phrasing at the door.
- Split campaigns by intent theme, because campaigns are where budget, geography and bidding actually live.
- Keep ad groups tight enough that one ad answers every keyword in the group without hedging.
- Run Keyword Planner forecasts before funding anything, and let volume and year-over-year direction decide which themes deserve their own campaign.
- Set match type from the evidence: exact for validated intent, phrase for the long tail, broad only once conversion data has earned it.
- Contain competitor bidding in its own campaign with its own comparison-aware landing page.
- Choose geography deliberately, targeting presence rather than interest, and excluding everywhere you cannot serve.
- Report the qualified lead back as the conversion, so the platform optimizes toward the leads sales actually wants.
Do that and the account becomes readable. Every dollar traces to an intent somebody deliberately chose to buy, which means it can be defended in a compliance review and scaled without a rebuild.
In verticals where clicks cost this much, structure is the difference between buying intent and renting noise. Sloppy accounts do not announce themselves. They just convert your budget into traffic that was never going to qualify.
If you are standing up paid search in an expensive or regulated category, architecture like this is what I build first. Let's talk.