A specialty auto parts retailer came to us with a 100,000+ SKU catalog and a problem. On the surface things appeared to be running well. Paid spend was going up. ROAS (return on ad spend) remained consistent. But, revenue wasn’t increasing and it became difficult for anyone to identify which products in the catalog were actually profitable.
12 months later, revenue increased by 22%; profit after advertising expenses increased by $530,000 year-over-year. The quality of visitors to the site improved. Nothing about that improvement came from spending more or bidding harder. It came from restructuring how demand was captured, measured, and filtered across a catalog too large and too uneven to treat as a single entity. That's the story this article is built around, because it's a more useful teacher than a generic list of best practices.
Most of what breaks ecommerce PPC accounts is a structure problem that only shows up once a catalog gets big enough to hide it.
This isn’t a listicle of "7 ways to increase your ROAS". Instead, we’re going to outline our actual processes used when working on a client's account: Google Merchant Center and feed structure, keyword and demand segmentation, channel sequencing, bid management, remarketing, and measurement. We’ll walk through the way it played out on a real account, with the tactics that support it.
Ecommerce PPC Management is defined as an ongoing process of creating, building, and modifying paid campaigns (primarily through Google Ads and Microsoft Ads) in such a way that your spend generates revenue, and ultimately profit when your spending is factored in.
For low-SKU catalogs, there are fewer variables to manage; you have a smaller amount of keywords covered, less cluttered feed data, rational bid pricing etc. When dealing with large/complex catalogs (10k+ SKUs, variable margin levels, fitment/compatibility requirements, retail vs trade buyers), it becomes a much larger problem. The account has to represent a catalog that no single dashboard metric can summarize accurately.
ROAS is just dividing the money earned from each campaign by the cost for that campaign. This is probably the statistic most likely to lie to you about performance in catalogs as they grow in size. We get a lot of clients that overly focus on ROAS to their paid account’s detriment.
Why?
ROAS represents an average. An account with 100,000 SKUs does not operate as one single unit. Instead it operates as 100,000 small units that are unevenly contributing toward profitability, a number of which are profitable while others are losing money. Some items are your cash-cows, generating significant revenue because a number of these items are producing much higher margins than other products.
That’s when ROAS lies to you. If there are enough of these high margin generating products then these would drive the overall ROAS for the entire account regardless of whether or not the lower margin products (the majority) were making money. The account may appear to be performing well based but the underlying business itself is not.
We saw this exact situation on our client’s auto parts account. Although the performance of the account had not fallen apart; spend was changing and ROAS remained constant. However, sales had stopped growing and profits after ad expenses were unpredictable. The platform reporting did not reflect how actual demand was being pulled through the product offerings across the catalog. The account was being operated off a metric that was technically correct but completely useless for providing directional guidance.
The fix isn't a better ROAS target. It's better questions
Instead of asking if ROAS is over 400%, I would instead ask:
Common ways this shows up at scale:
Don’t get me wrong, ROAS still matters. It's a useful constraint against waste. It just isn't the finish line. Profit after advertising costs, contribution by category, and revenue that actually sticks are what should guide the account.
Every platform in your marketing mix should have a defined job. On the auto parts rebuild for our 100k SKU client, here's how that played out:
Google Ads was positioned as the primary demand capture layer
It's where the highest-intent, most specific searches happen, and for a 100k SKU catalog, that's where the account had to be strongest.
Performance Max was structured to support discovery without collapsing into blended reporting
PMax is genuinely useful for surfacing demand a manually structured campaign might miss, but its reporting can obscure exactly the SKU-level and demand-type visibility a large catalog needs. The goal here was to use PMax wisely and not allow it to become a black box.
Microsoft Ads was introduced to capture incremental demand without cannibalizing what Google was already converting
Microsoft Ads is often overlooked, but there’s still opportunity here due to lower competition, generally lower CPCs for many auto parts categories and a buyer group that behaves relatively similarly to Google, so the structure can be duplicated.
Social channels like Meta and YouTube play a supporting role in the spec-driven market. These platforms help generate awareness and trust, and allow for remarketing. Don’t expect them to be primary demand engines for auto parts, marine supplies, etc.
A feed can be fully complete and still fail to tell Google's algorithm, or your own reporting, which products are actually worth the spend. Your feed could contain every attribute necessary, however your reporting tool doesn’t know how to read them.
For our auto parts client, the focus was on building the feed based on performance groups rather than making sure everything was included in the feed:
This last point is something most agencies miss because it seems counterintuitive to remove SKUs from paid campaigns. The logic behind removing underperforming SKUs from paid campaigns is that when you include every SKU in a feed and allow each SKU equal opportunity to advertise (regardless of whether or not it has ever been purchased previously), you’re wasting money on a SKU that is never going to generate a sale. Removing unprofitable SKUs allowed us to create a better performing feed that gave more opportunity for profitable SKUs to grow.
The feed may feel like a one-time setup but that’s where we see accounts go wrong. Things like price, availability, performance shift can change over time. Ongoing feed maintenance is crucial to keeping PPC profitable.
For large catalogs, the biggest failure with keyword strategy is reporting that mixes branded demand with non-branded demand and hides where growth really comes from. For our auto parts account example, search and shopping programs were reorganized specifically to separate branded demand from other demand.
People searching by brand name already know your store and convert at different rates and for different reasons versus those who are still undecided about where to buy. If you blend both types of searches together in reporting, performance for branded searches consistently hides what is really happening for acquisition of non-branded demand. Growth ceilings usually reside in non-branded acquisition.
Beyond that segmentation, the fundamentals still apply, especially in a spec-driven catalog:
You’re not looking for a long keyword list, that’s not the goal. You need a reporting structure that shows where growth is coming from; whether it’s from people that already know your brand or new customers you’re actually winning for the first time.
Responsive ads in large catalogs are based on responsive search ads (RSA), and can include up to 15 different headline options plus 4 different description options. These are considered a pool of assets that Google will combine to form an unlimited number of unique versions of your ad based on each individual’s search query, device and location. Unlike previous expanded text ad formats, you have no control over how many or which headlines are paired together. Thus, what is "good" ad copy when dealing with thousands of ad units is no longer simply writing the best possible three-headline version, but rather providing enough high-quality, varied content so that Google has ample opportunity to test all of its various combinations.
What holds up for a spec-driven catalog:
Extensions still help, and are going to be even more important as a catalog grows in size, since they give Google additional "surface area" to get your ads more closely matched with user intent. Sitelinks to relevant categories. Structured snippets will allow you to display information about product specs and certifications. Price extension will provide users with range transparency. Promotion extensions can be used if there is actually an active promotion (and not something manufactured by the marketing team).
Test at the Asset Level — Google has started reporting on clicks/conversions per individual headline and description vs reporting on ad performance. Use these metrics to eliminate poorly performing assets and avoid guessing.
In a large catalog what works in one category may NOT work well in another category. Therefore, testing should occur on each category individually — it's not a one-and-done process.
Landing pages carry more weight in spec-driven ecommerce than in most other categories, since landing pages are tasked with confirming the buyer landed on the correct item prior to beginning their buying decision.
In practice, that means:
Getting the right buyer to the right page matters, but so does what happens when that routing fails, which it sometimes will at catalog scale.
Automated bidding has a lot of value. The problem is that it will spend budget toward whatever the algorithm determines is a conversion, without regard to margins, seasonality or SKUs you actually want to push.
For our 100k SKU auto parts client, we prevented wasted spend by using ROAS as a constraint. But we didn’t use ROAS as the end all be all of success for the account. Stability, contribution, revenue, and profit behavior did. That distinction changes how you build guardrails around automation.
Don’t allow automation to dictate your account without guardrails. That means putting in place margin-based segmentation, negative keyword discipline, and regular audits sitting on top of whatever bidding automation you're using.
Remarketing requires segmentation. You shouldn’t be showing the same ad to everyone.
For spec-driven catalogs, the audience is not necessarily going to be swayed by lifestyle images. They want to see the same trust and specificity signals displayed on the site: confirmed compatibility, stock status, social proof. Someone who viewed a specific part number wants to see that part number again, not a generic ad to return to the site.
Update your segment lists by removing converted users, refreshing criteria and rotating creative. This will reduce ad fatigue that could be diluting performance.
The ads are living, data is coming in, but are you doing anything with it? This is where we see many ad accounts go wrong. They build a “dashboard” with a few KPIs and think they’re done. For our 100k SKU auto parts client, it was a huge leverage point during the account rebuild.
First, we fixed tracking. Most tracking setups are not correct, meaning there’s a discrepancy between platform reporting and business reality. We oriented reporting around profit, cost of sale, revenue and channel contribution.
Our goal is never perfect 1:1 attribution. That’s a fantasy, and chasing it is a time suck. Our goal was directional clarity that could support real decisions without constant reconciliation. We wanted to give our client a dashboard that finance and marketing could both look at and agree on, instead of two separate stories that had to be manually reconciled every time a decision needed to be made.
What that requires in practice:
One caveat in the age of LLMs and AI: visibility isn't only rankings and clicks anymore. Zero-click recommendations are a growing part of how buyers find products. They run on the same clean, structured product data that PPC and feed work depend on. A measurement approach that only accounts for traditional clicks is already missing part of the picture.
Here’s how we actually structure an ad account rebuild:
Weeks 1–2: Diagnostic. Audit tracking, feed data, and account structure. Identify where branded demand is masking non-brand performance, where spend is following volume instead of contribution, and where reporting and business reality have drifted apart.
Weeks 3–6: Structural fixes. Rebuild the feed around performance tiers. Separate branded and non-brand campaigns. Reconfigure tracking so reporting reflects profit, not just activity. Sequence channels (primary capture layer, discovery layer, incremental layer) instead of letting them compete for the same demand.
Weeks 7–10: Stabilization. Let the restructured account run. Spend should start following contribution instead of volume. This is usually where ROAS holds steady but the composition underneath it changes. Meaning fewer wasted clicks, tighter traffic quality, cleaner signal.
Weeks 11–13: Scale what's working. With clean measurement and a filtered catalog in place, growth becomes incremental instead of fragile because you're now scaling the parts of the catalog that were actually proven to convert profitably, not the parts that happened to have the most inventory or the most search volume.
This is essentially how we rebuilt the ad account for our 100k SKU auto parts ecommerce client. The $530K profit gain and 22% revenue growth built gradually, as the account stopped amplifying low-value demand and started reinforcing the paths through the catalog that were already profitable.
Three things changed structurally on the auto parts account, and they're the same three things worth checking in any large-catalog PPC account that's plateaued:
"Our ROAS and Revenue are improving and we don't have wasted spend on ads that don't have a return." — General Manager, Specialty Auto Parts Retailer
For a large catalog, we aim to build a paid media system that could scale without quietly breaking, which is a better goal than chasing a higher ROAS number every quarter.
See what your demand capture is actually doing.
The SCUBE Game Plan is a focused review for large, spec-driven catalogs — built to surface what's contributing to performance, what's masking underlying issues, and where structure is quietly working against you. You'll walk away with clear KPIs, campaign insights, and a 90-day roadmap.
[Get Your Game Plan]
Why is ROAS (return on ad spend) seemingly healthy even though profit is flat?
This is because ROAS is simply the total revenue generated by an entire account divided by the total dollars spent to drive that revenue. In other words, it's an account-wide average. A couple of extremely successful products could be pulling up your overall ROAS average at the same time as you're losing margins on every product in the rest of your catalog.
How long after you've implemented these structural improvements will I see the actual benefits in terms of profits?
Some of the early changes to your structure such as segmenting feeds, fixing tracking problems etc., will begin to show directional movement toward better profitability in a month or two. The kind of sustained profit growth we saw on the auto parts account (measured over 12 months) takes that long because it compounds gradually, not because any single fix is slow.
What's a realistic ROAS target?
It depends on margin, category, and catalog complexity. There's no universal number worth anchoring to. A more useful target is knowing what ROAS your account needs to sustain given your actual margins, and treating that as a floor, not a scoreboard.
How do you decide which SKUs to pull out of paid campaigns entirely?
Pricing competitiveness, availability, and realistic conversion likelihood. If a SKU can't realistically return profit given current pricing and availability, leaving it in paid campaigns just spreads the budget thinner across the products that can.
How does branded vs. non-brand segmentation actually change performance?
It’s not necessarily that it changes performance, it changes what you see in reporting. When branded and non-branded demand are separated, you can see if growth is coming from new customers or people who already knew your store. That’s going to change where you allocate budget next.
Does Performance Max work for large, complex catalogs?
Short answer, yes. Longer answer, PMax can work but it needs to be structured to avoid collapsing into blended reporting that hides SKU-level performance. Used without guardrails, PMax can obscure exactly the signal a large catalog needs most.
How do you keep automated bidding from working against your margins?
Layer margin-based segmentation, negative keyword discipline, and regular audits on top of whatever bidding automation you're using. Automation should scale a strategy you've already defined. Where it can go wrong is when it’s making the margin decisions on its own.
What does "directional clarity" in reporting actually mean?
A reporting setup where the ad account and the business's actual financials tell a consistent story, closely enough that decisions don't require manual reconciliation every time. You should not be aiming for 100% perfect attribution. You don’t want two conflicting pictures of the same business.
