Tag: Data Quality

Why MMM Still Demands Clean Data and Human Judgment
I see marketing mix modeling (MMM) becoming easier to access, but I do not think it has become easy to get right.

After several conversations about MMM adoption, I keep hearing the same concern: “We believe in MMM, but we do not know how to get started.”

My answer is that open-source platforms have lowered the barrier to entry in a meaningful way. What they have not lowered is the level of expertise required to produce results that are trustworthy, explainable, and useful for decision-making.

Open-source MMM has changed the starting point

I am seeing MMM adoption accelerate because marketers need more durable measurement methods. Almost half of U.S. marketers expect to invest more in MMM over the next year, and many now rank it as one of the most reliable measurement approaches available.

The open-source shift is real. Three production-grade libraries now give teams a practical way to approach MMM across a wide methodological spectrum.
- Robyn (Meta, R): I see this as the most approachable starting point because it includes automated hyperparameter search through Nevergrad, Pareto frontier model selection, decomposition, and response curve plots. It is also the one I use most often because it is highly customizable.
- Meridian (Google, Python/TensorFlow): I view Meridian as a more rigorous option, especially because it uses Bayesian inference, geo-level priors, and principled uncertainty quantification. The tradeoff is a steeper learning curve.
- PyMC-Marketing (PyMC Labs, Python): I consider this the most flexible path. It offers a full probabilistic model that comes closest to academic-grade Bayesian MMM, but it also demands the most statistical fluency.
This generation of tools has removed the old $150,000 to $500,000 consulting gate that used to be the primary path into MMM. A team with R or Python expertise and reasonably clean historical data can now run a model in-house.

Open-source R and Python tools have lowered the cost of starting with marketing mix modeling, but the expertise needed to produce trustworthy, actionable MMM remains the real ceiling.

The caveat I always make explicit is this: “free tool” does not mean “free model.” The software may be free, but the domain expertise needed to configure it correctly is not. That expertise is a major part of the value.

The vendor landscape is crowded and complicated

I also see a fast-growing SaaS layer built on top of open-source MMM. To evaluate it clearly, I find it helpful to separate vendors into a few practical groups.

Data-layer-first vendors

Platforms like Rockerbox and Northbeam started with attribution and data collection, then added MMM. Their advantage is usually pipeline speed and data access, not deep modeling flexibility or customization.

Measurement-first vendors

Platforms such as Measured, Analytic Partners, Ekimetrics, and Nielsen Gracenote tend to offer more rigorous modeling and enterprise-grade capabilities, usually at a higher price point.

Google Meridian and GA360

I think Google’s decision to open-source Meridian is both a generous contribution to the field and a strategic move. When a walled garden funds and packages a measurement methodology that can be used to evaluate its own channels, I believe it is worth maintaining healthy skepticism about priors, defaults, and assumptions, even when the code is transparent.

Open-source MMM tools now span a clear trade-off: Robyn offers the most approachable starting point, Meridian adds Bayesian rigor, and PyMC-Marketing pushes deepest into statistical flexibility.

The practical vendor question I keep coming back to is simple: who owns the data layer, and does that ownership create conflicts in the modeling layer?

Challenge 1: Data access can quietly break MMM

I think data access is the most underappreciated MMM implementation blocker. A well-specified model needs more than a quick export from a dashboard.
- I usually want two to three years of weekly data as a baseline, so the model can capture at least two full seasonality cycles and enough spend variation to learn from.
- I need consistent channel-level spend granularity, not just a broad “digital” bucket. Search, social, display, video, and other channels need to be separated.
- I need offline channels such as TV, OOH, radio, events, and direct mail, even though they often live in different systems, belong to different teams, and use incompatible time periods.
- I need external covariates, including macro indicators, competitor activity, pricing data, and product launch calendars.
- For B2B, I often need even more history because longer sales cycles and lower conversion volumes make the data requirements more demanding.
In practice, I often find that the real blocker is the six-week data archaeology project that happens before modeling begins. Finance owns revenue. The brand team owns TV. The agency owns digital spend. A spreadsheet from 2021 may be the only record of trade promotions.

The model is only as good as the data archaeology behind it, and that is rarely the part anyone highlights in a vendor demo.

Challenge 2: I still have to roll up my sleeves

AI assistants have lowered the syntax barrier. They can scaffold a Robyn run, generate a Meridian configuration, or help debug a PyMC model. What they cannot reliably do yet is make the judgment calls that determine whether an MMM is credible.

Rows of illuminated data cabinets and paper files stretch into the distance, capturing the pressure on marketers to turn fragmented customer data into a smarter performance engine.
- I still have to decide where to land on a Pareto frontier across hundreds of model solutions, balancing NRMSE against DECOMP.RSSD tradeoffs.
- I still have to know whether Nevergrad’s optimizer has meaningfully converged or simply landed in a local minimum.
- I still have to configure adstock transformation parameters, including Weibull shape and scale or geometric decay, so they reflect realistic channel behavior.
- I still have to diagnose why a model gives a channel an implausible contribution and decide whether the fix is a prior, a data correction, or a variable exclusion.
In other words, if I try to vibe code my way into MMM, I may end up with a model that appears to work but is wrong in ways I will not catch. The scripting is not the hardest part. The real work is validating the output, including using channel-specific incrementality experiments to calibrate the model.

Challenge 3: Human expertise is not optional

Even if the tools mature enough for AI to run a competent default MMM, I still see human expertise as essential. The irreplaceable work is encoding business context that no model can infer from the data alone.
- Adstock and carryover context: I need to know whether a TV buy carries over for four weeks, paid search carries over for three days, or a brand awareness campaign decays over months. That knowledge usually lives with channel experts, not inside the dataset.
- Saturation curve shape: I need to recognize when a channel is probably approaching diminishing returns before the model says so, and I need to question the model when it suggests something implausible.
- Guardrails and anomaly handling: I need to explicitly model or flag COVID troughs, product launches, pricing shifts, and macro disruptions as structural breaks. AI does not automatically know a client had a pricing crisis in Q3 2022.
- Interpretation sanity checks: If a model assigns 40% of contribution to TV for a brand spending $2 million on TV, I need the experience to say, “That feels wrong,” and investigate. That intuition is earned, not computed.
- Organizational translation: A technically correct model has little value if I cannot explain why it recommends moving 15% of search budget to CTV in language a CMO and CFO will act on.
I start with the groundwork before the model

The best place to begin is not the model itself. I start by understanding what data is needed, who owns it, and who can help interpret it in the context of real marketing decisions.

None of that is quick or easy, but it is essential if I want meaningful insight from MMM, whether I choose an open-source library or a subscription-based platform.

As a practical first step, I would download Robyn’s demo script and experiment with sample data before applying MMM to my own business data. That kind of hands-on testing makes the strengths, limits, and judgment calls much clearer.

Inspired by this post on Search Engine Land.
July 9, 2026
Bad Conversion Data Is Quietly Wrecking Google Ads

I used to think bad data mainly meant bad reporting. Now, in Google Ads, I see it as something much more expensive: bad delivery. When conversion data is wrong, it does not just make a dashboard confusing. It can train campaigns to spend budget chasing the wrong people.

As automation takes over more of the ad-buying process, from creative generation to bidding, data has become one of the few inputs I can still control. It may also be the most important one, because automation can only optimize toward the signals I give it.

I keep coming back to one question: what is worse, a brilliant ad shown to the wrong audience or an average ad shown to the right one? The first burns budget on people I do not want. The second may not win every click, but when someone does engage, at least they are closer to the customer I actually need.

That is why I have to ask myself a harder question before launching any automated campaign: did I spend more time verifying the data than writing the ad copy?

The cost of bad data has changed

A few years ago, bad tracking was mostly a reporting problem.

If a tag fired twice, a conversion was mishandled, a value came through incorrectly, or offline conversions stopped working for a few weeks, the main result was a dashboard that did not add up. It was frustrating, but the damage was usually limited. Someone would eventually question the numbers in a monthly review, I would trace the issue, fix it, and the next report would look cleaner.

That same data now feeds the algorithm buying paid media. Smart Bidding does not wait for me to interpret a report or sit through a monthly review. It reads conversion data and acts on it before I may even notice that something is broken.

The same wrong number now creates a very different outcome. A bad number in a report requires an explanation in a meeting. A bad number in a conversion action used for bidding costs money immediately, because the algorithm does not know the signal is wrong.

It simply optimizes toward that signal the moment it sees it, and it does so efficiently.

Google does not understand my funnel or my business

Google may let me label conversion actions as “lead,” “opportunity,” or something similar, but those labels are mainly for organization. The platform does not truly understand where each conversion event sits in my funnel.

What it sees is a conversion event with a numeric value attached to it, usually a currency value. It does not inherently know that a newsletter signup might be worth $2 in eventual value, a lead might be worth $60, and an opportunity might be worth $400. To Google, those are conversion events. Without better signals, it has no real context that one may be worth 200 times another.

The algorithm is not optimizing for my business outcome by default. It is optimizing for the data I provide. If that data is wrong, the optimization will be wrong too.

For example, if every form submission fires the same conversion with the same default value, I give the system no clean way to separate low-intent inquiries from high-value prospects. The algorithm treats them the same. And because low-quality leads are often cheaper to acquire, it can quickly flood the account with them.

The cost per lead may drop from $40 to $25, and the dashboard may make performance look more than 35% better. But behind that cleaner metric, the pipeline can dry up as genuinely qualified inquiries quietly fall by half.

Dig deeper: Why better signals drive paid search performance

3 ways bad data quietly wrecks delivery

Bad data can show up in different ways, but I see three issues that are especially likely to derail campaign delivery.

1. Wrong event

If I optimize for a top-of-funnel action like a page view while the real conversion events happen further down the funnel, the algorithm learns to buy more of those cheap events. The problem is that the lower-funnel activity may never follow.

2. Wrong value

If I count every conversion equally, or assign every conversion the same placeholder value, I hide the real differences in business value. When actual value can vary by 10 times or more, the algorithm will often chase the easier, lower-value conversions because they are cheaper to acquire.

3. No data

This problem does not get discussed enough. A complete break in conversion data can damage a campaign faster than almost anything else.

On Day 1, the algorithm starts wondering where the conversions went. By Day 2, it begins assuming they may not be coming back. By Day 3, it can start making serious bidding changes. Within a week, many campaigns can throttle themselves down to almost nothing.

How I pick the right signal for Google

So how do I fix this? I start by choosing the signal that best represents business value, not just the easiest action to count.

Take a typical lead generation business. Some leads will never convert, while others may be worth 10 times as much as the rest.

If the form asks the right qualifying questions, I may already know which leads are which. But if I optimize for every submitted lead using a target CPA, I am telling Google that all leads are equally valuable.

Imagine an account spending $20,000 a month at a $40 target CPA and generating about 500 leads. Only 150 qualify, and maybe just 50 are genuinely high value. A basic lead may be worth $60, a qualified lead may be worth $200, and a high-value lead may be worth $600. That is a 10 times spread in value.

In that situation, I have several ways to improve the optimization signal.

Optimize for a qualified lead: I can create a new conversion action, such as “qualified lead,” and fire it only when a lead has real value. Then I can move the target CPA strategy to that conversion action, knowing the campaign will ignore leads with no value. The advantage is that I train the campaign on a more meaningful signal. The downside is that every qualified lead is still treated equally.

Assign conversion values and use target ROAS: I can add a currency value to the qualified lead based on the potential revenue it could generate if it becomes a sale. Then I can switch the campaign to target ROAS, allowing Google to optimize for return instead of simply counting leads. The tradeoff is that it may still buy larger numbers of lower-value leads if it can acquire them at the right price.

Optimize for a high-value lead: I can create a “high-value lead” conversion event that fires only for top-tier leads, with or without a conversion value. Then I can optimize with either target CPA or target ROAS, depending on whether I care more about acquisition cost or return. The advantage is stronger lead quality. The downside is that, depending on spend and volume, the data may be too limited to support this approach until the account scales.

These are only a few possible optimization signals, and they do not even go deeper into the funnel. I can apply the same thinking to lower-funnel milestones by creating separate conversion actions for events such as contacted lead, qualified contact, or high-value contact.

Targeting and measurement can be different

This sounds simple, but the conversion event I optimize for and the one I report on are not always the same. In many cases, they should not be the same. One trains the algorithm. The other tells me how that training is performing.

In the example above, a client or internal stakeholder may still want to see cost per lead. That is a valid metric. But the campaign may be optimizing for the Qualified Lead conversion, not the original lead submission.

I can keep the original lead conversion running purely as a reporting metric, so stakeholders still get their cost-per-lead view while the campaign bids on the qualified lead signal that actually reflects business value.

Same campaign. Two conversions. Two very different jobs.

That brings me back to the question I started with: did I spend more time verifying the data than writing the ad? In an automated account, data is no longer just measurement. Data is strategy.

Inspired by this post on Search Engine Land.

July 3, 2026
Build Trust in Your Marketing Data to Eliminate Skepticism
As a marketer, I know how it feels to operate with a hidden skepticism tax. Trusting marketing data can be a challenge, often leading to countless hours spent cleaning spreadsheets and reconciling conflicting reports. And let’s not forget second-guessing those attribution models and AI outputs.

This lack of trust slows down execution, weakens team alignment, and results in decisions built on shaky foundations. A prime example is branded search, which often undeservedly takes credit for conversions that were likely to happen anyway. It’s like crediting a revolving door for everyone who enters a building. This gap between correlation and causation highlights a broader issue in modern marketing—a reliance on fragmented or low-confidence data.

The key isn’t just collecting more data, but building a foundation of data we can actually rely on—through verified identities, unified reporting, cleaner pipelines, and a robust measurement framework designed to distinguish true signals from noise.

Let’s break down some core concepts behind building this foundation and the types of data environments they foster.

Probabilistic vs. Deterministic

Consider a coffee shop loyalty app to explain probabilistic vs. deterministic data: When a customer logs in and orders, you know it’s Sarah. That’s deterministic. Conversely, if someone on the same Wi-Fi browses your menu without logging in, you might assume it’s Sarah based on the device and location signals—it’s probabilistic. Both have their uses, but assumptions can lead to inaccurate messages, like sending a “Happy Birthday, Sarah!” notification without certainty.

Using a data-to-confidence mapping, like the identity confidence thermometer, can help explain this concept effectively to clients.

Deterministic data sits at the top of the thermometer (100% confidence), with various probabilistic confidence levels descending down to the bottom.

Siloed vs. Holistic

Imagine the old tale of blind folks describing an elephant: Marketing describes the trunk as a hose, Sales sees the leg as a tree, and Finance calls the tail a rope. This illustrates the pitfalls of siloed data in ROI reporting. A holistic approach ensures everyone sees the whole elephant.

In a more practical example, a B2B SaaS company runs LinkedIn ads. Marketing registers 5,000 form fills, Sales finds only 2,000 worthy leads in the CRM, and Finance reports 1,200 closed deals attributed to organic traffic due to broken UTMs. Different teams, different truths, zero confidence.

Here’s what these inconsistencies look like, contrasted with a unified data spine approach.

Third, First, and Zero-Party Data

Think about buying a house:
- Third-party data: a nosy neighbor speculating about a move—it’s just hearsay.
- First-party data: a realtor who sees them attending open houses—observed behavior.
- Zero-party data: the buyer expressing intent on a form—it’s direct communication.
As cookies fade away, marketers will shift from widespread hearsay to less frequent but more valuable direct interactions.

Visualize this as a pyramid: third-party data at the base (widest, lowest trust), first-party in the middle, and zero-party at the top (narrowest, highest trust).

Big Data vs. Correct Data

Picture a cluttered kitchen where nothing is ever discarded. The fridge is full, but half the contents have expired, forcing you to sift through it all for a single fresh ingredient. Occasionally, you use something spoiled—this is ‘big data’ for you.

By contrast, ‘correct data’ is a well-organized pantry: fewer items, all fresh, accurately labeled, and easily accessible. Consider feeding an AI model a massive data set with duplicates and errors—it might mislead rather than help you make informed decisions.

Here’s a visual metaphor of raw data flowing into a ‘swamp’ versus passing through a filter into a clean, reliable reservoir.

Correlation vs. Causation

You’ve probably encountered this concept before. In marketing, branded search often seems like a high performer because it records conversions right before purchases, similar to a revolving door taking credit for everyone entering a building.

Correlation identifies that those walking through the door became customers, while causation asks whether they’d have entered regardless of the door. Incrementality testing is key here.

In this test, you hold out a group from seeing ads and compare conversion rates to the exposed group. If both groups convert similarly, ads may be taking credit rather than creating demand.

An example might show branded search with inflated ROAS compared to a more accurate, incrementality-adjusted view emphasizing prospecting channels.

Building a Stronger Marketing Confidence Layer

To establish cross-team confidence, consider these data foundation tools:
- Identity confidence thermometer: Go from probabilistic data (low confidence) to deterministic data (high confidence).
- Siloed vs. holistic: Transition from siloed data to a holistic view for greater confidence.
- Data trust pyramid: Move from third-party (low confidence) to first- and zero-party data (high confidence).
- Big data vs. correct data pipeline: Filter raw data to reliable outputs, moving away from a ‘confidently wrong’ AI.
- Correlation vs. causation ROAS: Shift from identifying correlations to proving causation with a scientific approach.
While AI can automate countless tasks, effective decision-making must be upheld by experienced marketers applying good judgment. These data foundations help us move closer to achieving that.

Inspired by this post on Search Engine Land.
May 13, 2026
Boost Your Data Insights with Google Analytics Task Assistant

When I first heard about Google Analytics introducing their new Task Assistant, I was intrigued. This tool promises to be a game-changer for those of us who want to maximize our use of Google Analytics without needing deep technical know-how.

It’s exciting to see Google simplify such a complex product. Task Assistant is designed to help advertisers and analysts like me gain more value from our data effortlessly.

What’s New. With the rollout of Task Assistant, Google Analytics offers a guided workflow tool that surfaces tailored recommendations. This means improving property setup, data collection, and reporting is easier than ever.

How It Works. Located in the left-hand navigation, Task Assistant organizes recommendations into clear categories like connecting accounts and enhancing reporting. I can mark tasks as complete or skip items not aligning with my goals, making the setup more flexible.

Why We Care. Identifying gaps in tracking quickly helps ensure I’m working with reliable data. Task Assistant minimizes the risk of missed insights or inaccurate reporting, allowing for confident optimization of campaigns and budgets.

Between the Lines. Analytics platforms, as powerful as they are, can be underutilized due to poor configuration. I’m glad Google is turning setup into a step-by-step process rather than leaving it as a daunting manual audit.

The Bottom Line. Task Assistant is all about making Google Analytics more actionable. It guides users toward better data quality and effective measurement, all with less guesswork.

Inspired by this post on Search Engine Land.

April 30, 2026
Mastering Paid Search: What to Optimize When Keywords Matter Less

In today’s digital landscape, I’ve noticed that paid search platforms are evolving to prioritize who sees my ads, often without depending solely on my chosen keywords.

This shift means I need to focus on optimization strategies beyond just keywords, such as leveraging audience data, enhancing landing page context, and understanding conversion behaviors. Recognizing this shift is crucial for me to know where to focus my efforts now.

A decade ago, keywords gave me a sense of control. Back then, hypersegmentation and single keyword ad groups were the norm.

We’d meticulously create unique landing pages for each keyword in every ad group, reveling in the manual process, convinced that we controlled the machine.

Times have changed, and the forecast of Google and Microsoft phasing out keywords feels more real than ever.

With tools like Performance Max and emerging AI Max solutions, along with contextual LLM-driven searches such as ChatGPT, I see the industry leaning towards a keywordless future.

Still, keywords remain vital as they reveal user intent and indicate where users stand in their journey:

If these signals are now managed behind a black box, my role as a marketer is evolving. So, what am I optimizing for?

Dig deeper: Beyond keywords: Mastering AI-driven campaigns

Intent is now inferred from a web of signals, relegating individual keywords to the background. My optimization focus should now be on three main pillars in 2026.

Google now emphasizes customer match and first-party data over mere queries. With Data Manager API integration, it identifies users in auctions matching my key deals.

No longer do I bid on “cloud security.” Instead, I target IT directors (sharing first-party data) investigating SOC 2 compliance, even if they search for something vague like “scaling infrastructure.”

B2B match rates can be challenging, but this is where I must innovate my strategy, broadening one-to-one list matching and collaborating with integration partners.

Clustering individuals by shared pain points and offering on-site experiences help me understand their verified intent before reaching the remarketing list.

My landing page serves as a vital data source. Google’s AI examines it to grasp the nuances of my offerings, making creative assets crucial signals that align with my target themes and keywords.

If my landing page effectively communicates “mid-market manufacturing,” AI identifies relevant users regardless of specific keyword use, transforming my “keyword strategy” into a content strategy.

Opting for a creative approach similar to Meta’s, where Andromeda elevates the creative as a primary targeting signal, is beneficial. These creative inputs define my audience, demanding a balance between creative and technical input.

Journey-aware bidding and value-based bidding mean algorithms now analyze a user’s journey beyond the final click.

Optimization now targets “high-value need states,” feeding the system data about mid-funnel behaviors that result in significant contracts.

Dig deeper: Why better signals drive paid search performance

The most profound change for digital marketers, including myself, is shifting focus from query-level to user-level intent.

While the previously ignored query “how to manage payroll” might not have targeted enterprise SaaS companies, AI now understands if that user is a financial VP at a large firm, indicating commercial intent.

If it’s the right user, the right signals should prompt AI to act on their purchasing stage.

As AI handles matching, my role shifts towards becoming a data architect.

Data quality determines my success. I must feed AI with valuable leads to optimize for value-based bidding effectively.

Assessing the health of my signal, from landing pages optimized for AI readability to correct technical content, ensures Google accurately targets my audience.

I now focus less on micromanaging search terms and more on managing brand exclusions and negative themes.

The future of search is about being the best solution for the right individual at their evolving need state.

Keywords served as training wheels, but it’s time to see how quickly my data can propel me forward.

Dig deeper: Why PPC teams are becoming data teams

Inspired by this post on Search Engine Land.

April 27, 2026
Unmasking AI: Is Your Data Truly Ready?

As I look around, it seems like everyone is scrambling to harness AI’s power. However, I’m realizing that fundamental identity gaps and issues like fraud and unreliable inputs are not getting resolved, but rather they are magnified by AI models.

AI has quickly become one of the most confidently discussed items in our modern marketing strategies. Budgets are reallocated, teams restructured, and vendors evaluated primarily by how “AI-powered” they appear. The belief is strong that once the right AI models are in place, performance metrics—such as targeting, segmentation, and conversion—will simply fall into place.

Yet, I’ve discovered a quieter truth. While organizations aren’t necessarily struggling with using AI, they face challenges feeding it adequate data. And often, the data they are supplying AI isn’t nearly as reliable as assumed.

This realization leads me to the uncomfortable truth about inputs. AI doesn’t produce truths; it magnifies what’s provided. If data is fragmented, outdated, or manipulated, AI doesn’t correct it—it scales it confidently.

Marketers have invested heavily in data infrastructures, only to find that an abundance of data and signals doesn’t necessarily equate to readiness. Large volumes do not guarantee validity. For instance, customer profiles built from various identifiers don’t assure a unified identity, and AI models are not inherently designed to question these flawed inputs.

Identity is at the core of this issue. Every AI-driven marketing effort assumes accurate identity for analysis and targeting, yet identity remains a fluctuating component in our data stacks. Consumers frequently move across devices and change profiles, making it tricky to track accurately over time. However, most systems treat a snapshot identity as a constant, and AI inherits this flawed assumption.

Additionally, not all data issues stem from outdated sources. Some are intentionally deceptive due to evolving fraud tactics, becoming more challenging to distinguish without additional context. Fraudulent behavior can significantly distort model outputs and performance metrics, creating a feedback loop where AI unintentionally perpetuates the very issues it should mitigate.

Traditional data strategies often focus on structure over substance, and clean data doesn’t equate to accuracy. AI demands an in-depth understanding of identity validity, activity authenticity, and risk awareness, which traditional strategies may overlook.

The illusion of AI readiness becomes apparent when dashboards show excellent match rates and models yield seemingly precise outputs. However, metrics of identity reachability and engagement accuracy become crucial yet often disregarded questions.

True AI readiness starts with ensuring that our data inputs are trustworthy. It focuses on verifying identity accuracy, validating meaningful activities, and acknowledging risks rather than simply accumulating data records.

By addressing these foundational elements, organizations can suppress low-value identities, optimize outreach, and mitigate misuse before it skews results. Over time, this creates a structural advantage for AI operations, leading to more reliable predictions and efficient campaigns.

I’ve come to understand that AI’s impact on marketing is undeniable, yet it cannot independently resolve inherent data challenges. Organizations need to prioritize and invest in understanding the integrity of their data systems.

The real question isn’t about applying AI but assessing whether our data is worthy of AI. This deeper level of scrutiny defines true readiness and distinguishes the truly prepared from those merely rushing ahead.

Inspired by this post on Search Engine Land.

April 20, 2026