Digital marketing

Search is where demand you already created gets collected

This is the part most often misread. Paid search rarely creates demand — it captures it. Somebody sees your television spot, hears the radio ad, gets the mailer, and then searches your category two days later. If you are not there, the click goes to whoever is, and the channel that created the demand gets blamed for not working.

Which is why we do not sell search as a standalone product bolted onto someone else's plan. It is the collection layer under everything else, and it should be budgeted as such. Everything below is planned the same way: as a line with a job, not a product with a price.

Search and discovery

PPC management

Pay-per-click is the one channel where the platform will happily run itself if you let it. Performance Max, broad match and fully automated bidding all work by taking decisions away from the advertiser, and they are genuinely good at finding volume. What they cost you is visibility — into which queries the money went to, which placements served, and which audience actually converted.

Active management is the work of putting those decisions back. In practice that means reading the search terms report every week rather than every quarter, adding negatives against the queries automation keeps buying, keeping campaign types separated so brand and non-brand performance can be told apart, and holding conversion actions to things that are actually worth money rather than the cheapest available click.

Two things worth saying plainly:

• Percentage-of-spend billing is a conflict of interest. An agency paid a share of your budget earns more when you spend more, whether or not it returned anything. We will tell you when an account should spend less.
• Microsoft Ads is smaller and usually cheaper. Lower volume, older audience, and in some categories a materially lower cost per lead. It is worth testing and it is not worth a separate retainer.

What PPC cannot do: create demand that does not exist, rescue a weak offer, or outrun a landing page that does not deserve the click. If the category has no search volume, the honest answer is that the budget belongs in Connected TV or direct mail instead.

SEO

Two different jobs get sold under one name. Organic site SEO is about whether your pages can be found and understood — structure, speed, content that answers a real question, and links from places that matter. Local SEO is about your Google Business Profile and the map pack, and it runs on a different set of signals entirely: correct primary category, review volume and recency, consistent name-address-phone data across the web, and proximity.

Proximity is the one nobody wants to say out loud. For map-pack queries, how close your business physically sits to the person searching is a ranking factor you cannot buy your way around. A remodeler in Lakeville will not outrank a closer competitor for “bathroom remodel near me” searched from north Minneapolis, at any budget. We do not sell rankings we cannot influence, and that is the clearest example of one.

What we can influence is the set of pages that exist to be found in the first place. That is why this site carries a page for each of the ten largest cities in the metro rather than one page claiming to serve everywhere — a real page about a real service area can rank; a list of city names in a footer cannot.

SEO is slow. It is measured in months, and anyone guaranteeing a ranking on a timeline is either buying ads and calling it SEO or has not told you which keyword they mean.

AEO — answer engine optimization

AEO is the work of making a page easy for an answer engine to lift a direct answer out of — Google's AI Overviews, featured snippets, voice assistants, Perplexity-style tools. The mechanics are concrete enough: structured data that tells a machine what kind of thing a page describes, headings phrased as the question a person actually asks, the answer stated plainly in the first sentence underneath rather than four paragraphs down, and enough consistency across the web that the engine is confident which business you are.

Here is the honest part. Most of what genuinely helps is good SEO plus proper schema markup. It is largely the same work, and an agency selling AEO as a separate discipline with its own retainer is charging twice for one job. The evidence base for tactics specific to answer engines is thin, changes constantly, and is mostly published by companies selling AEO services.

We do the parts that are defensible and cheap: schema on every page, question-shaped headings where a real question exists, and a plain answer directly under it. This site's pages carry Service, LocalBusiness and BreadcrumbList markup, and the Connected TV page carries FAQPage markup on its questions. That is verifiable rather than asserted — view the source.

GEO — generative engine optimization

GEO is the newer version of the same question: whether ChatGPT, Gemini, Claude or Perplexity mentions your business when somebody asks for a recommendation. It matters more every year and almost everything written about it is guesswork sold with confidence.

What is actually understood about the mechanism: when one of these tools answers a question about local businesses, it usually retrieves live results from a search index at the moment of asking rather than recalling something from training. Which means the search-visibility work is upstream of the answer. Being findable, correctly categorised and consistently described across the open web is what feeds the retrieval. Mentions on pages the index trusts do more than anything done on your own site.

What nobody can do is guarantee it. There is no ad slot, no bid, no ranking report and no API that reports your position. Visibility can be sampled — ask the same question repeatedly over weeks and record whether you appear — but that is a survey, not a ranking, and it moves when a model is updated for reasons no one outside the lab can see. Anyone selling guaranteed placement inside ChatGPT is selling something they cannot deliver.

So our position is deliberately unexciting: we do the upstream work, which is the same work that earns organic and map-pack visibility anyway, and we do not bill it as a separate product.

Paid media

Programmatic media buying

Programmatic means buying impressions through an automated auction: a demand-side platform bids on each impression as it becomes available, against a geography, an audience and a price you set. It is how Connected TV, most display and much digital video are actually purchased.

One widely repeated thing is now out of date. The industry spent years preparing for third-party cookies to disappear from Chrome. Google reversed that decision in April 2025 and shut down most of the Privacy Sandbox project in October 2025, retiring the replacement technologies it had spent six years building. Third-party cookies remain in Chrome today with no announced removal date. That does not mean cookie targeting is healthy — Safari, Firefox and Brave still block them, so a substantial share of traffic was never reachable that way and still is not. It means anyone still selling you an urgent cookie-deprecation strategy has not read the news in eighteen months.

The real problem with programmatic is not targeting. It is that money disappears inside it. Between your budget and the publisher sit a demand-side platform, an exchange, sometimes a reseller, and a verification vendor, each taking a percentage that is often undisclosed. A meaningful share of open-market inventory is made-for-advertising sites — pages built to carry ads rather than to be read — and they will absorb budget while reporting perfectly acceptable metrics.

What we insist on: the actual site list, not a category description; take rates disclosed in writing; exclusion lists applied before launch rather than after the first month's report; and reporting detailed enough to show where impressions ran. Our whole position is that plans are negotiated, not ordered, and programmatic is the channel where that matters most, because it is the easiest place in media to spend money invisibly.

Display advertising

Banner advertising, bought programmatically or direct. It is the cheapest inventory on a media plan and the easiest to sell badly, so it is worth knowing exactly what you are buying.

Start with what “viewable” means. Under the Media Rating Council standard, a display impression counts as viewable when at least 50% of the ad's pixels are on screen for one continuous second. For video it is 50% of pixels for two continuous seconds, and for very large formats the threshold drops to 30% of the area for one second. Most advertisers have never been told that this is the bar. A campaign reporting strong viewability is reporting that half of each ad was on screen for about as long as it takes to scroll past it.

Which tells you what display is for. It is a frequency and reminder layer — useful for staying present between television flights, and as the delivery mechanism for retargeting. It is not a direct-response channel, and a plan leaning on display to generate leads is usually a plan that ran out of better ideas.

We will often recommend less of it than you were expecting, or none. On a small budget it is frequently the weakest line available, and the same money in radio or direct mail does more.

Video advertising

Digital video means YouTube and in-stream placements elsewhere: skippable and non-skippable pre-roll, six-second bumpers, in-feed units and Shorts. It is bought through Google Ads or, for more control over placement and frequency, through a demand-side platform.

It is a genuinely different product from Connected TV, and the difference is not the file. CTV is a television commercial on the main screen in the house, generally non-skippable, sound on. Digital video is frequently a phone, held at arm's length, skippable after five seconds, sound off until something earns it. Both are worth buying and they do not do the same job.

What digital video is better at: reach per dollar, long-form explanation for a considered purchase, and intent — YouTube is a search engine, and someone searching how to fix or choose the thing you sell is a better prospect than anyone a demographic target will find. What it is worse at: holding attention it has not earned, and being measured honestly. Completion rate is a delivery check, not a result. A high completion rate on non-skippable inventory tells you the ad played, nothing more.

Paid social

On a digital plan, paid social's job is cheap targeted reach against a defined audience — the layer that builds familiarity before search has anything to collect. It sits between programmatic and search: broader than a keyword, more precise than a display buy.

The planning problem specific to running it alongside everything else is frequency. Each platform caps and reports frequency inside its own walls, so a household hit by Meta, YouTube and a programmatic display line has been reached three times while three dashboards each report a comfortable number. Nobody is over-serving on purpose; it happens because no single platform can see the others. Somebody has to hold the total, and that is a planning job rather than a platform setting.

The full detail on how we run it — creative fatigue, why recut broadcast spots underperform in-feed, and suppression — is on the social media page.

Targeting and audiences

Retargeting

Retargeting shows ads to people who have already been to your site, engaged with your content, or appear on a list you uploaded. Audiences are built from a site pixel, a customer file, or platform engagement, and served through display, video, social or Connected TV.

It is also the most over-credited line in digital advertising, and this is the useful thing to understand about it. Retargeting reports beautifully because it is aimed at people already in the market — they visited your site last week. Many of them were going to buy anyway. When your television and CTV campaigns create demand, retargeting is standing at the end of the funnel collecting the credit on a last-click report, and the channels that did the work look like failures. The question worth asking is not what retargeting's cost per conversion was. It is how many of those conversions would not have happened without it, which requires a holdout test, not a dashboard.

Three things most accounts get wrong:

1. No frequency cap. This is the difference between a reminder and the sensation of being followed, and the second one damages the brand it is promoting.
2. Converters not excluded. Without a suppression audience you keep paying to advertise to people who already bought.
3. One window for everything. A thirty-day window is right for some purchases and absurd for others. It should follow the actual length of the buying decision.

Also worth knowing: the reachable pool is smaller than the platforms imply. Safari, Firefox and Brave block third-party cookies, and Apple's app tracking permissions cut mobile reach further, so a retargeting audience is a fraction of your real traffic rather than all of it.

Geofencing

Geofencing draws a boundary around a physical place and targets devices inside it — or, more usefully, devices that were inside it recently. The location signal comes from apps that pass coordinates into the advertising bid stream. Accuracy varies enormously depending on the source: GPS from an app with permission is precise to a few metres, wifi triangulation is roughly building-level, and an IP address is often only accurate to a city. Vendors quote the best case. A plan built on the assumption that every impression is metre-accurate will not deliver what was promised, and we would rather set that expectation before the invoice than after.

Used well it is genuinely good at a few things: reaching people who visited a competitor's showroom, targeting attendees of a specific event or venue, and covering a tightly-defined trade area that a broadcast buy would overshoot by forty miles.

It also carries real legal exposure that most agencies never mention. Several states now restrict geofencing outright:

• Washington's My Health My Data Act makes it unlawful to use a geofence — defined there as a virtual boundary within 2,000 feet — in or around any facility providing in-person health care, in order to identify or track people seeking care, collect health data, or send them related advertising. There is no consent exception, and the act carries a private right of action.
• Connecticut prohibits a geofence within 1,750 feet of any mental health, reproductive or sexual health facility for those purposes. Nevada restricts geofencing around in-person medical facilities, and New York has passed a standalone prohibition covering health care facilities.

The Federal Trade Commission has been active on the supply side too. It has brought actions against several location data brokers — among them Outlogic, formerly X-Mode Social, and InMarket Media in 2024 — and in May 2026 announced a settlement with Kochava barring the sale of sensitive location data covering medical facilities, places of worship, schools and childcare, domestic violence shelters and military installations without express consent.

None of that makes geofencing unusable. It does mean we will not run a health-adjacent geofence, we will ask where a vendor's location data came from, and we will say no to a targeting idea that is legally exposed even when a vendor is willing to sell it.

Custom audience targeting

Custom audiences are built from data you already own. A customer list is hashed — converted to an irreversible string before it leaves your side — and matched against platform accounts. From there you can target those people, build a lookalike audience modelled on them, or, most usefully, exclude them.

Expect the match rate to disappoint. Lists go stale, people register with a personal address and buy with a work one, and plenty of customers have no account on the platform at all. A list of ten thousand rarely produces ten thousand reachable profiles, and an agency that does not warn you about that in advance is setting up a difficult conversation for later.

Suppression is the underused direction and the one we push hardest. Upload your customer list as an exclusion on an acquisition campaign and you stop paying to advertise to people who already bought. It costs nothing, it is available on every major platform, and in our experience most accounts arrive without it configured. On a business with meaningful repeat custom it is the single cheapest improvement available.

Two limits worth knowing. Lookalikes built from a small or messy seed list produce noise that looks like precision, so the seed matters more than the size of the audience it generates. And targeting is restricted by policy in some categories — housing, employment and credit advertising all carry limits on how narrowly an audience may be drawn, and health and financial-status targeting is constrained. Those are platform rules, not preferences, and they apply regardless of who is running the account.

The broader point: your own customer data is the only audience asset that does not change under you. Rented targeting comes and goes with platform policy, browser defaults and regulation. A clean, current customer list keeps working.

Reporting that survives a hard question

Every account we run reports the same way: what was spent, what came back, and what it cost per lead — month against last month, not a rolling average that hides a bad four weeks.

Two rules we hold to on reporting, because they are where most agency numbers quietly fall apart:

• A number we could not measure shows as “n/a”, never as zero. Those are opposite findings. One is a paused campaign; the other is broken tracking. A report that renders them identically will eventually be used to make a decision it never supported.
• Conversion tracking gets verified before results get read. Reporting zero leads for an account whose tag is not firing is a fabricated number arrived at by omission.

What we look at first in an underperforming account

1. Where the money actually went. Usually a handful of broad-match terms are taking a third of the spend and returning nothing. That is a search-terms problem and a negatives problem, not a bidding problem.
2. Whether the account is capped. Lost impression share to budget means more money buys more volume. Lost share to rank means it will not — that is ad relevance or the landing page. They call for opposite responses, which is why we never add them together.
3. Whether the landing page deserves the click. Traffic up and conversion rate halved is not a media problem.

What we do not do

We do not sell rankings we cannot influence, we do not report impressions as an outcome, and we do not run a channel we cannot explain the purpose of. Eleven disciplines are listed on this page and very few businesses should be buying all eleven — part of the job is saying which four matter for yours.

If digital is not the right place for the next dollar, we would rather say so and put it into Connected TV or direct mail.

Sources for the figures on this page: the Media Rating Council's viewable ad impression standard (50% of pixels for one continuous second for display, two seconds for video); Google's April 2025 announcement that it would not deprecate third-party cookies in Chrome and its October 2025 retirement of most Privacy Sandbox technologies; Washington's My Health My Data Act and the Connecticut, Nevada and New York geofencing provisions; and Federal Trade Commission actions against location data brokers including Outlogic (formerly X-Mode Social) and InMarket Media in 2024 and the Kochava settlement announced in May 2026. Verified 4 September 2026. Regulation in this area moves quickly — tell us if something here has gone stale and we will correct it.