An enterprise SEO agency spends most of its effort making search work executable inside a company that was never built to execute it. Finding the opportunity is the cheap half of this job. On a site carrying tens of thousands of addresses, a reasonable list of what would help usually already exists in a document somebody wrote eighteen months ago and nobody has shipped.

Five things the role covers once a site reaches that size.

1. Deciding what deserves engineering time. Not what would help, which is always a long list, but which handful of items justify displacing something a product team has already committed to.
2. Turning findings into specifications. A finding written as an observation dies in a backlog. The same finding written as acceptance criteria against a named template gets built.
3. Writing the standards other people work to. Merchandising, regional marketing, product and localization teams make hundreds of search-relevant choices a month without asking anybody. Those choices are either governed or they are noise.
4. Making the money case. Large budgets move on a model somebody in finance can check, and a projection with hidden assumptions gets discounted to zero the first time it is questioned.
5. Watching a market that changes without telling you. At this scale competitors move by template, and one of those changes can take revenue off a quarter before it surfaces anywhere.

Brand Vision was established in 2018, with more than 15 years of practice behind the founding team, and roughly 150 brands and 500 or more projects have gone through the studio since. The outside verification is 250 or more five-star reviews, a 5.0 on Clutch built from 64 or more recorded client interviews, four Awwwards, a Webby, RGD and CGD certification, and an A plus standing at the Better Business Bureau. Leads have risen an average of 250% across past programs, which is a record of completed work at very different scales and not a projection for yours, and nobody here will hand you a percentage before seeing your own numbers. You can read the practitioners who would run it, look through work already shipped at this size, and note that our search people sit with the group that deploys your code instead of filing tickets into somebody else's queue.

The uncomfortable part of the role is that much of it is not search work. It is writing, negotiating and standard-setting, and a firm that only knows the technical half will hand you one more document.

Nine, and no engagement has ever used all nine at once. Which of them get funded comes out of where your throughput is actually blocked, and on most accounts that is two or three places instead of nine.

Technical SEO at estate scale. Crawl and index control across hundreds of thousands of addresses, server log evidence of where crawl effort really goes, rendering, and speed measured per template instead of per page. The underlying discipline is the one described under technical SEO, and the unit of work is different, because nothing here gets repaired one page at a time.

Template and platform-level optimization. Titles, headings, markup, copy blocks and metadata written as rules against a named component, then proven on a subset before the rule reaches the whole estate.

Content inventory and consolidation. Every address scored against what it earns, what it costs to keep and what it competes with, ending in a keep, merge, redirect or retire decision with a named owner attached.

Internal linking systems. Links generated by rules inside components, monitored as a system, with the small number of editorial placements that genuinely need judgment still done by hand.

International and multi-region work. Which markets get funded, how their addresses are arranged, language annotations returned correctly by every version they name, and a decision per market about whether it justifies material written for it or an adaptation of something else.

Governance, standards and training. The written rules, checklists and review gates that let a distributed team make correct calls without us in the loop, plus the sessions that teach people why each rule exists instead of handing them a document.

Forecasting and business-case modelling. Ranges built from your own data, with every assumption sitting somewhere your finance team can change it and watch the output move.

Stakeholder and procurement support. Being in the rooms where this gets approved, answering security and legal questions properly, and giving your internal sponsor something they can defend when we are not present.

Executive reporting. One view for the people holding the budget and a deeper one for the people doing the work, both read against the forecast instead of against last month.

Two tracks usually run beside these instead of inside them. Editorial production at volume belongs with SEO content strategy, which is a different staffing question. Being cited inside generated answers is AI search optimization, and at this scale it is largely a templating and evidence problem, which is why it fits here better than most people expect.

Enterprise SEO is the same discipline running under four constraints a smaller site never meets, and the constraints matter far more than a page count does. Companies with four hundred pages sometimes need this. Companies with ninety thousand addresses are sometimes served perfectly well by a two-person retainer.

What changes once those constraints bite.

• Nothing can be done one page at a time. Improvement has to be expressed as a rule against a component and applied by the build, which turns a copy edit into an engineering request with a queue in front of it.
• The volume of decisions exceeds any single owner. Dozens of people publish continuously and each of them makes choices about addresses, headings, canonicals and navigation without knowing a choice was made.
• Permission is distributed. A title belongs to marketing, the template belongs to product, the address structure belongs to platform engineering, and regional copy belongs to a country manager who does not report to your CMO. One change needs four yes answers and any single no stops it.
• Mistakes are expensive in both directions. A correct rule earns across thousands of addresses at once and a wrong one loses across the same thousands, so the tolerance for shipping on instinct disappears.

What does not change, which is where money gets wasted.

• Demand research and intent still decide what is worth pursuing.
• Page quality still decides whether an eligible page wins.
• Independent authority still sets the ceiling on all of it.
• There is no separate algorithm for large sites, so buying the discipline in its general form buys most of what matters and the enterprise part is the delivery mechanism wrapped around it.

The proxies worth using instead of page count are unglamorous and better. How many distinct templates render your commercial pages. How many content management systems and platforms are in play. How many markets and languages. How many teams can publish without asking permission. And how long it currently takes to get a one-line change into production, which is the single most predictive number in the whole assessment. A retailer whose filter system generates hundreds of thousands of addresses is an ecommerce SEO problem at estate scale. A software business with 900 pages, six languages and a weekly release train is an enterprise problem with a small site attached to it.

Scope is built from your delivery constraints and never from your page count, because the honest drivers of cost are template count, platform count and market count. A 200,000 address catalogue rendered from fourteen templates on one platform is a smaller job than a 6,000 page corporate estate spread across four content management systems and nine countries.

How the phases usually run.

Access, inventory and baseline. Three to six weeks. A crawl configured for scale, server or content delivery network logs, analytics and Search Console across every property, a register of every template with its owner, and a benchmark of indexation, positions, revenue and crawl behaviour taken before anything is touched. Access alone routinely takes a fortnight, because it needs three departments to act.

Diagnosis and sequencing. Two to four weeks. Findings read against your competitors, your demand, your commercial priorities and your release calendar, then ordered by value against cost to ship. Every item names the team that has to do it, which is what makes the list usable instead of impressive.

Specification and pilot. Four to eight weeks. The top items written as build specifications, applied to one template group, and measured before the rule goes wider.

Rollout, governance and monitoring. A quarter onward, continuous. Paced by your release cadence far more than by ours, which is why we ask about it early.

What moves the number is mostly structural. More templates cost more than more pages. Each additional platform adds another set of constraints. Each market adds address structure, annotation work and localization judgment. Whether raw logs are obtainable changes what can be diagnosed. Whether a replatform is already in flight matters most of all, since auditing an architecture queued for deletion is spending you should not do, and a replatforming engagement is the better conversation. And how much execution sits with your team is the largest lever available and the one most clients underuse.

On timing, two things are true together. An indexation or rendering block can turn around in weeks once fixed. Everything else compounds, and the first movement worth showing a board lands between month four and month twelve. A commitment to first-page positions inside a month is a tell, meaning the vendor is either repairing something already broken or chasing phrases with no buyers behind them.

Ongoing work carries a six-month minimum, which at this scale is close to academic, since specification and rollout rarely finish inside two quarters. Every account, property and file stays in your name throughout. Where you want the picture before committing anything, a full diagnostic or a scoped review will size it for a small share of what a program costs.

Nearly every enterprise search program that stalls, stalls on calendar access and not on ideas, and the way through it is not escalation. A product organization does not decline your work because it disagrees with you. It declines because the request arrived with no size, no value and no owner, while everything competing with it arrived carrying all three.

Search work loses these arguments for reasons that are usually deserved. It gets described as best practice, which reads as an opinion. It carries no number, so a product manager weighing it against a feature with a forecast has nothing to compare. It is scoped loosely, so engineering cannot estimate it and will not commit to it. And it often arrives as a list of forty items, which reads as a project nobody has room to start.

What we bring instead is a model your finance team can pull apart. It starts from your own data. Current impressions and clicks by template group, a movement estimate grounded in what comparable changes have actually produced, a click-through curve applied to that movement, your own conversion rate for those page types, and your own margin or contract value. The output is a low, expected and high case, and every assumption sits in a visible cell somebody can change to watch the answer move. A model nobody is allowed to interrogate deserves to be discounted to zero, and it usually is.

Then the ask gets made small. One template, one sprint, a defined measurement window, and a number attached to it. Shipping one change that returns something is worth more politically than any roadmap document, because the second request comes from somebody with a result behind them.

Three tactics that work in practice.

1. Attach search work to releases already scheduled. A template being opened for a design change is the cheapest moment in the year to correct its markup, its headings and its linking.
2. Ask for a standing allocation instead of favours. A fixed slice of every sprint, agreed once at a senior level, removes the need to win the same argument every month.
3. Write acceptance criteria and not findings. Engineers build against criteria, so our specifications go into your tracker in the format your team already uses, which is why we work directly with whoever owns deployment.

The honest part. Sometimes the highest-value item genuinely will not be built this year. The right response is to plan around it, name the cost of the delay in the reporting, and stop repeating a request that has already been answered. Where the business needs volume inside that gap, paid search is the place to buy it and we will say so.

For some enterprise work we are the right team and for some we are plainly the wrong one, and the difference is worth naming before you brief us. Brand Vision is a senior group of practitioners with no bench of juniors underneath it, and that shapes what we are good for.

What a small senior team is genuinely good at. Diagnosis and sequencing across a large estate, where the value sits in judgment and the deliverable is a decision. Writing specifications an engineering team will accept without translation. Template and platform-level work, where four correct rules outperform a great deal of manual effort. Being in a room with your engineering leads and your executives in the same week without a handover between them. And any decision where being wrong is expensive, because that is exactly where seniority pays and where somebody else's junior layer costs you money.

What it is not good at. Production headcount, meaning sixty pages a month across nine languages with a translation workflow attached. Coverage across every time zone. Being your review queue. And any tender carrying a minimum staff count or a certification we do not hold, which we would fail honestly and would rather fail in week one.

The review queue point deserves the space, because it is the whole argument. A large organization makes hundreds of search-relevant decisions a month, and no agency of any size can approve them all without becoming a worse bottleneck than the one it was hired to remove. So the standard is the deliverable, and that is not a consolation prize. It means a written rule set, checklists built into the tools your teams already open, review gates on the small share of decisions that genuinely need judgment, and training that explains the reasoning so people can extend the rule to a case nobody anticipated. Roughly one decision in twenty needs us. The other nineteen need a rule and somebody who understands why it exists.

Two situations where you should hire somebody else, said plainly. If what you need is volume execution at a scale requiring forty people, a large agency or an internal team is the correct purchase, and that gets said in the first meeting. And if you already have a strong internal search lead, the highest-return version of us is a standards and specification engagement beside them instead of a full program, which costs considerably less and works better.

You can read the people who would be assigned and how we staff before committing anything, and where the honest answer is a smaller piece of everything the practice covers, that is what gets proposed.

A template change is a bulk edit that looks like a small one, so it gets released like code and never like a content edit. One rule on one component can rewrite the title on forty thousand addresses overnight, and a wrong rule is wrong forty thousand times before anybody notices.

Nothing that goes wrong here throws an error, which is why these failures reach production intact. A rule reads a field that is empty on a third of your records, so a third of your titles arrive truncated or half-formed. A pattern that reads well on one product reads as duplication across a category. Markup describes an attribute the page stopped displaying two releases ago. A canonical rule collapses two page types you needed kept apart. A heading rule strips out the one sentence on the page containing the words people actually search. All of it silent.

The sequence we use.

1. Write the rule against named fields, with a fallback for every empty case. A specification and not a suggestion, in the format your engineers already accept.
2. Generate the output before anything ships. Every affected address rendered into a file so a person can read the edge cases. This step finds most of the failures above and it costs a day.
3. Release to a cohort instead of the estate. A comparable group of pages inside the same template, matched on traffic and seasonality, with the remainder held back for comparison.
4. Hold the window. Four to eight weeks at minimum, longer where seasonality is strong, and no verdict inside the first fortnight.
5. Keep a one-deploy revert. Anything shipped without a way back is a change nobody will agree to test again.

The honest limit. Site-side testing in search is noisy. Cohorts are never truly identical, an algorithm update can land in the middle of your window, and the engine assesses the whole domain instead of your test group. So results get read as directional evidence and we still run them, because the alternative is changing forty thousand pages on somebody's opinion and the cost of a bad rule is measured in quarters.

Platform constrains all of this more than people expect. What a stack lets you control decides what a specification can even ask for, which is why we establish the crawl and index layer and the template register before proposing anything, and why a hard platform limit sometimes turns the conversation into a rebuild. Where a change alters what a visitor does instead of what a crawler reads, it needs behavioural testing as well as a search measurement, since a title that earns clicks and loses buyers is not a win.

On most large estates the fastest available gain is removal and the second fastest is merging. Adding pages to a site already carrying thousands nobody owns makes the underlying problem worse and charges you for the privilege.

Why estates end up this way is worth understanding before anybody starts deleting. Campaign pages built for a launch in 2019 and never retired. Product lines that no longer exist. Regional copies made by duplicating a page and editing the phone number. Four articles on one subject written by four teams who never met. Documents published as files instead of pages. None of it was unreasonable at the time and all of it now competes for the same finite attention.

The inventory is one table before it is a decision. Every address joined to what it earns, what it converts, how many internal links point at it, how many external links it holds, when it was last touched and who asked for it. That join is tedious and it is where the argument gets settled, because opinions about which pages matter rarely survive contact with a conversion column.

Then four decisions, applied against criteria agreed before anybody looks at a specific page.

• Keep and invest. Earning now, or clearly capable of it, and worth genuine work.
• Merge. Several pages circling one subject, consolidated into the address with the strongest history, with everything unique from the others carried across properly instead of appended at the bottom.
• Redirect. No longer needed, still holding links or still receiving visits, so the value moves permanently to the closest relevant page.
• Retire. Earning nothing, linked by nobody, better gone. Removed with a correct status code and taken out of the sitemap and the navigation at the same time.

Setting the criteria first is a political act as much as a technical one. Every page has a person who asked for it, and a deletion list with no published standard behind it becomes a series of individual arguments you will lose one at a time.

What removal buys is crawl attention concentrated on pages that earn, a clearer signal about what the site is about, fewer of your own pages competing with each other, and a smaller surface to maintain on every future change. What it does not buy is demand. Pruning removes drag and creates nothing, which is why it runs beside editorial work and independent coverage instead of in place of either. Where you want the size of the opportunity before committing to the work, an audit will measure it.

At estate scale internal links are a build feature with an owner, a specification and a monitoring dashboard, and your editorial team places almost none of them. A person can link thoughtfully across three hundred pages. Nobody links thoughtfully across sixty thousand.

The system has to deliver outcomes you can measure instead of intentions. Every commercially important page reachable within about three clicks of the homepage. Links concentrated on pages that earn instead of spread evenly. No orphans, meaning no page that nothing links to. Anchor text that describes the destination and varies naturally. And stability, so adding a thousand products next quarter does not starve the pages you care about.

Where the links come from once it works properly.

Breadcrumbs generated from a real taxonomy. The cheapest structural win available on a large site, and it depends entirely on the taxonomy being right, which is frequently the actual project hiding underneath the request.

Contextual modules driven by attributes. Related items chosen by category, use case, specification or region, and never by publication date. Recency-driven modules withdraw links from every evergreen page you own, and nobody notices.

Curated hubs. A small number of deliberately built pages that group a subject the way a buyer thinks about it. This is where editorial judgment still beats any rule.

Filter and pagination rules. Which narrowed views get linked at all. This is where a link graph either concentrates or dissolves.

Body copy links placed by people, on a few hundred pages. The deliberate exception, kept manual because it carries the most weight per link.

The failure modes are worth naming because they ship constantly. Footer blocks carrying eighty sitewide links, which pass almost nothing and dilute everything. Automated linking that produces the same anchor phrase a thousand times. Modules assembled in the browser, where a crawler that never runs the script never sees a single link. And modules that connect every page to every page, which flattens the site and tells a search engine nothing about what you consider important.

Reporting on it is the part almost everybody skips. Click depth distribution, internal links per page across your top two hundred commercial pages, orphan count, and the share of internal links pointing at revenue pages. All four move visibly when the system is correct.

One honest limit. Internal linking redistributes authority you already hold and creates very little of its own. On a site with no external standing it is a rounding error. The effect is largest on deep category and filter architectures and on estates carrying many near-identical place pages, a pattern documented in local SEO.

Most international search problems are funding decisions misfiled as tagging problems. Companies arrive asking how to configure language annotations across eleven markets when the useful answer is that three deserve investment and eight deserve one well-built page.

Prioritization comes first and it is arithmetic. Search demand in that market for the thing you actually sell. Your ability to fulfil there, including payment, shipping, support hours and any licence the category needs. Competition, which varies enormously by country for one identical product. Existing recognition, since an unknown brand starts a long way back against entrenched local players. And whether anybody in your company can answer a phone in that language. Score the candidates, fund a few properly, and stop pretending eleven thin markets are a strategy.

Then the structure decision, which is expensive to reverse.

Country domains. The strongest signal about which market a site serves, the most expensive to maintain, and each one begins its standing from almost nothing.

Subdirectories on one domain. Cheapest to operate, inherits the authority you have already built, weakest geographic signal of the three. Correct for most companies.

Subdomains. Between the other two on every axis, usually chosen for internal reasons instead of search ones.

Language and region annotations are the mechanical layer underneath, and they declare which version serves which language and country. Each version has to point at every other version including itself. The failures we find most are one-way declarations nothing returns, tags aimed at addresses that redirect or carry a no-index instruction, a missing default for visitors who match nothing, and template-generated tags listing translations that were never published. Two English pages for two countries is the most common duplication problem on any large estate, and these annotations are the only thing preventing it.

Localization past translation is where the return actually sits. People search in the words their own market uses, which are often not the words a translation memory produced. Units, sizes, standards, regulatory language, holidays and payment names differ. Proof has to be local, because a case study from another continent persuades nobody. And Spanish written for Mexico is not Spanish for Spain. So the decision per market is whether it earns material written for it or a careful adaptation of something existing, and that gets decided one market at a time.

Two adjacent notes. A bilingual domestic market carries every one of these problems inside one country, and our Montreal work settles address structure and language before anybody commissions a translation. And generated answers behave differently by language, so visibility has to be sampled per market instead of once in English, using the same consistency discipline that keeps business details matching across platforms.

Buying an agency inside a large organization is a compliance process before it is a marketing decision, and it routinely takes longer than the first phase of the work. Six to twelve weeks between a verbal yes and a kickoff is normal, and the teams that plan for it start earning a quarter earlier than the teams that do not.

How it lands on us, in order.

1. Sourcing. Sometimes a formal tender with weighted criteria, sometimes a sole-source justification your sponsor has to write and defend. We will tell you early which of those we can realistically win, because a tender with a headcount floor is not winnable by a firm our size.
2. Security and information technology review. A questionnaire covering where data lives, who holds access and how it is granted and revoked, which subprocessors are involved, incident response, retention and deletion, device management, and multi-factor or single sign-on requirements. We answer at the level we can genuinely support and we do not claim a certification we do not hold. Where a control is absent we name it and propose a compensating one.
3. Legal and contracting. A master services agreement, a data processing agreement, insurance certificates, liability caps, intellectual property assignment and confidentiality terms. We work from your paper where we can and mark up what we cannot accept instead of signing quietly.
4. Regulatory and accessibility review. Public bodies and universities and colleges commonly require a conformance statement before price is ever discussed, and clinics and health organizations add a privacy review that governs what your analytics may collect at all.
5. Access provisioning. Named accounts inside your own identity provider instead of shared logins, least privilege on every system, and a documented offboarding path agreed before the first credential is issued.
6. Commercial setup. Supplier registration, purchase order, invoicing format and payment terms. This sounds trivial and it delays more kickoffs than security review does.

How Brand Vision handles it. One named person owns your questionnaire and your paperwork, so your procurement contact is never chasing a shared inbox. We keep a standing pack of the documents this process asks for, maintained instead of assembled from scratch each time. Contracting can run through our Canadian entity or a US one depending on where your data and your legal team need it to sit, which is worth settling early and is part of why our head office arrangement is documented instead of improvised.

The honest note. Your internal sponsor has to defend this decision in rooms we are not in, so we write the summary they will actually forward instead of leaving them to build one out of a proposal.

Reporting runs at two levels, and the executive version answers three questions on a single page. What did we say would happen, what happened, and what changes now. Everything else belongs in the practitioner view underneath it.

What the executive view carries. Revenue and qualified pipeline from organic search read against the forecast, with the variance explained instead of merely displayed. The two or three initiatives that actually moved the number. What is blocked, who owns unblocking it, and what the delay costs per month. And the next decision leadership has to make, with a date beside it. No impression counts, because that total climbs whenever a page appears at position fifty for a query nobody acts on, which makes it the easiest figure to present and the least useful to read.

What the practitioner view carries. Performance by template group and by market instead of one blended line. Indexation and crawl behaviour tracked continuously. Positions across a defined set, including the terms we chose to leave alone and the reason. The internal linking measures. Specification and publishing throughput against plan, with any delay attributed honestly, because the wait is frequently on your side of the process and hiding that helps nobody. And technical regressions caught, which on a site shipping weekly is a monthly event.

Competitor monitoring, which at this scale is its own discipline. Watching positions alone is too slow to be useful. What gets tracked weekly is competitor title and markup patterns, new template deployments, movement in how many of their pages are indexed, new sections and hub pages appearing, and their share of a fixed keyword set. A rival changing one category template can take positions across thousands of terms inside a fortnight, and a monthly summary surfaces that in week six of a quarter already lost. Step changes trigger an alert instead of waiting for a report.

Three commitments hold the whole thing together. Every number ties to leads, qualified pipeline or revenue and never to activity volume. Where attribution is genuinely directional, that gets marked on the line instead of dressed up as a precise figure. And a bad quarter reaches you from us first, with the revised forecast and the reason it moved, because a model quietly restated is worse than a model that was wrong out loud.

Every task, specification and decision sits on a board open to you continuously, so the state of the program never depends on a meeting. You can see how this has read across previous programs, and where you want an outside opinion on what your existing reports leave out, send them over, including the case for spending the budget elsewhere.