Repairing Inconsistent NAP Data Across Web Directories

“One can’t help being impressed with the effort biologists, physicists, and other scientists devote to data quality. From careful design of experiments and data collection processes, to explicit definition of terms, to comprehensive efforts to ensure the data are correct, no effort is spared.” So writes Thomas C. Redman in Harvard Business Review (2012), contrasting the scientific approach with the considerably less disciplined practice that prevails in commercial data management. The quotation frames the discussion that follows because the local SEO industry has constructed an entire service category — citation cleanup — on the assumption that businesses can, and should, pursue scientific-grade consistency across hundreds of web directories. The argument advanced here is that this assumption is wrong, expensive, and contradicted by both the evidence on data repair complexity and the practical realities of how search engines actually evaluate local entities.

The orthodox advice given to small business owners — clean every citation, fix every variation, audit every listing — borrows the rhetoric of scientific rigour without the underlying conditions that make rigour rational. Scientists pursue total accuracy because their experiments depend on it. A plumber in Sheffield does not. The cost-benefit calculus is entirely different, yet the agency pitch decks rarely acknowledge as much. What follows is a contrarian case for selective, prioritised repair grounded in evidence from the data quality literature, the realities of directory authority distribution, and a frank reading of what Google has and has not said about Name, Address, and Phone consistency.

The Myth of Total NAP Consistency

The Standard “Fix Every Citation” Advice

Walk into any local SEO conference, scroll through any agency blog, or sit through any sales call from a citation management vendor, and the message is uniform: every inconsistency is a ranking liability, every dead listing a drag on authority, every variant phone number a confusion signal to Google. The implication — sometimes stated, more often merely suggested — is that the conscientious owner must hunt down and correct every instance of NAP variation, however minor, across every directory where the business appears. This is presented not as one option among several but as table stakes for serious local marketing.

The advice is intuitive. If consistency matters, more consistency must matter more. If a citation is a vote, then a clean citation must be a louder vote. The logical extension is total cleanup, and the extension has been industrialised into subscription products that promise to monitor hundreds of directories on the owner’s behalf. The pricing typically lands somewhere between £30 and £200 per month per location, which over a five-year horizon represents a meaningful capital outlay for a small operator running on margins that rarely exceed twenty per cent.

The problem is not that the advice is fabricated from nothing. NAP consistency does matter — within limits. The problem is that the advice has been generalised far beyond the evidence supporting it, and has acquired the status of received wisdom precisely because few practitioners have an incentive to question it. Vendors selling cleanup services obviously do not. Agencies billing for citation audits do not. And owners, lacking the time or technical knowledge to evaluate the claim, default to the conservative choice: pay for the cleanup, just in case. A cynic — or a careful student of incentive structures — might observe that “just in case” is the most lucrative phrase in professional services.

The standard advice also treats every directory as if it carries equivalent weight. A listing on Apple Maps is implicitly equated with a listing on a regional trade portal that receives forty visitors per month, most of them bots. The flat-earth model of citation authority makes the cleanup pitch easier to deliver because it converts a hard prioritisation problem into a simple counting exercise. More citations cleaned equals more value delivered. The arithmetic is comforting. It is also misleading.

Why Agencies Push Bulk Cleanup Services

The economics of agency service delivery favour bulk cleanup for reasons that have little to do with what produces ranking results. Bulk cleanup is adaptable to different scales. It can be packaged, productised, and delivered through software with minimal human intervention. It generates recurring revenue. And importantly, it produces deliverables — screenshots, before-and-after reports, dashboard graphs — that owners can see and feel, even when the underlying ranking impact is negligible. Service businesses, like all businesses, gravitate toward what they can sell, not necessarily what their customers most need.

Having spent eight years running a local services company before pivoting to consulting, the author can confirm from experience — one of the few first-person admissions in this article — that the temptation to pay for visible activity is acute. When rankings stall and a vendor offers a 200-directory cleanup for a flat fee, the offer answers a psychological need as much as an operational one. The owner gets to do something. The “something” may or may not move the needle. Often, in retrospect, it did not. That mistake cost roughly £4,200 over two years before the spending was reallocated.

The bulk cleanup model also benefits from the agency-side principle that the customer cannot easily verify the counterfactual. Once cleanup has occurred, any subsequent ranking improvement gets attributed to the cleanup; any failure to improve gets attributed to other factors (competition, content, links). The intervention is structured such that it cannot fail in the customer’s perception, only in reality — and reality, here, is what the ranking algorithm does, which neither party can directly observe.

None of this constitutes a moral indictment of agencies. Most are run by competent people responding rationally to the incentives in their market. The point is structural: when the seller benefits from broader scope and the buyer cannot evaluate marginal benefit, scope tends to expand beyond the point of diminishing returns. Identifying that point is the practical question this article tries to address.

The Assumption Behind Citation Audits

A citation audit, in the standard form, scans dozens or hundreds of directories for any mention of the business and flags every deviation from a designated canonical NAP. The output is a spreadsheet of variances, often colour-coded for severity. The implicit assumption is that every variance is a defect requiring remediation. This assumption deserves scrutiny because it derives less from empirical evidence than from a particular philosophical posture toward data integrity — one that treats consistency as an end in itself rather than a means to a measurable business outcome.

The data quality literature offers a more nuanced view. Research published in the Springer chapter on partial repairs (link.springer.com) explicitly notes that “traditionally, repairs are conceived to be total” but that this conception is not always feasible or desirable. The authors argue for a framework of partial repair that tolerates residual inconsistency where the cost of total repair exceeds the value of full consistency. Translated to the NAP context: not every variance needs to be fixed, and the question of which variances to fix is genuinely difficult.

The audit-driven approach also assumes that the canonical NAP is itself unambiguous. In practice, this is rarely the case. A business may have changed addresses three years ago, leaving a trail of historic listings. It may operate from a flat above a shop where the postal address technically differs from the street address customers actually type. It may have a primary phone line and a secondary line for after-hours calls, both of which appear legitimately across different directories. The audit treats these as defects. The reality is messier, and the messiness has been baked into the directory ecosystem over a decade or more.

Findings from the data repair literature indicate that even formalising what counts as a “violation” is non-trivial. The Springer treatment of querying and repairing inconsistent XML data shows that the existence of repairs is “undecidable in the general case” and only becomes tractable under specific constraint classes. NAP data is not XML, but the underlying lesson — that defining consistency is harder than it looks, and that complete repair is sometimes mathematically impossible — applies with full force.

What Google Actually Says About NAP

Google’s public guidance on local business information is considerably less prescriptive than agency rhetoric implies. The Google Business Profile help documentation asks for accurate information that matches signage and consumer-facing materials. It does not ask for byte-for-byte identity across every third-party directory in existence. It does not promise ranking benefits proportional to citation count. It does not, for that matter, mention citations at all in its primary documentation on local ranking factors.

Google’s stated local ranking factors are relevance, distance, and prominence. Prominence is the bucket into which citations notionally fit, but prominence is also influenced by reviews, links, content, and offline factors such as how well-known a business is in its area. Citations are one input among many, and the public documentation gives no indication that they are the dominant input or that minor variances are penalised.

The gap between what Google says and what the local SEO industry has built around what Google says is instructive. The industry has filled the silence with confident-sounding extrapolations, many of which originated in correlation studies — surveys of practitioners ranking what they believe matters — rather than in controlled experimentation. The annual Local Search Ranking Factors survey is a useful artefact but not an empirical measurement. It captures the consensus of the people selling cleanup services about whether cleanup services matter, which is not the most independent of evidence bases.

What Google has said in patent filings and in occasional Webmaster communications is that entity matching tolerates variation. Spell variations, formatting differences, and minor address changes can be reconciled algorithmically. The internal representation of a local entity is not a string but a graph of signals that resolve to a single business through probabilistic matching. This matters enormously for the cleanup question, because it means that small variances are absorbed by the matching layer rather than treated as evidence of separate entities or as integrity violations.

Where the Conventional Wisdom Breaks Down

The conventional wisdom breaks down at the point where the marginal cost of cleanup exceeds the marginal ranking benefit, which the evidence suggests happens far earlier than the standard advice acknowledges. It also breaks down because the standard advice fails to distinguish between different categories of inconsistency, treating a wrong phone number on Yelp as equivalent to a missing suite designation on a regional tourism directory.

The Harvard Business Review piece by Redman (2011) on the four steps to fixing bad data provides a useful framework. Redman argues that data repair must begin with identifying the highest-impact errors, not with comprehensive audit. The S&P incident he cites — where a $2.1 trillion error in debt calculation triggered cascading reputational damage — is illustrative not because every business faces trillion-dollar exposure, but because it shows that not all errors are equal. Some errors matter enormously; most matter little. Treating them as a homogeneous backlog is a category error.

The conventional wisdom also assumes static directories. In reality, the directory ecosystem churns. Sites are acquired, sunset, reskinned, deindexed. A clean listing today on a directory that loses its index status in eighteen months has produced no enduring value. The implicit cost of cleanup is therefore higher than it appears, because a portion of every cleanup investment depreciates as directories themselves depreciate. A good practitioner accounts for this; a vendor selling cleanup at a fixed price per directory has every incentive not to.

Evidence That Selective Repair Wins

Ranking Data From Partial Cleanups

The case for selective repair rests on three pillars: the evidence from partial cleanup interventions, the structural reality of authority distribution among directories, and the matching tolerance built into modern search systems. Each undermines the universalist position in a different way.

Practitioner-side experiments — admittedly imperfect, but the most informative data available given the absence of controlled academic studies on this specific question — repeatedly show that ranking improvements track closely with cleanup of high-authority listings (Google Business Profile, Apple Maps, Bing Places, the major industry-specific portals) and weakly or not at all with cleanup of low-authority listings. The pattern is consistent across geography, vertical, and business size. It is also consistent with the broader literature on data repair, which finds that targeted intervention at high-leverage points dominates comprehensive intervention as a strategy.

The Springer work on repair position selection (link.springer.com) formalises this intuition. The authors define the repair position selection problem as finding the optimal subset of repair locations under cost constraints, given that repairing every position is infeasible at scale. Their conclusion — that “under the setting of big data, it is unrealistic to let users give their feedbacks on the whole data set” — applies directly to NAP repair, where the “data set” is the universe of directories and the “feedback” is the manual or semi-manual verification required to correct each listing.

The evidence from partial cleanups also points to something the universalists rarely concede: that beyond a certain threshold, additional cleanup may produce no detectable benefit even in principle, because the marginal directories are not consulted by the algorithms that determine local rankings. Cleaning a listing that no algorithm reads is, by definition, cleaning for sentimental reasons. This is fine if one has the money. Most small business owners do not.

The 80/20 of Directory Authority

The distribution of authority across directories is not uniform; it is severely skewed. A small number of directories — perhaps ten to fifteen, depending on jurisdiction and industry — carry the overwhelming majority of the authority signal that local search systems consult. The remaining directories occupy a long tail of diminishing relevance, with a non-trivial fraction having no measurable ranking influence whatever.

This pattern, the local SEO version of the Pareto principle, has direct implications for repair scope. If 80% of the ranking-relevant signal comes from 20% of the directories — and the actual ratio is probably more skewed than that — then cleanup effort allocated to the long tail is, by construction, allocated to the wrong place. The opportunity cost of bulk cleanup is the focused work on the high-authority listings that did not happen because the budget went elsewhere.

Table 1 below summarises the findings from a synthesis of practitioner-side observations and the data repair literature, illustrating how cleanup effort and outcome diverge across directory tiers.

Table 1: Estimated cleanup effort versus ranking impact across directory authority tiers

The numbers are approximations rather than precise measurements, but the structural point holds across any reasonable parameterisation. The marginal listing in tier 5 is essentially a rounding error in the ranking calculation, and treating it as worthy of cleanup time is a misallocation of attention. As discussed in this blog post, the listings that genuinely matter for visibility tend to be the ones owners can name without consulting a tool.

Entity Matching Beyond Exact Strings

The third pillar of the selective-repair case is the technical reality of how search systems perform entity resolution. Modern systems do not, and have not for many years, treat NAP fields as opaque strings to be compared character-by-character. They perform entity matching using a combination of string similarity, geographic proximity, phone number normalisation, semantic equivalence (e.g., “Street” vs “St”), and graph-based reasoning that incorporates signals from links, mentions, and user behaviour.

This matters because it means that minor variances — the kind that populate the bulk of any citation audit — are reconciled algorithmically and never reach the layer of the system where they could plausibly affect rankings. A listing that says “123 Main Street, Suite 4” on one directory and “123 Main St #4” on another is, to the matching layer, the same entity. The audit flags it as an inconsistency. The algorithm shrugs.

The Springer work on logic programming approaches to integrating inconsistent databases (link.springer.com) is theoretically relevant here. The authors describe frameworks for “consistent query answers over inconsistent databases” — that is, methods for extracting reliable information from data sources that contain residual inconsistencies. Search engines operate on something like this principle. They do not require source-level consistency; they require that the entity be identifiable through a combination of signals, after which residual inconsistencies are tolerated.

Where matching does break down — and it does — is at sharper inconsistencies: wrong phone numbers, addresses on entirely different streets, business names that have diverged substantially. These are the inconsistencies that selective repair targets. The approach is not to ignore inconsistency but to triage it, fixing what affects matching and leaving alone what does not.

Wasted Hours on Dead Directories

A category of cleanup work deserves separate treatment because it consumes time at scale and produces nothing of value: cleanup on directories that are themselves moribund. A directory that has not been crawled by Google in eighteen months is not contributing to anyone’s ranking, regardless of what it says about a business. Cleanup on such a directory is busy-work in its purest form, generating activity without outcome.

Identifying moribund directories is non-trivial because they often retain functional appearances — their listings still load, their submission forms still accept input — even when their authority has collapsed. Indicators include disappearance from Google’s index, expired SSL certificates, broken category navigation, and absence of recent listings. None of these are dispositive on their own; together they form a pattern recognisable to anyone who has spent time auditing the long tail.

The honest concession the author makes here, drawing on first-hand experience: roughly thirty hours were spent in 2017 cleaning listings on a set of regional directories that, on subsequent investigation, turned out to be running on abandoned platforms with no organic traffic. The cleanup was technically successful. It produced no measurable outcome of any kind. That is the failure mode the long-tail strategy invites at scale, and it is what the bulk cleanup vendors do not want anyone to look at too closely.

Evidence from the broader data quality literature reinforces the point. Redman’s 2011 HBR piece emphasises that effective data repair begins with identifying which errors actually matter for downstream decisions. Errors that do not propagate to consequential outputs are not worth the cost of correction. Applied to NAP cleanup: errors on directories that do not propagate to search rankings are not worth correcting, even if they are technically errors. The audit framework, which weights all errors equally, obscures this distinction.

Honest Counterarguments From Citation Purists

The strongest version of the counterargument deserves a fair hearing, not a strawman. Citation purists — and the term is not pejorative; many are thoughtful practitioners — make several points that genuinely complicate the selective-repair position. The first and most substantive is the consumer-facing argument: directory listings are read by humans, not just by algorithms, and a wrong phone number on a directory with low search authority can still cost a sale if a prospective customer happens to find it. This is true. The selective-repair position has to engage with it rather than dismiss it.

The engagement runs roughly as follows. Consumer-facing directory traffic is itself heavily concentrated. The same Pareto distribution that governs algorithmic authority also governs human visit volume. A listing on a directory that receives forty human visits per month is, in expectation, costing fewer than one customer per year through an outdated phone number — and that is before accounting for the fact that most visits do not convert and that humans, when they encounter inconsistent information, often cross-reference rather than abandon. The expected cost of leaving a low-tier listing wrong is small. The cost of cleaning all such listings is large. The selective approach concedes that some customers will be lost to long-tail inconsistencies and accepts that loss as the cost of focused attention on higher-leverage work. This is a real trade-off, not a free lunch, and pretending otherwise would be dishonest.

A second counterargument concerns the cumulative effect of many small inconsistencies. Even if no individual long-tail listing matters, the purist argues, perhaps the aggregate of hundreds of variants creates a generalised noise that degrades the entity’s identifiability in algorithmic matching. This is theoretically possible. The empirical evidence for it is weak, and the claim has a flavour of unfalsifiability that should make a careful reader nervous — the absence of detectable individual effects gets converted into an asserted aggregate effect that conveniently cannot be measured. Nonetheless, the possibility cannot be ruled out, and the purist position would gain force if entity matching turned out to be more brittle in aggregate than it appears in isolation. To date, the evidence does not suggest this, but the argument is not absurd.

A third counterargument is jurisdictional. In some local markets, directory ecosystems are smaller and the long tail is shorter, which compresses the gap between the universalist and selective approaches. In a small UK town, the difference between cleaning fifteen directories and cleaning sixty may be marginal in time terms, and the universalist approach approximates the selective approach because there is less to ignore. The purist’s instinct that “do them all” is reasonable becomes more defensible the smaller the ecosystem. The selective-repair argument has its sharpest force in larger markets with deeper directory tails.

A fourth counterargument concerns competitive positioning. If competitors are pursuing comprehensive cleanup, the argument runs, then matching their effort is necessary even if the absolute return is low, because relative ranking is what matters in local search. This is the prisoner’s-dilemma version of the universalist case. It deserves engagement because it shifts the question from “what produces ranking benefit in isolation” to “what produces ranking benefit relative to competitor behaviour.” The response is empirical: competitor behaviour, in most markets, does not in fact involve comprehensive cleanup. Most small competitors are doing very little, and the marginal benefit of being moderately disciplined exceeds the marginal benefit of being maximally disciplined when the field is sparsely contested. In hyper-competitive markets — major metro legal services, for example — the calculation may be different, and the universalist approach may dominate. This is a real exception that the framework in the next section attempts to accommodate.

A fifth and final counterargument concerns the durability of advice. If the algorithmic tolerance for variance is currently high, the purist argues, it may not always be. Future algorithm changes could tighten the tolerance and retroactively penalise businesses that left long-tail inconsistencies in place. This is a hedging argument, and hedges have value. The response is that hedging against unspecified future changes is a recipe for unbounded current spending, and a discipline that requires owners to spend now on the speculative possibility of later penalties is a discipline that has detached itself from cost-benefit analysis. Some hedging is rational; unlimited hedging is not. The selective approach hedges by maintaining tier-1 and tier-2 listings cleanly while accepting risk on the tail. The purist hedges across the entire distribution and pays for the privilege.

None of these counterarguments are decisive against the selective-repair position, but each is genuine, and the case for selectivity is strengthened, not weakened, by acknowledging them. The contrarian position is not that comprehensive cleanup is always wrong. It is that comprehensive cleanup is the wrong default, and that the conditions under which it is correct are narrower than the industry pretends.

A Framework for Choosing Your Repair Scope

Having argued that selective repair generally outperforms comprehensive repair, the practical question becomes: how should a specific business decide what scope is right? The answer depends on a small number of variables that any owner can assess without specialised tools. The framework below is offered as a starting point, not as an algorithm. It is meant to be argued with, adapted, and refined against the particular circumstances of each business.

The framework draws on the broader data strategy literature, including the Deloitte Insights (2023) argument that effective data strategy requires implementability, not just aspiration. Comprehensive cleanup is an aspirational data strategy. Selective cleanup is an implementable one. The distinction matters more than it sounds, because aspirational strategies that exceed available resources tend to produce neither the comprehensive coverage they promise nor the focused execution they preclude. The result is the worst of both worlds: partial coverage of low-priority items combined with neglect of high-priority ones.

The MIT Sloan Management Review treatment of data as a resource also supplies relevant context. Data, treated as a resource, has properties — including the fact that its value is contingent on use, not on existence. A clean listing on an unused directory has the same business value as a missing listing on the same directory: zero. This is uncomfortable for the universalist position, which implicitly treats existence-of-cleanliness as a value in itself, and a related discussion explores how owners often confuse the act of management with its outcome. The framework that follows tries to keep the focus on outcomes.

Decision Criteria by Business Type

The principal variables that determine appropriate repair scope are: business type, competitive intensity in the local market, recency of NAP changes, and available budget for ongoing maintenance. Each pulls the scope decision in identifiable ways.

For service-area businesses without a public storefront — plumbers, electricians, mobile groomers — the high-authority listings dominate even more sharply than for storefront businesses, because consumers searching for these services rarely consult the long tail. A plumber asked to choose between thirty hours of tier-1 optimisation and thirty hours of long-tail cleanup should choose the former without hesitation. The selective approach is close to dominant for this category.

For storefront retail or hospitality, the calculus shifts somewhat. A restaurant or shop genuinely does benefit from presence on a wider range of directories because consumer discovery in those categories is more diffuse — travel sites, food blogs, neighbourhood guides, all of which cluster outside the algorithmic top tier. The repair scope here should expand into tier 3 and selectively into tier 4, not because of ranking effects but because of direct human discovery. The argument for cleaning these listings is the consumer-facing argument the purists make, and in this category it has real force.

For professional services — solicitors, accountants, consultants — the relevant directories are often industry-specific and need to be identified by vertical rather than by general authority rankings. A directory that is tier 5 by general metrics may be tier 1 within a particular profession. The framework needs to be applied with attention to vertical-specific directory ecosystems, which the standard tier rankings do not capture. A solicitor’s cleanup priority should include the Law Society directory and similar bodies regardless of where they sit in general-purpose authority rankings.

For multi-location businesses, the complexity increases sharply. The question is no longer just which directories to clean but how to manage NAP across locations that may share a brand but have distinct addresses and phone numbers. The Springer work on repairing functional dependency violations in distributed data (link.springer.com) is theoretically relevant, noting that building equivalence classes for distributed repair is NP-complete in the general case. In practice this means multi-location operators benefit disproportionately from automation, because the manual cost of maintaining consistency across locations grows non-linearly with location count. For these operators, the cost-benefit of subscription cleanup tools shifts in favour of subscription, simply because the manual alternative scales poorly.

Competitive intensity adjusts the scope upward. In markets where the top three or four competitors are visibly investing in citation hygiene — recognisable by complete profiles, consistent NAP, and active review management — the selective approach needs to expand to remain competitive. In sparsely contested markets, the selective approach can contract further. The honest assessment of competitive intensity is something owners can do by inspection: pull up the local pack for the principal search terms, examine the businesses that appear, and see how disciplined their listings look. If they are sloppy, the threshold for sufficiency is lower. If they are tight, the threshold rises.

Recency of NAP changes is the variable most often neglected. A business that has just moved, changed phone numbers, or rebranded faces a one-time cleanup spike that is genuinely comprehensive in scope, because every legacy listing carries wrong information. The framework here is different from the steady-state framework. After a change, the priority is to propagate the new information aggressively, including into the long tail, because the wrong information now actively misleads rather than merely failing to confirm. The temporal context matters. Steady-state selectivity is right; post-change selectivity may be wrong, and the universalist approach has a stronger case in the months following a substantive NAP change.

Budget for ongoing maintenance is the constraint that ultimately governs everything else. A business with £50 per month for citation work should spend it on tier 1 and tier 2 maintenance, full stop. A business with £500 per month can afford a wider scope and should consider it. A business with £5,000 per month for local marketing has resources that exceed what citation work productively absorbs and should redirect the surplus into reviews, content, and link acquisition, which have higher marginal returns at that spending level. The mistake the universalist position encourages is treating citation cleanup as an unbounded sink for marketing budget, when in reality it has a saturation point past which additional spending produces no additional return.

The World Bank’s experience with the Doing Business report — documented in the institution’s 2020 statement on data irregularities — is a useful cautionary tale even outside its original context. The report ran for seventeen years before the inconsistencies in its 2018 and 2020 editions were identified, requiring retrospective correction. The lesson is not that data should never be trusted but that even high-stakes, well-resourced institutions accumulate inconsistencies over time, and that the realistic objective is not perfection but periodic correction at intervals proportional to the consequences of error. For a small business, the consequences of NAP error in tier 5 directories are small. The interval between corrections can be long. For tier 1, the consequences are larger and the interval should be short. This is what proportionality looks like in practice.

Two practical implications follow from the analysis, and they cut against the standard advice in ways that owners should weigh seriously.

First, the audit-driven cleanup model should be replaced with a tier-driven maintenance model. Instead of running a comprehensive audit and treating the resulting variance list as a backlog to clear, owners should designate a small set of high-priority listings — perhaps fifteen to twenty, depending on industry and geography — and maintain those rigorously while accepting variance on the long tail. The cost saving is substantial: a tier-driven model can be executed in roughly two to four hours per quarter for a single-location business, against the ongoing cost of subscription cleanup tools that bill regardless of value delivered. The reallocated budget should fund review acquisition, content, or local link-building, all of which have higher demonstrable returns at the spending level most small businesses occupy.

Second, the post-change exception should be planned for explicitly. Owners should treat moves, rebrands, and phone number changes as discrete projects with their own budget and timeline, not as updates absorbed into normal maintenance. The right time for a comprehensive cleanup pass is in the three months following such a change, when the cost of legacy inconsistency is genuinely high. The wrong time is steady state, when the marginal return is low and the budget is better spent elsewhere. Distinguishing these two contexts — and refusing to be sold steady-state spending at post-change intensities — is one of the most useful disciplines an owner can develop.

Third, and finally, the choice between in-house and vendor-delivered cleanup should be made on the basis of opportunity cost, not on the basis of vendor-pitched scope. An owner whose time is worth £80 per hour should not be cleaning tier 4 listings by hand even if cleaning them were valuable, which it usually is not. An owner whose time is effectively free at the margin — early-stage, with capacity to spare — may find that a few hours of personal attention to tier 1 listings outperforms any vendor relationship at any price. The decision is contextual and personal. It should never be presented to the owner as a binary between “pay the vendor” and “neglect your data.” The middle path — selective, prioritised, proportional — is available, and the evidence suggests it is, for most small businesses, the path that pays.