Every channel available to a small service business costs money, hours, or both. Ranking them honestly means knowing what each one demands and how much evidence stands behind it. On both counts the marketing literature is worse than it looks.

Start with the numbers used to justify the ranking, because three of the most quoted do not survive being traced.

“46 percent of searches have local intent.” Attributed to Google everywhere. It appears in no Google publication. The chain ends at a social media post from 10 October 2018 in which a conference attendee reported that a Google representative had said it during a presentation. A search industry blog repeated the post the same day with no link to any document, and everyone since has cited Google directly. No study, no method, no sample, and no update in eight years.

“76 percent of people who search locally visit a place within 24 hours.” This one has a real citation, and reading it is instructive: “smartphone users=1000, local searchers=634, purchases=1,140, May 2016.” So it rests on 634 people, it is a decade old, and the page carrying it no longer loads. A different Google study from 2014 put the same figure at 50 percent on a different method.

“The average cost per lead in your trade is X.” No United States public institution publishes a cost per click or a cost per acquisition for any industry. Neither major advertising platform publishes benchmarks by sector; both route advertisers to account-specific forecasting tools instead. Every average you have been shown was assembled by an agency from the accounts it happens to manage, which is a convenience sample of its own client list.

That is not a reason to give up on comparison. It is a reason to compare on what you can observe yourself.

Provenance of three statistics commonly used to justify local marketing spendProvenance of three statistics commonly used to justify local marketing spend by small service businesses. The first, that forty six percent of searches have local intent, is attributed to Google in thousands of articles but appears in no Google publication. Its citation chain ends at a social media post dated the tenth of October two thousand eighteen in which a conference attendee reported that a Google representative had stated the figure during a presentation at Google’s offices; a search industry blog repeated that post the same day with no link to any document, and every subsequent citation attributes the figure to Google directly. There is no study behind it, no published methodology, no sample size and no update in eight years. Google’s own signed and dated document from August two thousand seventeen states instead that nearly one third of all mobile searches are related to location, citing internal data from April two thousand sixteen. The second, that seventy six percent of people who search locally on a smartphone visit a physical place within twenty four hours, does carry a real citation which reads smartphone users equals one thousand, local searchers equals six hundred and thirty four, purchases equals one thousand one hundred and forty, May two thousand sixteen. It therefore rests on six hundred and thirty four people, is a decade old, and the page carrying it no longer loads, while a separate Google study published in two thousand fourteen put the equivalent figure at fifty percent using a different method. The third, that the average cost per lead in a given trade is a specific number, has no institutional source whatsoever, since no United States public institution publishes a cost per click or a cost per acquisition for any industry and neither major advertising platform publishes benchmarks by sector, both routing advertisers instead to account specific forecasting tools. Every average in circulation was therefore assembled by an agency from the accounts it happens to manage, which constitutes a convenience sample of its own client base with no sampling frame and a strong selection bias.Three numbers, traced”46 percent of searches have local intent”A 2018 post reporting an unattributed remark at a conference. No study, no method, no sample,no update in eight years, and absent from every Google property.”76 percent visit within 24 hours”Real citation: 634 local searchers, May 2016. The page no longer loads, and a 2014 Googlestudy put the same figure at 50 percent.”The average cost per lead in your trade”No public institution publishes one. Neither platform publishes sector benchmarks. Everyaverage comes from an agency describing its own client list.
One traces to a conference post. One rests on 634 people and a dead page. The third does not exist in any public statistic. Source : Search Engine Roundtable, 10 October 2018; Google Purchased Digital Diary, May 2016; verified absence of any published CPC or CPA benchmark on Google Ads Help and Meta Business Help Center (2026)

What each channel actually demands

One arbitration sits upstream of this ranking and is treated separately: whether to buy demand from a lead platform or build your own channel. That is the subject of home services, renting demand or building it. What follows assumes you are building.

Since the published comparisons are unusable, compare on inputs instead. Every channel here costs something you can count before you commit to it.

ChannelMoneyYour hoursCompoundsEvidence behind it
Completing your listingnoneone afternoonyesone controlled study
Asking every customer for a reviewnoneminutes per jobyespolicy rules, no effect size
Referral from past worknoneongoing qualityyesnone published at this scale
Paid leads from a platformhigh, per leadlownono published effect size
Paid searchhigh, ongoingmedium to set upnorandomized evidence, small businesses
Organic contentnonehigh, sustainedyesnone published at this scale
Social postingnonehigh, sustainedpartlynone published at this scale

Two columns deserve explanation.

Compounds means the effort leaves something behind when you stop. A review sits on your profile permanently. A purchased lead does not exist the day after the job. Google’s own definition of the prominence ranking factor makes this concrete: it is “based on info like how many websites link to your business and how many reviews you have”, both of which accumulate.

Evidence is deliberately harsh. Most of these channels have no published effect size at this business scale, which does not mean they do not work. It means anyone quoting a number for them is quoting a vendor.

Customer acquisition channels for a small service business compared by cost, persistence and evidenceCustomer acquisition channels available to a small service business, compared by what each one costs in money and hours, whether the effort compounds by leaving something behind when it stops, and how much published evidence supports it. Completing a business listing costs no money and roughly one afternoon of time, it compounds, and it is supported by one controlled study published by Google. Asking every customer for a review costs no money and a few minutes per job, it compounds, and it is governed by published platform rules although no effect size has been published. Referral from past work costs no money and requires ongoing quality of delivery, it compounds, and no effect size has been published at this business scale. Buying leads from a platform costs a high amount per lead and little time, it does not compound since the relationship ends with the job, and no effect size has been published. Paid search costs a high ongoing amount and a medium amount of time to set up, it does not compound since a randomized experiment found effects ceased the moment advertising stopped, and it is supported by randomized evidence from an experiment on small businesses. Organic content costs no money and a high sustained amount of time, it compounds, and no effect size has been published at this scale. Social posting costs no money and a high sustained amount of time, it compounds partly, and no effect size has been published at this scale. The compounding column matters because it distinguishes effort that leaves an asset behind from effort that does not, and Google’s own definition of the prominence ranking factor makes this concrete by describing it as based on information such as how many websites link to a business and how many reviews it has, both of which accumulate over time. The evidence column is deliberately harsh, since most of these channels have no published effect size at this business scale, which does not mean they do not work but does mean that anyone quoting a number for them is quoting a vendor.Seven channels, four questionsChannelMoneyHoursCompoundsEvidenceComplete the listingnonean afternoonyesone studyAsk everyone for a reviewnoneminutes per jobyesrules onlyReferral from past worknoneongoing qualityyesnone publishedPaid leadshigh, per leadlownonone publishedPaid searchhigh, ongoingmedium to set upnorandomizedOrganic contentnonehigh, sustainedyesnone publishedSocial postingnonehigh, sustainedpartlynone published”None published” does not mean it does not work. It means anyone quoting a number for it is quoting a vendor.
Two columns decide most of it: whether the effort compounds, and whether any published evidence supports it. Source : Google Business Profile Help on prominence and reviews; Dai, Kim and Luca, Marketing Science 42(3), 2023; verified absence of published benchmarks on both advertising platforms (2026)

The one channel with a measured effect

Google ran a controlled comparison in July 2014 and published it: 1,000 respondents, five business categories including auto repair and hair salons, a control cell shown a minimal listing and a test cell shown a complete verified listing for the same business.

The difference in stated perception:

Statement, share who agreeMinimal listingComplete listing
Is a reputable business36%69%
Is well-established38%68%
Knows what it’s doing35%67%
Offers quality products or services32%66%
Is a business I would visit44%61%

Two caveats the document itself supplies. It was run in what it calls “a laboratory environment”, so these are perceptions rather than visits or jobs. And it was conducted in 2014 on the search results pages of 2014.

Even discounted for both, this is the strongest evidence available for the cheapest action on the list. Finishing a listing takes an afternoon and costs nothing.

Perception differences between a minimal and a complete business listing in a controlled comparisonPerception differences between a minimal and a complete business listing, from the only controlled comparison Google has published on the question. The study was conducted in July two thousand fourteen with one thousand respondents aged eighteen to sixty four across five business categories including a bakery, an auto repair shop, a hair salon or barber, a florist and a hardware store, using a twenty minute online survey. A control cell of four hundred and ninety four respondents was shown a minimal listing while a test cell of four hundred and ninety six respondents was shown a complete verified listing for the same business, with all differences reported as statistically significant at ninety five percent. The share agreeing that the business is a reputable business rose from thirty six percent with a minimal listing to sixty nine percent with a complete one. The share agreeing that it is well established rose from thirty eight to sixty eight percent. The share agreeing that it knows what it is doing rose from thirty five to sixty seven percent. The share agreeing that it offers quality products or services rose from thirty two to sixty six percent. The share agreeing that it is a business they would visit rose from forty four to sixty one percent. Two caveats are supplied by the document itself. It was conducted in what the document calls a laboratory environment, so the results are stated perceptions rather than measured visits or jobs won. And it was conducted in two thousand fourteen on the search results pages of two thousand fourteen. Even discounted for both limitations, this constitutes the strongest published evidence available for the cheapest action on a small service business’s channel list, since finishing a business listing takes an afternoon and costs nothing.Minimal listing against complete listingShare agreeing with each statement. Control n = 494, test n = 496.Is a reputable business36%69%Knows what it’s doing35%67%Offers quality work32%66%The two caveats the document supplies itself”a laboratory environment”, so perceptions rather than jobs. And 2014 results pages.
The cheapest item on the list is also the only one with a published controlled test. Perceptions, not jobs. Source : Google / Ipsos MediaCT, Impact of Search Listings for Local Businesses, August 2014, control n=494, test n=496 (2014)

What paid advertising did, and for whom

There is one randomized experiment on businesses this size, and it is worth more than any case study.

Free advertising was given to 7,209 businesses randomly drawn from 18,294, for three months. Page views rose 19 percent, direction requests 14 percent, telephone calls 7 percent, reviews 5 percent.

Three findings sharpen that.

The least known gained the most. Independent operators gained more than national chains with comparable attributes. Advertising works by informing, and there is nothing to inform someone of if they already know you.

Nothing carried over. Effects stopped the moment the advertising stopped.

And orders did not move measurably. Effects on bookings were not statistically significant, and the authors state they were underpowered on that outcome. The return figure quoted from this study elsewhere comes from an appendix calculation on pre-experiment tax data matched to 13 percent of the sample, which the authors themselves call “a back-of-the-envelope calculation”.

Why you cannot settle this with a test

The reasonable response is to measure your own channels properly. The arithmetic is against you, and someone has computed exactly how much.

Across 25 large field experiments, each with more than half a million users, researchers calculated what precision those experiments achieved. To distinguish a highly profitable campaign from a break-even one, the median campaign would have to be nine times larger. To resolve a ten percentage point difference in return, which is an ordinary threshold for a spending decision, it would have to be 62 times larger, which the authors describe as “nearly impossible for a campaign of any realistic size.”

That is for advertisers spending millions. For a business running a van, formal measurement of return is simply not available.

What is available is cruder and still useful: switch one channel off for a fortnight and count the calls. It will not give you a return on investment. It will tell you whether the phone notices.

Sample sizes required to measure advertising return on investment with useful precisionSample sizes required to measure advertising return on investment with useful precision, from a study of twenty five large field experiments conducted with major retailers and brokerages, each covering more than five hundred thousand unique users and collectively representing two point eight million dollars of digital advertising spend. The authors computed what precision those experiments actually achieved and what would be required for a decision. To reliably distinguish a highly profitable campaign delivering a fifty percent return from one that merely breaks even at zero percent, the median campaign in their sample would need to be nine times larger. To resolve a ten percentage point difference in return on investment, which is an ordinary threshold for an investment decision, the median campaign would need to be sixty two times larger, which the authors describe as nearly impossible for a campaign of any realistic size. Their summary statement is that informative advertising experiments can easily require more than ten million person weeks, making experiments costly and potentially infeasible for many firms. The underlying reason is variance in individual purchase behavior, which commonly has a coefficient of variation of ten. A campaign delivering a twenty five percent return had to raise mean sales per person by thirty five cents on a variable with a mean of seven dollars and a standard deviation of seventy five dollars, giving an explanatory power on the order of five millionths. The median standard error on return on investment across the retail experiments was twenty six point one percent, producing a confidence interval roughly one hundred percentage points wide. For a business operating a single van, formal measurement of advertising return is therefore not available at any level of effort or discipline. What remains available is cruder and still useful: switching one channel off for a fortnight and counting the calls, which will not produce a return on investment figure but will indicate whether the telephone notices.How much bigger the test would need to beComputed on 25 field experiments, each with more than half a million usersTo tell +50 percent return from break-even9x largerTo resolve a 10-point difference in return62xThe authors’ own summary”informative advertising experiments can easily require more than 10 million person-weeks”What is left for a one-van businessSwitch one channel off for a fortnight and count the calls. Not a return figure. A signal.
The threshold is not effort or discipline. It is sample size, and the required size is out of reach at this scale. Source : Lewis and Rao, The Unfavorable Economics of Measuring the Returns to Advertising, Quarterly Journal of Economics 130(4), 2015 (2015)

A ranking that survives the evidence

Ordered by evidence and by what stays behind, not by what a vendor recommends.

1. Finish the listing. Free, one afternoon, and the only item on the list with a published controlled test behind it. Google’s implicit-category rule also means you can stop adding categories: selecting a specific one automatically includes the broader ones above it.

2. Ask every customer for a review, and script nothing. Free, minutes per job, and it accumulates on the one ranking factor Google describes as reputation. Ask everyone rather than only the satisfied ones, which is both the platform rule and the federal test.

3. Make the work referable. No published effect size exists at this scale, and it remains the channel with the lowest cost per job in almost every trade. Turning up when you said you would is a competitive position in a field where four out of five operators have no employees at all.

4. Buy leads to fill a gap, and budget them as fuel. They work, they cost per unit, and they leave nothing behind. Useful for a slow month, dangerous as a foundation.

5. Run paid search once the first three are in place. The experiment says it moves calls and that the gain concentrates on people who did not already know you. It also says the effect stops when the money does. When those three are done and the diary still has gaps, that is the point at which our B2B paid acquisition work is worth buying, with the spend read against calls per week rather than against a sector average.

6. Treat content and social as long projects or not at all. Both cost hours rather than money and both compound, but nothing published measures their effect at this scale. Start them because you can sustain them, not because a figure said so.

And keep one habit that costs nothing: whenever a number is used to justify a spend, ask where it came from. Three of the four figures most commonly used in this market do not survive that question.

Recommended order for building customer acquisition channels in a small service businessRecommended order for building customer acquisition channels in a small service business, ranked by the strength of published evidence and by whether the effort leaves an asset behind rather than by vendor recommendation. First, finish the business listing, which is free, takes about one afternoon, and is the only item on the list supported by a published controlled test, noting that Google’s implicit category rule means additional categories are unnecessary since selecting a specific category automatically includes the broader ones above it. Second, ask every customer for a review and script nothing, which is free, takes minutes per job, and accumulates on the one ranking factor Google describes as reputation, remembering to ask every customer rather than only the satisfied ones since that is simultaneously the platform rule and the federal test. Third, make the work referable, for which no published effect size exists at this scale but which remains the channel with the lowest cost per job in almost every trade, and noting that turning up when promised is a competitive position in a field where four out of five operators have no employees at all. Fourth, buy leads to fill a gap and budget them as fuel rather than as equipment, since they work, cost per unit, and leave nothing behind, making them useful for a slow month and dangerous as a foundation. Fifth, run paid search once the first three are in place, since the randomized experiment shows it moves calls with the gain concentrated on people who did not already know the business, and also shows the effect stops when the money does. Sixth, treat content and social posting as long projects or not at all, since both cost hours rather than money and both compound, but nothing published measures their effect at this business scale, so they should be started because they can be sustained rather than because a figure recommended them. One habit costs nothing and should be kept throughout: whenever a number is used to justify a spend, ask where it came from, since three of the four figures most commonly used in this market do not survive that question.Build in this order1Finish the listingFree, one afternoon, one published controlled test2Ask everyone for a reviewFree, minutes per job, accumulates on prominence3Make the work referableFree, ongoing, lowest cost per job in most trades4Buy leads to fill a gapBudget as fuel. Useful for a slow month, not a foundation5Paid search, after the first threeMoves calls. Stops when the money stops6Content and socialLong projects. Start them because you can sustain them
The first three cost no money. The last three cost money or a lot of hours, and only one of them has evidence. Source : Synthesis of the sources cited throughout this article (2026)