00 HOW TO USE THIS

Start with the right problem

Demand is always the problem. Even when it’s not. Every startup, enterprise, coffee shop, and multi-billion dollar monopoly wants more demand. Forever. It’s never enough. Yet most demand programs fail before execution begins because the diagnosis of what’s missing and how to increase demand is incorrect. Twenty minutes of rigorous diagnosis here saves a quarter of misdirected effort downstream.

There are only three ways your demand engine can be constrained, and each demands the opposite intervention. If that sounds confusing, well, that’s why you’re not generating the demand you had hoped and why you’re here.

You may have insufficient demand: everything that reaches you converts at a reasonable rate, but volume is inadequate. Your pipeline is healthy and undersized. Adding conversion-rate optimization to this constraint achieves nothing, because the bottleneck sits upstream of every rate you would optimize.

You may have insufficient capture: interest exists, but it escapes. Prospects arrive, evaluate, and leave. Investing in more attention at this stage is expensive performance. You are paying to fill a vessel with a known leak.

Or you may have insufficient truth: you cannot distinguish between the first two because your metrics describe attribution, not causation. This is the most common condition and the least diagnosed, because the dashboards appear sound. Section 05 explains why.

Complete these questions before proceeding. They require more honest effort than they appear to.

Diagnostic: before anything else

Where your answers point:

The majority of readers place themselves in the third row while assuming they belong in the first. This reflects no failure of analytics capability; it is inherent to how attribution functions, and Section 05 provides the evidence.

Which numbers you can trust

This guide originated from a research initiative to establish defensible benchmark ranges for demand generation. It uncovered a different issue.

Most industry benchmarks have no source at all. Not weak sourcing, zero sourcing. The pattern is remarkably consistent: a precise figure surfaces without a named study, twenty other sites republish it (some referencing each other, some referencing nothing), and within a few years it becomes accepted truth. No one ever measured it.

Three illustrations, each traced to its earliest appearance:

This is why every number in this guide is tagged.

What this guide will not tell you

The following figures have almost certainly been presented to you as fact. Each was pursued to its origin during this research pass. None has an origin.

• The number of touches required to close a B2B deal. Every source publishing a number (eight, twelve, fifteen) does so without any citation.
• Win rate by inbound, outbound, and partner-sourced pipeline. The most requested breakdown in demand generation, and no study exists.
• “Six to ten stakeholders” in a buying committee. Commonly attributed to a major analyst firm; no report containing that range with a disclosed sample could be located. Section 05 instead uses that firm’s actual measured figure.
• “Buyers spend 17% of their time with suppliers.” This appears to derive from an illustrative graphic that divides six buying activities into roughly equal portions. Seventeen percent equals one-sixth. Secondary sources cite 5%, 17%, and 20% interchangeably, which reveals the problem.
• Cost per acquisition for content and SEO. The channel where most companies concentrate effort has no credible public economics.
• Attribution models “diverge by 30–50%.” Directionally accurate, factually unsourced. Section 05 substitutes three field experiments that properly quantified divergence.

01 CREATION AND CAPTURE

Two different jobs, one budget

Capture monetizes demand that already exists. Creation establishes awareness that a problem is worth solving. Nearly all budgets overweight capture, not because it performs better but for entirely different reasons.

Demand capture engages people who already recognize the problem, understand they have it, and are actively seeking a solution like yours. Search ads, comparison pages, review sites, most outbound, and every referral you’ve ever closed constitute capture. The demand pre-existed. You harvested it.

Demand creation helps someone recognize a problem is worth solving, or that a superior approach exists, before active search begins. It operates more slowly, resists attribution, and determines how much demand will be available to harvest in six months.

Most budgets skew toward capture not because capture is more effective, but because capture is measurable within a quarterly review cycle. Search reports in a week. Creation reports over quarters, through a measurement system Section 05 will demonstrate is structurally blind to it. When one metric moves and another doesn’t, budget flows to movement.

This produces a distinct failure mode worth naming because it resembles success until it doesn’t: you compete on bid price in an auction for buyers who have already selected a category, while remaining invisible to everyone who hasn’t. Your cost per lead appears defensible. Your growth is not.

Capture is a tax on demand you already created. If you haven’t created any, you’re paying that tax on someone else’s demand.

What the effectiveness research says

The strongest evidence on the creation-capture balance does not originate in B2B marketing. It comes from advertising effectiveness research, which most demand generation discourse overlooks.

Les Binet and Peter Field’s The Long and the Short of It, published by the IPA in 2013, analyzed 996 case studies encompassing approximately 700 brands across 83 categories spanning roughly thirty years. [1] The campaigns generating the greatest business impact allocated approximately 60% to brand building and 40% to sales activation. Their 2018 follow-up, Effectiveness in Context, refined this to 62/38 overall and, more usefully, demonstrated substantial variation by category.

The optimal ratio is not a single number. It shifts according to purchase consideration:

Field, Effectiveness in Context, IPA (2018), refining Binet & Field (2013). Basis: the IPA Effectiveness Databank, 996 cases. [1]

The more considered and infrequent the purchase, the more the optimal mix shifts toward building mental availability well before the moment of need, because that moment is rare, brief, and saturated with competitors.

And the case against it

The 60/40 principle is advertising’s most influential benchmark, and one of the field’s most respected voices considers it flawed. Omitting that critique would violate this guide’s standards.

Byron Sharp of the Ehrenberg-Bass Institute has described it as “terrible, very misleading.” His objection is methodological and substantive: the IPA databank consists of award entries. Campaigns are submitted by agencies, selected for win potential, and written as persuasive narratives for juries. This is not a random sample of advertising effectiveness; it is a sample of advertising agencies believed would win awards. As Sharp frames it: if you wanted to properly answer this question, you would never use this data source. [1]

A second issue is causal direction. A brand-heavy budget correlates with strong business performance in the databank. It is entirely plausible that strong performance provides the confidence and resources to fund brand investment, rather than brand investment driving performance. The methodology cannot distinguish these explanations.

Field’s public response to Sharp characterized the criticism as “tacky self-publicising.” Readers may assess that themselves.

People wanted a number.

Sharp, on why 60:40 spread as fast as it did.

The B2B cut, and why it says the opposite of what you’d expect

Binet and Field subsequently analyzed B2B cases within the same databank, in research commissioned by the LinkedIn B2B Institute covering cases from 1998 to 2018. The result contradicts most second-hand citations: the B2B optimum emerged at approximately 46% brand and 54% activation. [2] Activation commands a larger share in B2B than the 60/40 aggregate, not smaller.

Three qualifications must accompany that figure in the main text, not footnoted away.

• The authors themselves label the finding tentative. Fewer than fifty B2B cases exist in the databank. They acknowledge this explicitly.
• LinkedIn’s own assets contradict each other. The report cites 46/54. LinkedIn’s marketing page for the same research cites 50/50.
• LinkedIn sells B2B brand advertising. The research is rigorous and the authors credible. The sponsor still holds a commercial position.

The legitimate application of this finding is narrow but still valuable: it argues against importing a B2C ratio into a B2B plan, and demonstrates that the optimal mix is category-specific rather than universal. It is not a target to adopt.

Most of your market is not buying today

The case for creation in B2B rests on a straightforward observation about buying cycles. If organizations replace a given system every four or five years, then in any quarter the portion actively in-market is small. All capture spend competes for that sliver, while creation spend addresses the remainder.

This is commonly referenced as the 95:5 rule: roughly 5% of B2B buyers are in-market at any moment. It originates from John Dawes at the Ehrenberg-Bass Institute, published via the LinkedIn B2B Institute. Dawes himself has stated on record that the figure is not intended as precise. It is a heuristic communicating that most businesses are not buying in most periods. [2]

He also publishes a method to calculate your own figure, which represents a more productive use of twenty minutes than debating whether 5% is exact.

Your category’s in-market share

The foundations underneath

The 95:5 framing is a heuristic. The research underlying it is not, and distinguishing the two matters:

Your split

Before Section 02 evaluates whether your channels can carry your position, establish how you actually allocate spend across each job. Most teams find the result surprising, and the surprise consistently points one direction.

Guidelines for completing this honestly: a channel qualifies as capture if the buyer arrived already searching for something in your category. Branded search is capture. So is most content that answers a query someone typed. So is any outbound sequence premised on the problem already existing. If uncertain, ask what belief the buyer had to hold beforehand for the activity to function. If the answer is “that they have this problem,” it’s capture.

Your creation / capture audit

The PDF prints this as a worksheet with 11 blank rows to fill in.

02 CHANNEL–POSITION FIT

A channel is a format before it is an audience

Every channel requires the buyer to hold a prior belief. If your positioning seeks to change that belief, the channel cannot carry it, regardless of its performance elsewhere.

Channel selection is typically framed as an audience question: where does my buyer spend time? That question is valid, but secondary. The primary question is what the channel format requires the buyer to already understand.

Search requires that the buyer can articulate the problem. Not your product, but the problem itself, in language sufficiently aligned with yours to trigger a keyword match. If your positioning argues the buyer has misdiagnosed the problem, no query exists to purchase. No one searches for a category they don’t know exists, and no bidding strategy corrects that.

Paid social demands value legibility within roughly two seconds, without context. It is the strongest available channel for interrupting someone not actively searching, which makes it the natural home for demand creation. It is also merciless toward positions requiring a paragraph to convey.

Content and SEO require active research behavior and, increasingly, selection by an answer engine. Events require a problem worth traveling to discuss. Outbound requires that you can identify the problem externally before any buyer disclosure. Partner channels require credibility transfer from another party, which only occurs when your position is adjacent to theirs rather than reframing it.

You cannot buy a keyword for a category the buyer doesn’t know exists.

The fit matrix

Positions listed vertically, channels horizontally. Read across a row: these are the channels your position can and cannot support. Position types follow the taxonomy in The Ultimate Positioning Playbook; if you’ve completed that work, your row is already clear.

Cannot is intentional and literal: the channel format renders the position inexpressible, and increased budget worsens rather than improves performance. Strains indicates feasibility with disproportionate effort, typically requiring a supporting asset to carry the core argument.

Three ways this goes wrong

Illustrative constructs, not case studies. They are composites rather than specific companies, written to make the failure patterns recognizable.

What channels cost

Median cost per click and cost per lead from Google Ads, by category. This represents the most thoroughly documented channel cost data available in B2B: disclosed sample, defined window, medians rather than means.

WordStream / LocaliQ, 2026 Google Ads Benchmarks. n = 13,474 US search campaigns, April 2025 – March 2026, minimum 52 campaigns per category, medians. WordStream sells PPC management. [2]

Focus on the direction column. Cost per lead declined in three of four categories year over year. The assumption that paid acquisition becomes monotonically more expensive is not supported by the best-sampled data. If your paid costs are rising, that is specific to your account. Identify which component is driving it.

Channel–position audit

One row per funded channel. The third column is decisive and most frequently skipped.

Audit: every channel currently funded

The PDF prints this as a worksheet with 9 blank rows to fill in.

Any row marked “no” in the final column represents spend acquiring something other than what you believe you’re buying. That alone does not mandate cutting it. In some cases the correct response is to build the asset that supplies the missing belief. It should never remain accidental.

03 THE ALLOCATION MODEL

Deriving a split rather than copying one

What organizations actually invest, where those dollars flow, and how to construct a defendable range for your own budget from four inputs you already possess.

There is no single correct marketing budget. There is a defensible one, and what makes it defensible is that you can show how you derived it.

What companies actually spend

Two longitudinal surveys track this figure. They disagree, and that disagreement is more instructive than either number alone.

Nine percent versus 7.8% is not a contradiction to reconcile. They survey different populations with different instruments: Gartner’s panel is almost exclusively large enterprise, while the CMO Survey covers a broader size distribution and separates B2B from B2C. These are two distinct time series, not two points on a single line. Conflating them is how spurious trends are manufactured.

The principle generalizes beyond this example: when benchmarks conflict, examine the panel before deciding which is wrong. Usually, neither is.

Acquisition budgets run 26% larger than retention budgets, while retention outperforms.

The CMO Survey, 35th edition, January 2026. n = 308. [1]

That gap is the most quietly damning finding in the entire research review. It is not an argument to cut acquisition; Section 01’s penetration evidence points the other way. It is evidence that almost no one has actually derived their split from first principles.

Where the money goes

Within the marketing budget, allocation across people, agencies, technology, and media has shifted consistently in recent years.

Gartner CMO Spend Survey, 2024–2026 editions, n ≈ 400 per year. Gartner sells advisory services to this audience. Figures before and after 2024 reflect a definitional change and should not be read as a continuous series. [2]

More than two-thirds of media investment is now digital. The more significant shift is martech falling to a ten-year low while labor rises: after a decade of buying tools to perform the work, budgets are moving back toward people to operate them.

Deriving your own split

This is a derivation, not a recommended allocation. By this page you have, or will shortly have, all four required inputs.

The output is a range, and it should be. A point estimate here would be false precision masquerading as rigor. Each input is itself a range, and honest arithmetic on ranges yields a range. What this derivation delivers is not a single number, but the ability to articulate what must change for the number to change, which is the only thing that survives a finance review.

Allocation: derived, not copied

The PDF prints this as a worksheet with 10 blank rows to fill in.

04 THE MATH

Backwards from revenue

Every demand number you require derives from one target and four rates. Most teams own the target and borrow the rates from someone else’s business.

The chain runs in a single direction, and it runs backwards from revenue.

Illustrated with round numbers, for a business selling at a $40,000 average contract value in a mid-market motion:

The coverage figure of 4.2× emerged from the arithmetic. No one selected it, and it is not 3×. That is the subject of the next page.

Coverage, derived rather than quoted

Every board requests 3× pipeline coverage. The rule’s origin is undocumented: it is widely attributed to a named research firm, yet no such publication could be located. UNSOURCED You don’t need the rule, because the arithmetic behind it is one line.

If pipeline P at win rate w must produce revenue X, then P × w = X, therefore P = X ÷ w. Required coverage is simply 1 ÷ your win rate.

Three times is what that equals at a 33% win rate. Nothing more. If your win rate isn’t 33%, the 3× rule isn’t conservative or aggressive. It’s arithmetic describing a business that isn’t yours.

Arithmetic, not a benchmark. Dave Kellogg has published the same relationship, noting that 3× is what most boards default to requiring. [2]

Win rate and cycle length by deal size

Both rates in the derivation move with deal size. The relationship is consistent across every source reviewed, even when absolute values differ.

Optifai, n = 939 B2B SaaS companies, deal-level CRM data, updated April 2026. Vendor panel data, not independently audited. Coverage column is arithmetic. [2]

And how long capital is deployed before it returns:

The two sources differ by roughly 50 days at the median, and the reason is instructive rather than disqualifying: they start the clock at different points and survey different populations. ICONIQ’s panel is growth-stage and well-capitalized, which is not the market median. Use the shape from both (cycle length increases steeply with deal size) and your own data for the level.

An enterprise motion at a 15% win rate requires 6.7× coverage and six months of cycle. Plan for one and you will miss the other.

Stage conversion, and why these numbers are soft

The rates that convert demand into opportunities are the least reliable numbers in this guide, and it’s worth understanding precisely why before applying them.

First Page Sage, Sales Funnel Conversion Rate Benchmarks: 2026. Client data 2017–2025, roughly 65% B2B. No sample size has ever been published. First Page Sage is an SEO and demand generation agency. [2]

The companion finding, from Forrester: inquiry-to-closed-won in a conventional lead-centric process runs below 1%. [2] If that appears low relative to the table above, it is because the table isolates each stage while the 1% figure compounds them across a real population that includes everyone who never reached stage one.

What you can afford to pay

The ceiling on acquisition spend is defined by how long you can afford to wait for payback. Payback benchmarks with distribution, not a median that obscures it:

High Alpha, 2025 SaaS Benchmarks Report, n = 800+, self-reported, 69% US. High Alpha is a venture firm; the report is the successor to OpenView’s, which concluded when that fund wound down after 2024. [2]

Interquartile ranges are enormous. At $20–50M ARR the middle 50% of companies spans eleven to twenty-seven months. Any single “good payback” figure conceals that dispersion. Payback also lengthened from 2022 into 2024 in longitudinal data, contrary to the efficiency-era narrative. [2]

On the 3:1 rule

You will be asked about LTV:CAC. Understand what you’re being asked: the 3:1 target originates from a single blog post circa 2009, whose author characterizes it as “a very rough rule of thumb” and justifies it by noting a handful of public companies exceeded it. UNSOURCED No dataset underpins it, and no methodologically disclosed source has ever published the observed distribution of realized LTV:CAC across real companies.

You don’t need it. You already have retention and payback in this spread, and those are measured. Net revenue retention shows a median of 101% with the middle half at 96% to 109%; gross retention median 91%, middle half 82–95%. [2] A business retaining 110% annually can rationally pay more to acquire a customer than one retaining 95% at the same payback target. You can calculate precisely how much more from your own numbers instead of importing a ratio from a blog post.

Your engine

The complete chain, blank. This is the page to bring into a planning meeting.

Work it backwards

If line six is materially larger than your assigned target, you’ve identified the conversation to have before the quarter begins, not after it ends.

05 MEASUREMENT

Measurement without lying to yourself

Three randomized field experiments, executed at scale and published in peer-reviewed journals, converge on the same conclusion: attribution systematically overstates advertising’s impact, and short windows systematically miss its impact in B2B.

This section contains the strongest evidence in the guide with the fewest caveats. Everything below derives from experiments where spend was actually withheld from randomly selected populations and the difference measured, rather than from modeling, a survey, or a vendor’s platform data.

The shared finding, stated once then demonstrated three times: attribution identifies who was present at the sale. It does not identify who caused it. Those are distinct questions, and only one carries economic value.

The eBay experiment

In 2012, eBay paused paid search advertising in 68 of 210 US media markets for sixty days, retaining 142 markets as controls. Findings were published in Econometrica.

Blake, Nosko & Tadelis, “Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment,” Econometrica 83 (2015), 155–174. [1]

Observational methods estimated the advertising returned forty-one dollars per dollar spent. The experiment found it lost money. The gap is not a rounding error or modeling preference. It is the difference between measuring correlation and measuring causation, on the same company’s data, in the same period.

Branded keywords performed worse: no measurable benefit whatsoever. The ads primarily purchased clicks that would have arrived via organic results regardless. The mechanism the authors identified explains it completely. Attributed conversions were dominated by frequent existing customers who would have purchased anyway. The ad was present at the sale. It did not drive it.

The Facebook experiments

A larger, more general study. Researchers executed fifteen randomized field experiments on Facebook, encompassing roughly 500 million user-experiment observations and 1.6 billion ad impressions, then compared experimental ground truth against the observational methods commercial attribution tools employ.

Gordon, Zettelmeyer, Bhargava & Chapsky, “A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook,” Marketing Science 38(2) (2019), 193–225. [1]

Across all fifteen: in half the studies, every observational method was wrong by a factor of three or more. Adding covariates did not reliably improve accuracy. Richer demographic and behavioral controls failed to recover the experimental result. This bias is not a data volume problem you can collect your way out of.

Attribution tells you who was present at the sale, not who caused it.

Your attribution window is too short

The third experiment matters most for B2B because it was conducted in B2B.

Researchers ran geography-based switchback experiments with a semiconductor manufacturer, randomizing display advertising timing across regions for a high-consideration industrial product, the kind requiring engineering evaluation. Published in Management Science.

Thomas, Goic & Kalyanam, “Long Lags and Large Returns: Experimental Evidence from Advertising to Businesses,” Management Science (2024), DOI 10.1287/mnsc.2023.02661. Single-company field experiment. [1]

Take the second row seriously, because it indicts a near-universal practice. A thirty-day attribution window on a ninety-day sales cycle is not conservative. It is systematically incorrect, and incorrect in a specific direction: it undercounts precisely the slow-acting, difficult-to-attribute activity Section 01 argues you already underfund.

This creates a loop worth naming. Creation work pays back late. The measurement window closes early. The activity is therefore logged as ineffective and defunded. That reduces demand available to capture next year, which increases capture costs, which is then misread as ad market inflation rather than a self-inflicted consequence.

Models disagree, and the platforms have conceded it

Beyond the experiments, two peer-reviewed findings and one factual event are worth keeping at hand.

• Attribution models produce materially different channel contribution estimates from identical data. An individual-level study across display, search, referral, and email, validated against a field experiment, found differences significant, and found retargeting reduced conversion probability for certain journey patterns, which no last-touch report would reveal. Li & Kannan, Journal of Marketing Research 51(1) (2014), 40–56. [1]
• Last-touch attribution systematically over-credits early-funnel publishers relative to causally optimal allocation, distorting bidding as a result, though efficiency loss magnitude depends heavily on conditions. Berman, Marketing Science 37(5) (2018), 771–792. [1]
• Google removed first-click, linear, time-decay, and position-based attribution from Google Ads and GA4 in October 2023, retaining only data-driven and last-click. Its stated rationale (fewer than 3% of conversions used the deprecated models) reveals how little trust anyone placed in rules-based models. [1]

Note the absence of a figure quantifying how far models diverge. The ubiquitous “attribution models vary by 30–50%” line traces to nothing. UNSOURCED The experiments above measure the divergence that actually matters, between every model and ground truth.

What is genuinely invisible

A controlled experiment published sixteen tracked URLs across eleven networks and measured what analytics reported for approximately 1,100 resulting visits.

SparkToro with Really Good Data. 16 tracked URLs, 11 networks, roughly 100 panelists, 1,113 visits over 10 days. Small panel, run by an analytics vendor. Cite for mechanism, not as population estimate. [1]-adjacent

Every private conversation about your company that concludes with a click arrives labeled “direct.” This is not dark social theory; it is a measured property of how referrer headers function in messaging applications.

What remains unestablished is aggregate magnitude. No rigorous study quantifies total gap in B2B, and the widely cited “90% measurement gap” figure does not appear on the page it is attributed to. UNSOURCED What that source actually reports: across 620 conversions over twelve months, attribution software credited 78% to web search while a mandatory “how did you hear about us” field credited 85% to dark social. [2] One company, non-experimental, free-text self-report, from a consultancy whose model is validated by the conclusion. Use that, named, or use nothing.

How buying actually happens

The first row replaces the ubiquitous “six to ten stakeholders,” which has no locatable source. UNSOURCED Five to sixteen is broader and less satisfying. It is also the range someone actually measured.

What to do instead

A hierarchy, ranked honestly by what each method can and cannot establish.

On modeling: open-source marketing mix modeling is materially more accessible than previously. Meta’s Robyn launched in R in 2021 with a Python version announced in December 2024; Google’s Meridian became generally available in January 2025; PyMC-Marketing continues active development. [1] These are legitimate tools.

The question no credible source has answered is the minimum spend below which this approach ceases to function. Vendors quote thresholds ranging from a few hundred thousand to five million annually, with modeling service vendors quoting the lowest floors. UNSOURCED No vendor-independent threshold exists. If your spend shows little variation across periods and conversions are sparse, the model will still produce an answer. You should be skeptical of it.

06 WHAT COMPOUNDS

Building demand that survives the click

The mechanism most demand capture relies on is measurably eroding. What endures is the work Section 01 identified as late-paying.

For twenty years, the default B2B demand plan assumed a stable mechanism: someone has a problem, searches, finds a page, clicks. Every layer of the conventional playbook rests on that assumption. It is now shifting, and by the standards of this guide the evidence is exceptionally strong: one independent research organization and two vendors with transparent methodology, all in agreement.

Read the final row twice. That is not a level, it is a rate of change. Erosion nearly doubled within eight months. Whatever figure is current when you read this is worse than the one printed here.

Ranking first is worth roughly half what it was worth eighteen months ago, and the trend has not flattened.

What endures

If the click itself is eroding, the assets worth building are those whose value doesn’t depend on one. That lands squarely on the work Section 01 identified as late-paying and first to be defunded.

A symmetry here deserves explicit statement, because it represents the argument this entire guide has been building toward.

Section 01’s case for demand creation rested on the observation that most of your market is not buying today, and that recall at the moment of need is built long before that moment arrives. That was always true. It was also always deferrable, because search allowed you to buy your way into the moment of need after the fact, expensively but reliably.

That option is being repriced. Zero-click data transforms Section 01’s long-term argument into an immediate one. Mental availability was the patient case; it is now the pragmatic one.

What would still work

The engine you’re building is not a collection of channels. It is the relationship between what you stand for, who can recognize it, and how long you’re willing to wait to determine whether it worked. Most companies get the first right, guess at the second, and never reach the third.

— SOURCES

Everything cited, with its tier

Grouped by what kind of evidence it is. Where a figure could not be traced to a primary source, it appears in section 00 rather than here.

PEER-REVIEWED RESEARCH  ·  TIER 1

• Blake, T., Nosko, C. & Tadelis, S. (2015). Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment. Econometrica 83, 155–174.
• Gordon, B., Zettelmeyer, F., Bhargava, N. & Chapsky, D. (2019). A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook. Marketing Science 38(2), 193–225.
• Thomas, M., Goic, M. & Kalyanam, K. (2024). Long Lags and Large Returns: Experimental Evidence from Advertising to Businesses. Management Science. DOI 10.1287/mnsc.2023.02661.
• Li, H. & Kannan, P.K. (2014). Attributing Conversions in a Multichannel Online Marketing Environment. Journal of Marketing Research 51(1), 40–56.
• Berman, R. (2018). Beyond the Last Touch: Attribution in Online Advertising. Marketing Science 37(5), 771–792.
• Goodhardt, G., Ehrenberg, A. & Chatfield, C. (1984). The Dirichlet: A Comprehensive Model of Buying Behaviour. Journal of the Royal Statistical Society.
• Romaniuk, J. & Sharp, B. (2004). Conceptualizing and Measuring Brand Salience. Marketing Theory.

BOOKS  ·  TIER 1

• Binet, L. & Field, P. (2013). The Long and the Short of It. IPA.
• Binet, L. & Field, P. (2017). Media in Focus. IPA.
• Field, P. (2018). Effectiveness in Context. IPA.
• Sharp, B. (2010). How Brands Grow. Oxford University Press. Part 2 (2016).
• Romaniuk, J. (2018). Building Distinctive Brand Assets.

SURVEYS WITH DISCLOSED METHODOLOGY

• The CMO Survey (Duke Fuqua, Deloitte, American Marketing Association), 35th edition, January 2026. n = 308. [1]
• Gartner CMO Spend Survey, 2024–2026. n ≈ 400 per year. [2]
• Gartner B2B buying surveys, May 2025 (n = 632) and March 2026 (n = 646). [1]
• Pew Research Center, AI Overviews and search behaviour, 2025. n = 900. [1]
• The Bridge Group, 2025 SDR Models, Metrics & Motions. n = 351. [1]
• Pavilion / Benchmarkit, 2025 B2B SaaS Performance Metrics. n = 583. [2]
• High Alpha, 2025 SaaS Benchmarks Report. n = 800+. [2]
• KeyBanc Capital Markets / Sapphire Ventures, 15th Annual SaaS Survey. n ≈ 104. [2]
• WordStream / LocaliQ, 2026 Google Ads Benchmarks. n = 13,474 campaigns. [2]
• Optifai sales cycle and win rate study. n = 939. [2]
• ICONIQ, 2026 State of Go-to-Market. n = 150+. [2]
• 6sense, 2023 B2B Buyer Experience Report. n = 900+. [2]
• Ahrefs, AI Overviews and click-through, December 2025. 300,000 keywords. [2]
• SparkToro with Similarweb, zero-click search, 2026. [2]
• SparkToro with Really Good Data, dark social referrer study. 1,113 visits. [1]-adj
• First Page Sage, 2026 Sales Funnel Conversion Benchmarks. No sample disclosed. [2]

COMMISSIONED RESEARCH

• Binet, L. & Field, P. The 5 Principles of Growth in B2B Marketing. Commissioned by the LinkedIn B2B Institute. Fewer than 50 B2B cases; the authors describe the findings as tentative. [2]
• Dawes, J. The 95-5 rule. Ehrenberg-Bass Institute, published via the LinkedIn B2B Institute. Described by its author as a heuristic rather than a measurement. [2]