00The problem, stated precisely

Event ROI has a numerator problem, not a denominator problem. Cost is knowable to the cent — our own dataset of 1,916 priced event budgets puts the median corporate event at €102 per guest. The return is where measurement collapses: it is multi-dimensional (pipeline, relationships, brand, retention), lagged (weeks to years), and partially latent (memory structures in buyers who are not in-market today). The academic literature settled one thing three decades ago: returns from business events can be measured and quantified under controlled conditions (Gopalakrishna, Lilien, Williams & Sequeira, “Do Trade Shows Pay Off?”, Journal of Marketing, 1995) — but only if you decompose the return, choose the right level of causal evidence for each component, and instrument the measurement before the event.

That is what this paper does. The discipline is borrowed from investment management: no fund reports a single blended number without component attribution and stated assumptions, and an event programme deserves the same standard. No composite “event ROI %”. A scorecard of measured components, each with a formula, an instrument, and a stated limitation.

R(event) = R_pipeline + R_relationship + R_brand + R_retention
           [ + R_internal — deliberately unpriced ]
ROI is reported as a component scorecard, not a single quotient.

01Attention: the input currency, priced

Attention research has upended media measurement: on average, roughly half of digital ad “time-on-screen” receives no human attention at all, and 85% of digital placements earn under 2.5 seconds of active attention (Nelson-Field / Amplified Intelligence; Lumen–Brand Metrics meta-analysis). The same literature shows aggregate attention time — not impressions — is the metric that predicts brand outcomes. An event is, in attention terms, an extreme asset: hours of voluntary, undivided, senior attention. Price it the way media buyers now price attention:

EAT = Σᵢ tᵢ × qᵢ   — Earned Attention Time (attentive-hours)
    tᵢ = attentive hours of guest i
    qᵢ = attention quality (1.0 undivided · 0.5 divided)
CPAH = C_total / EAT   — Cost per attentive hour
Digital: CPAH = CPM / (1000 × s̄ / 3600)
    s̄ = mean attentive seconds per impression

Worked comparison with stated inputs: a €15,300 summit (150 guests × €102) delivering ~3 undivided hours per guest → EAT ≈ 450 attentive-hours → CPAH ≈ €34 per attentive hour. A LinkedIn campaign at a €20 CPM with 1.5 attentive seconds per impression prices at CPAH ≈ €48 per attentive hour — of fragmented, solitary, second-long exposures. The event buys cheaper attention, and categorically different attention: continuous, social, multi-sensory, senior. This is the correct first answer to “why events” — and unlike a revenue projection, every input here is either your invoice or a published attention benchmark.

02Relationship capital: measuring intimacy

Your instinct that a 30-guest VIP experience creates something a 300-guest conference cannot is correct, and it has a measurement. Relationship-marketing theory treats trust and commitment as the mediators of B2B exchange (Morgan & Hunt, 1994); interaction depth is how events build them. Define a depth scale and score every contact:

d: stage = 1 · group discussion = 2 · hosted table = 4 · extended 1:1 = 8
RC = Σₐ wₛ(a) × d(a)   — Relationship Capital (contact-level)
    wₛ = seniority (IC=1 · manager=2 · director=3 · C-level=5)
RDI = RC / N   — Relationship Depth Index (intimacy per guest)

Reach–depth frontier chart: event formats positioned by guest count and interaction depth, with iso-relationship-capital curves — Figure 1 — Formats positioned by reach (median guest count, from 1,861 costed events) and interaction depth (defined scale). Dashed curves are iso-relationship-capital lines, RC = N × d̄.

Our cost data sharpens the frontier: per-guest prices barely separate the two strategies — executive dinners median €112/guest, conferences €102 — but totals diverge with headcount, and reach gets structurally cheaper at scale (per-guest cost falls from €114 at 5–30 guests to €58 at 400+, as the venue amortises from 50% of budget to a third). Depth costs linearly; reach compounds. The frontier makes format choice a measurement decision instead of a taste decision: a 300-guest summit and a 12-seat private dinner can produce similar total RC through opposite routes — reach versus depth. What you cannot do is get both from one event, which is why serious activation programmes pair a reach event with a depth event in the same week. Post-event, RC is validated behaviourally, not by feel:

Δresponse = r_attended − r_matched_non-attended
    r = reply rate to identical outreach, 30 days post-event

That reply-rate delta is a quasi-experiment you run for free in your own CRM, and it is the cleanest early signal that the intimacy was real.

03Pipeline: the attribution ladder

Attribution evidence ladder: from self-reported CRM influence up to randomised holdout with difference-in-differences — Figure 2 — Four levels of causal evidence for event → pipeline claims. Most teams sit at L4; L2 is achievable for any company with a defined target-account list.

“Event-influenced pipeline” (L4) is the number most CRMs print and the one a CFO should trust least — it credits the event with every deal it touched. The incrementality literature is unambiguous: causal lift requires a counterfactual. For B2B events the practical designs are:

L1: from matched target accounts, invite subset T; hold out subset C
Estimator (difference-in-differences):
    τ = (Y_T,post − Y_T,pre) − (Y_C,post − Y_C,pre)
    Y = pipeline € | opportunities | engagement score
    windows: 90 days pre / 90 days post
Guard-rails: match on firmographics + pre-engagement;
check parallel pre-trends; report τ with CI, not a point

Two honesty notes. First, holdouts cost something real: you are deliberately not inviting accounts you want at the event — which is why L2 (matched controls without randomisation) is the workable standard for most companies. Second, sample sizes in B2B are small; a 40-account event will produce a directional τ, not statistical significance, and should be reported as such. Complementary velocity metrics that need no control group: Δ days-in-stage and meeting-acceptance-rate delta for attended accounts versus their own trailing baseline.

04Brand: measuring memory, not intent

The Ehrenberg-Bass 95:5 rule reframes what most of your event audience is: roughly 95% of B2B buyers are out-of-market at any moment (Dawes / LinkedIn B2B Institute). For them the event cannot create pipeline — it creates mental availability: the probability your brand surfaces when a buying situation eventually arises. Buyers overwhelmingly purchase from the shortlist they hold on day one, so the measurable question is whether the event moved you onto more day-one lists. Three instruments, in ascending cost:

(a) Share of Search — free, weekly, geo-filterable (Google Trends):
    SoS_geo = searches(brand) / Σ searches(category brand set)
    filtered to activation country; 8wk pre vs 8wk post → ΔSoS_geo
(b) Owned audience: Δfollowers (target geo) · ΔSOV in category
    conversation (Brandwatch / Meltwater)
(c) Micro brand-lift survey — target-account contacts, 4 questions,
    fielded T−2wk and T+4wk: unaided recall · aided recall ·
    day-one-list inclusion · category-entry-point association

Share of Search is the strongest cheap instrument on this list: Binet’s and Hankins’ work shows it correlates strongly with market share and leads it by months, with Hankins finding it accounts for ~83% of market share across 30 case studies. It is not a perfect predictor — price and distribution intervene — but as a pre/post delta in a specific country around a specific activation, it is exactly fit for purpose, and it costs nothing.

05Retention: the quiet fourth engine

For events attended by existing customers, the measurable outcome is retention economics, using the same matched-control logic as Section 3:

ΔNRR = NRR_attended − NRR_matched_control   (12-month window)
Δchurn = churn_attended − churn_matched_control
Also report: expansion-revenue delta for attended accounts

This is often the largest true return of flagship customer events and the least measured — it requires only your own revenue data and a matched comparison set, and no one builds it because the event team and the CS team report to different people.

06The protocol: instrument before, measure after

The T−8 week start is not arbitrary: in our booking data the median event is locked just 31 days out, yet events planned 2–3 months ahead run ~6x larger — the measurement window and the quality window are the same window.

T−8 weeks: lock the target-account list; split T/C if running a holdout; baseline SoS_geo, social, engagement scores, pipeline per account; field the pre-survey (T−2wk).

T0 (event): capture the raw measurement material — attendance and dwell logs (badge/RFID where scale justifies), interaction records per contact (who sat where, who met whom — the seating chart is a data instrument), and content assets. An unmeasured event cannot be measured retroactively; this is why capture budget (absent from 97% of the 1,916 budgets we analysed) is a measurement cost, not a nice-to-have.

T+2 days: outreach begins (the Δresponse quasi-experiment starts here). T+4 weeks: post-survey; first SoS read. T+90: DiD estimate τ on pipeline; velocity deltas. T+180: rebooking/re-engagement rate (in our booking data, ~4 in 5 repeat decisions occur inside 6 months). T+365: ΔNRR for customer events.

Report all of it as a one-page scorecard: EAT and CPAH · RC and RDI · Δresponse · τ (with CI) · ΔSoS_geo · Δday-one-list · ΔNRR. Seven measured numbers beat one fabricated percentage.

07Cross-border adaptation: activating in a foreign market

For an international company activating in a market where it has thin presence — the typical brief we produce for — the framework tilts:

Budget the market before the metric. Activation economics vary ~3x by city in our dataset — from €185 median per guest in Cannes and €149 in Paris to €76 in Lisbon and €62 in Bangkok — which changes the reach–depth position the same budget can buy in each market.
Geo-filter everything. SoS, social SOV and press monitoring restricted to the activation country and language; a global brand-search series will not move on a single-city activation, but the Thai- or Portuguese-language series can.
Baseline is near zero — treat that as an asset. Low baselines make deltas detectable: going from ~0 to measurable branded search in a new market is a cleaner signal than a 2% move in your home market.
Weight the partner ecosystem, not just prospects. Market entry runs through local partners, regulators, press and talent. Score them in RC with their own seniority weights; count local-media citations as a separate line.
Expect longer lags. Cross-border B2B cycles stretch the T+90 window; pre-register T+180 as the primary pipeline read so nobody declares failure at day 30.
Control for the calendar. If the activation rides a major conference week, part of any lift is the week, not your event. A same-country non-event-week comparison, or a competitor-brand SoS control series, separates the two.

08Limits, stated plainly

Depth weights are defined, not discovered. The d-scale and seniority weights are declared conventions (as are Gopalakrishna & Lilien’s stage weights). Their value is consistency across your events, enabling comparison — not cosmic truth. State them once; never tune them post hoc to flatter a result.
Small samples are the B2B condition. Most single events cannot reach significance on pipeline lift. Aggregate across an event programme (n = contacts across 4–6 events) before drawing causal conclusions; report single-event reads as directional.
Share of Search is a proxy with known critics. Its correlation with market share is well-evidenced but methodologically contested at the margins; use it as a delta detector, not a valuation input.
Internal-event value stays unpriced here. Offsites and celebrations produce retention and culture effects that this framework deliberately does not monetise; forcing them into ROI arithmetic manufactures false precision.
Our own data is supply-side. The cost benchmarks (€102/guest median, capture-budget incidence, rebooking clustering) come from our 1,916-budget dataset; the causal machinery and attention/brand benchmarks come from the cited literature, and each client’s funnel constants must come from their own measured events.

09References

Gopalakrishna, S. & Lilien, G.L. (1995), “A Three-Stage Model of Industrial Trade Show Performance”, Marketing Science 14(1) · Gopalakrishna, Lilien, Williams & Sequeira (1995), “Do Trade Shows Pay Off?”, Journal of Marketing 59(3) · Dekimpe et al. (1997), “Generalizing About Trade Show Effectiveness”, Journal of Marketing 61(4) · Morgan, R. & Hunt, S. (1994), “The Commitment–Trust Theory of Relationship Marketing”, Journal of Marketing 58(3) · Binet, L. (2020), “Share of Search as a Predictive Measure”, IPA EffWorks · Hankins, J. (2021), share-of-search / market-share meta-analysis · Dawes, J. / LinkedIn B2B Institute, “The 95:5 Rule” · Nelson-Field, K. / Amplified Intelligence, attention-economy research incl. “Hacking the Attention Economy” (2025) · Lumen Research & Brand Metrics attention–outcome meta-analysis · CH3 group dataset: 1,916 priced event budgets, 7,046 line items, 51 cities, 2022–2026.

CH3 — event production agency for crypto, gaming and tech. 680+ events · 400+ clients · 40+ cities · 12 hubs.

Measuring What Events Actually Do

TL;DR