A purchase taking three quarters and six signatures defeats the arithmetic most marketing reporting rests on. What to measure instead, and where attribution genuinely stops working.
Attribution assigns credit to recorded touchpoints inside a window. In industrial markets the window is routinely shorter than the cycle, and everything downstream of that mismatch is unreliable in ways that are not visible on the dashboard.
Three things go wrong. The record decays, because tracking identifiers expire, individuals change roles and the buying group is larger than any single contact record. The decisive influence often happens where nothing is logged, in a distributor conversation, at a trade fair, or inside a plant standard written years earlier. And revenue booked this quarter reflects decisions taken three or four quarters ago, so optimising against it means steering with a stale signal.
The failure compounds because the cheapest events to capture — impressions, sessions, form fills relabelled as marketing qualified leads — are also the ones least connected to the outcome. They dominate reporting by availability rather than by usefulness, and once they are in a board pack they are difficult to remove.
The replacement is not a better attribution model. It is a smaller number of leading indicators measured consistently, a separate and slower cadence for revenue questions, and an explicit statement of what the numbers cannot tell you.
The indicators worth tracking share one property: they cost the buyer effort, and that effort only makes sense if a project exists. Effort is the filter. A page view costs nothing; requesting a sample costs an internal approval.
The set below is a menu rather than a checklist. Most manufacturers can instrument four or five of these well, and four measured properly beat twelve measured loosely.
Each of those indicators is a proxy for a stage in someone else's internal process, and each is corruptible. CAD libraries get scraped, students download datasheets, competitors pull certificates. Totals are therefore close to meaningless.
Count by named account instead. Three different documents pulled by three different people at one site inside a fortnight is a live project and should reach a sales engineer that week. Thirty downloads spread across thirty unrelated domains is probably noise, and treating it as a thirty-fold better result is how measurement starts misleading the people it serves.
This changes what the reporting system has to do. It needs company-level resolution, a way to associate anonymous activity with a known account, and tolerance for the fact that a meaningful share of activity will never resolve. Partial resolution is normal and is not a reason to abandon the approach.
Where one specification win can outweigh a year of small orders, the number of enquiries tells you almost nothing. Volume targets in that setting actively distort behaviour, because the cheapest way to raise the count is to lower the threshold.
Score at intake on a few observable fields rather than on a subjective rating. Whether an application was described. Whether a volume or a project timeline was given. Whether the enquirer works at a company in a served sector. Whether the part or specification named is one the firm actually makes. Four or five fields is the practical ceiling, because sales will not complete more than that reliably.
Then report the distribution rather than the average. The question that matters is whether the top band is growing in absolute terms, not whether a composite score drifted upward. An average can improve while the number of serious projects falls.
Track decline reasons with the same discipline. Enquiries the firm refuses are direct evidence about where the positioning is reaching the wrong audience, and they are usually the fastest route to fixing targeting.
Over a long cycle, most of the measurement problem sits at the seam. An enquiry that is passed on informally, worked for a quarter and then abandoned leaves no record that anything was ever wrong, which is why the handover between marketing and sales deserves more instrumentation than the channels feeding it.
Record three things at the moment of transfer: when it happened, what state the enquiry was in — what was asked for, in the enquirer's own words — and whether it was accepted or returned, with a reason. Two rates then become available: time to first contact, and return rate by source. If enquiries from one source are returned consistently, that is a targeting or positioning fault rather than a sales performance fault, and you will only ever see it if returns are recorded.
Insist on a closed loop on outcome, even a coarse one. In many industrial firms the real outcome is not a CRM stage at all: it is a part number appearing on a drawing or in an approved vendor list. Ask sales engineers to record separately from orders. The specification happens quarters earlier, it is the thing marketing can genuinely influence, and it is the earliest honest evidence that the work is compounding.
An executive report answers three questions: is qualified demand growing, where is it coming from, and what changed since the last report. Anything that does not serve one of those is decoration.
Include named-account activity over time, high-quality enquiry counts with the qualifying definition printed on the same page, specification and sample events, pipeline value created in the period as distinct from pipeline closed, and a short note on what was tried and what it cost. Include an uncertainty line — what the report cannot yet show.
Leave out impressions, reach, follower counts, aggregate MQL totals, single-touch attribution splits presented as fact, and any chart whose axis or definition has changed since the previous edition. A silently rebased chart destroys more credibility than a bad quarter.
One discipline is worth enforcing above the others: state the definition of every metric on the report itself. Over a multi-year cycle, the report will outlive the person who designed it, and an undefined metric drifts until it means whatever the current reader assumes.
The common starting position is a half-populated CRM, web analytics that have been reconfigured twice, and a sales history distributed across individual inboxes. The temptation is to reconstruct the past. Resist it, because an estimated baseline becomes an unfalsifiable comparison that every later result is judged against.
Set the measurement baseline forward instead. Pick a start date, fix the definitions in writing, and freeze them for at least two quarters. Frozen definitions are worth more than accurate ones at this stage; you can refine a definition later, but you cannot repair a period during which it kept changing.
Do one cheap retrospective exercise in parallel. Reconstruct the last twenty to thirty won opportunities with source, first contact date and close date. That yields a cycle-length distribution and a rough sense of where wins originate. It is anecdotal, and should be labelled as such, but it tells you how long the reporting window has to be before anything can reasonably be judged.
Say plainly in the first report that the opening two quarters are calibration. Setting that expectation early is often what keeps a long-cycle programme alive past its first review.
Attribution over a multi-quarter, multi-signature purchase is not a solvable problem. Spending heavily on instrumentation in pursuit of a certainty that the buying process cannot produce is a common and expensive mistake, and it usually ends with an elaborate system nobody trusts.
What is achievable is narrower and still useful: correlation at the account level, direction of travel over time, and the ability to say which activities preceded the projects that closed. Two practices get reasonably close without pretending. Ask at intake and again at close how the buyer first came across the company — imprecise, self-reported, and still better than nothing when read as a trend. And run occasional holdouts, pausing an activity in one region or segment and watching the leading indicators rather than the revenue, since revenue will move too late to inform the decision.
Be direct with the executive about where the line falls. You can show whether qualified demand is rising, and you can attribute at account level with moderate confidence. You cannot assign a defensible fraction of a closed order to a single campaign. Saying that at the outset costs a difficult conversation once; discovering it in the third quarterly review costs the programme.
Key takeaways
Long enough to cover at least one full sales cycle, which usually means reconstructing the cycle length from recent won deals first. Judge leading indicators from the first quarter and revenue only after the cycle has run once. Setting that window in advance prevents the programme being cancelled on a signal that could not have arrived yet.
As a total they are close to useless, because the number rises whenever the qualifying threshold falls. If the term is already embedded, redefine it against observable intake fields and report the top band as an absolute count. Otherwise replace it with specification, sample and account activity measures.
Named-account activity trend, high-quality enquiry counts with the definition printed alongside, specification and sample events, pipeline value created in the period, and a note on what was tried and what it cost. Add a line on what the data cannot yet show. Leave out impressions, reach and single-touch attribution splits.
Track distributor-originated enquiries by partner rather than pooling them, and agree a minimum record at handover so the origin is not lost. Ask distributors what they are being asked for, since that is a leading indicator you cannot instrument directly. Expect partial visibility and report it as partial.
