Attribution across forty locations, where the averages lie to you
A multi-location business that reports one national attribution number is hiding its most important finding. The same campaign performs differently in every market, and the aggregate is an average of things that should never have been averaged.
There is a specific report that most multi-location businesses produce, and it is the wrong one. It shows marketing spend, leads and revenue at national level, broken down by channel, with a return figure at the bottom. Everybody nods. Nobody can act on it, because nothing in that business happens nationally — it happens at forty sites with forty managers, forty catchments and forty answering habits.
The national attribution number for a multi-site business is an average of things that should never have been averaged. It is arithmetically valid and operationally empty.
The useful finding in a multi-location business is almost never which channel works. It is that the same channel works three times better at one site than another, and that the difference has nothing to do with the marketing.
Why the aggregate hides the finding
A national figure blends locations whose conversion economics differ by large multiples. Averaging them produces a number that describes no actual site, and it conceals the variance that contains all the value.
Consider a chain where paid search returns 6x nationally. That single figure is entirely consistent with eight sites returning 12x, twenty returning 6x, and twelve returning under 2x. The national number is correct and the twelve sites destroying value are invisible inside it. Nobody at the centre has any reason to look, because the headline is healthy.
The variance is not noise. In practice the spread between the best and worst performing locations on the same campaign is routinely three to five times, and it is stable enough over time to be a property of the site rather than a statistical accident.
Which means the single highest-return analysis available to a multi-location business is not channel optimisation at all. It is finding out why the bottom quartile converts the way it does, because closing half of that gap is usually worth more than any plausible improvement in media efficiency.
The location field is the whole trick
Attributing by location requires one additional column on the sales export: which site closed the sale. Everything else is the same matching process, sliced differently.
This is much easier than teams expect, and the expectation is usually the obstacle. People assume location-level attribution requires location-level campaigns, geo-targeted tracking, or separate accounts per site. It does not. The campaigns can stay national. The revenue is already tagged with a site in the sales system because the business needs that for every other purpose.
Where it does get harder is when the sales system records the salesperson but not the site, or records a region rather than a branch. Both are usually solvable with a lookup table maintained once. It is worth insisting on the granularity, because regional roll-ups reintroduce exactly the averaging problem you are trying to escape.
The one genuine complication is a sale that spans sites — an enquiry taken centrally and fulfilled locally, or a customer served by two branches. Pick a convention, write it down, and apply it consistently. Any consistent convention beats an argument each quarter about the right one.
The routing problem, which usually comes first
In many multi-site businesses the biggest measurable loss is not media waste but enquiries reaching the wrong location, or no location at all. Attribution by site surfaces this immediately, because the pattern is unmistakable.
The signature is a site with a high volume of enquiries and a poor close rate sitting next to one with fewer enquiries and a much better rate. Occasionally that is a capability difference. Frequently it is that a central number routes by postcode rules written four years ago, and a chunk of a busy catchment is being sent ninety minutes away.
Call routing rules, opening hours, overflow behaviour and voicemail handling are all invisible in a marketing report and enormously consequential in a location report. An enquiry that rings out at six in the evening is a lost sale that appears in your data as a lead the marketing generated and the site failed to convert.
The reason this matters for measurement rather than just operations is that it corrupts every conclusion drawn about channels. A channel that generates evening enquiries will look worse than one generating morning enquiries, at every site with the same voicemail policy, and no amount of campaign optimisation will fix a telephone.
Small samples and the temptation to over-read
A single location produces far fewer sales than the group, so its attribution figures move violently month to month. Treating that movement as signal produces constant, pointless intervention.
| Site | Matched sales | Revenue | One extra large deal | Apparent change |
|---|---|---|---|---|
| Group total | 1,840 | 18,400,000 | 18,460,000 | +0.3% |
| Large site | 210 | 2,300,000 | 2,360,000 | +2.6% |
| Medium site | 64 | 640,000 | 700,000 | +9.4% |
| Small site | 22 | 185,000 | 245,000 | +32.4% |
One deal moves the smallest site by a third. If site managers are held to monthly attributed revenue, the small sites will appear to be wildly improving and collapsing in alternation, and a great deal of management attention will be spent explaining variance that is entirely arithmetic.
The remedies are unexciting and effective: report small sites on rolling twelve-week windows rather than months, show the sale count beside every revenue figure, and set a threshold below which a site's figures are explicitly marked as directional. Making the uncertainty visible is far better than smoothing it away.
Comparing sites fairly
Ranking locations by attributed revenue rewards catchment size rather than performance. Compare on conversion of matched enquiries and on revenue per enquiry, which are within a site's control, rather than on totals which are not.
This is a fairness point and also an accuracy one. A city-centre branch with three times the population in its catchment should produce more revenue, and a league table that ranks it first has measured geography. The manager of the small rural site cannot do anything about that and will reasonably disengage from a scoreboard that only measures where they happen to be.
The metrics that survive this objection are close rate on matched enquiries, average value, speed to first contact, and the share of enquiries that were never contacted at all. Every one of those is actionable at site level, and the last is frequently the most shocking number in the whole exercise.
It is worth publishing the comparison openly within the group. Sites benchmark themselves against each other far more energetically than against a target, and a well-constructed comparison generates more improvement than any centrally mandated programme.
What head office should actually look at
The centre should look at the distribution rather than the average: which sites are in the bottom quartile, whether they are consistently there, and what is different about them. The national figure belongs in the budget conversation and nowhere else.
A useful centre report has four elements. The group figure for budget-setting. The distribution across sites, so the spread is visible. A watch list of sites persistently below the median. And the two or three operational variables that correlate with position in that distribution, which in most businesses are response time, contact rate and opening hours.
What the centre should avoid is issuing site-level targets derived from the national average. A target that a third of sites cannot reach because of catchment structure will be ignored by exactly the sites that most need attention, and it discredits the measurement along with the target.
The other thing to avoid is monthly reforecasting of channel mix based on site-level noise. Media allocation should move on group-level evidence over meaningful windows; site-level data is for operational intervention, which is a different lever with a much faster response.
Local budgets and the co-operative fund question
Where sites contribute to a shared marketing fund, attribution by location becomes a governance question as well as an analytical one, because the fund's fairness can now be measured rather than asserted.
This is uncomfortable and useful in equal measure. A national campaign funded proportionally by all sites but generating enquiries concentrated in a few catchments has always been an implicit transfer between locations. Nobody could see it before, so nobody argued about it. Location-level attribution makes it explicit.
The right response is usually not to abandon shared funding, which exists for good reasons of scale and brand consistency, but to be honest about the distribution and to use it to inform where incremental local spend goes. A site consistently under-served by national activity is a candidate for a local budget rather than a complaint.
It does require the centre to be willing to publish numbers that will be scrutinised by the people paying into the fund. Businesses that are not prepared for that conversation should think carefully before producing the analysis, because it does not go back in the box.
CloseRev reports by any dimension present on your sales file, so location-level breakdowns need one extra column rather than a separate deployment per site. The unmatched bucket is shown per location too, which is usually where a routing problem first becomes visible.
The multi-brand and multi-format complication
Groups that operate several brands or formats need attribution split by brand before it is split by site, because a weak brand with strong sites and a strong brand with weak sites produce identical group averages and require opposite interventions.
This is a common structure in mid-market groups that have grown by acquisition, where three or four fascias share a back office and a marketing team but not a customer base. Reporting them together produces numbers that describe no market anybody actually competes in, and the temptation to do so is strong because the sales system finally unified them.
The practical approach is to treat brand as the outer dimension and site as the inner one, and to resist any roll-up that crosses brands except for budget-setting. Comparing a premium fascia's close rate against a value fascia's is not a performance comparison; it is a description of two different businesses.
Where the brands genuinely share demand — the same customer considers both — that overlap is itself worth measuring, and it will show up as enquiries matched to one brand closing at another. Systematic leakage in one direction is a positioning finding, and it is invisible at group level.
What to give the site manager
A site manager needs their own enquiries, their own close rate, their own uncontacted count and a comparison against the group median. They do not need channel-level media analysis, because they cannot act on it.
The single most valuable item on that list is the uncontacted count, and it is the one most often omitted. Telling a manager that forty-one enquiries last month were never contacted at all is concrete, actionable and immediately motivating in a way that a return-on-spend figure is not. It also tends to be the finding that produces the fastest revenue improvement anywhere in this exercise.
Keep the report short and stable. A one-page view that arrives on the same day every month and never changes format will be read; a rich dashboard that requires exploration will be opened twice and then ignored. This is a well-understood pattern and it is routinely designed against.
Finally, give the numbers to the manager before they appear in any comparison seen by their regional director. Being surprised by your own figures in a meeting is the fastest way to make somebody an opponent of the measurement rather than a user of it.
The mistake that wastes a year
The commonest failure in multi-location attribution is trying to build per-site tracking infrastructure — separate accounts, separate numbers, separate tags — before doing the simple group-level match with a site column on it.
It is an understandable instinct, because splitting the marketing feels like the prerequisite for splitting the reporting. It is not, and the projects that start there tend to spend six to twelve months on plumbing and arrive with something fragile that breaks whenever a site is opened, closed or rebranded.
The two-export approach gets you a location breakdown in a week, on data you already hold, with no changes to any campaign. If that breakdown shows the variance described in this article — and it will — you then have an evidence-based case for whatever infrastructure genuinely turns out to be needed, which is usually far less than was originally proposed.
That sequencing matters commercially as well as technically. A measurement programme that produces a finding in its first month keeps its budget. One that asks for a year of build before showing anything is competing for patience it has not earned.
The sequence that works
Start with the group match, add the location dimension, look at the distribution before the averages, and investigate the bottom quartile operationally before touching media. Most of the available value is found before any campaign is changed.
- Match sales to sources at group level first, so the data quality issues are found once rather than forty times.
- Add the site column and produce the distribution, with sale counts beside every revenue figure.
- Mark sites below the sample threshold as directional and report them on a longer window.
- Investigate the bottom quartile on operational variables — response time, contact rate, hours — before concluding anything about channels.
- Only then compare channel performance across sites, and expect the differences to be about local conditions more often than about media.
In a multi-site business, the marketing report that changes the most money is usually the one that turns out to be about telephones.
Questions people actually ask
- How do you attribute revenue by location when campaigns run nationally?
- Match sales to source records first, then assign each matched sale to the location that closed it, using the location field already on the sales record. The campaign does not need to be split by site; the revenue does.
- Why do the same campaigns perform differently at different locations?
- Because the location is part of the conversion mechanism. Local competition, how quickly the phone is answered, staffing, catchment demographics and the site manager's follow-up habits all sit between the enquiry and the sale, and they vary far more than the marketing does.
- How many sales does a location need before its attribution numbers mean anything?
- Enough that a handful of deals cannot swing the result. As a rough working rule, treat a site with fewer than about thirty matched sales in a period as directional only, and compare it over a longer window rather than month to month.
- Should each location get its own attribution report?
- Each location should get its own view of its own numbers, and head office should get the comparison. Sending every site the full national report produces either indifference or an argument about somebody else's figures.
- How do you handle a lead that arrives at one location and closes at another?
- Attribute the revenue to the location that closed it and record where the enquiry landed, so the transfer is visible. Systematic transfers in one direction usually indicate a routing problem worth more than any campaign optimisation.
- Is a national attribution number useful at all for a multi-site business?
- It is useful for budget-setting at the centre and misleading for anything operational, because it averages markets with genuinely different economics. Report it, but never let it be the only view anybody sees.