First touch versus last touch on a show that sells for nine months

The audience director wants next year's media plan approved in three weeks and asks the data team for the channel numbers. The data team asks which channel numbers, and the room finds this pedantic, right up to the point where two reports arrive and the top of the ranking is different in each.

That is first touch versus last touch, and on a show that opens registration nine months before the doors and takes 40 per cent of its file in the final fortnight, the two reports are further apart than they are for almost any other kind of business.

Two questions, not two versions of one question

First touch answers a question about discovery. Of the people who registered, which channel introduced them to this show?

Last touch answers a question about conversion. Of the people who registered, which channel was in front of them when they finally did it?

Both are answerable from the same file. Both are correct. They diverge because a nine month buying cycle has room for a channel that finds people in January and a different channel that pushes them over the line in September, and no single touch report can hold both facts at once.

CEIR presented preliminary findings from its Attendee Acquisition Study on 31 January 2023, reported by Trade Show Executive the following day, and two figures from it are worth carrying into this argument. 98 per cent of respondents used one or more digital tactics, and email was described as one of the most effective tactics for generating registration. A show whose registrations are produced by many overlapping tactics is exactly the case where a single touch report, whichever end you pick, throws away most of what happened.

What happens when you run both over one file?

Take a completed edition with 12,000 registrations, all of them carrying at least one recorded touch.

Under first touch, paid social holds 3,360 registrations, which is 28 per cent. Organic search holds 2,640, or 22 per cent. Email holds 1,800, or 15 per cent. Paid search holds 1,440, direct holds 1,440 and partner referrals hold 1,320, which are 12, 12 and 11 per cent.

Under last touch, on the same 12,000 rows, email holds 4,080, or 34 per cent. Direct holds 2,760, or 23 per cent. Paid search holds 1,680 at 14 per cent, paid social holds 1,320 at 11 per cent, organic search holds 1,200 at 10 per cent and partner referrals hold 960 at 8 per cent.

Paid social goes from first place to fourth. Email goes from third to first, more than doubling from 1,800 to 4,080. Nothing about the media buy changed between the two reports. The same 12,000 people did the same things.

The direction of the flip is predictable enough to state as a rule. Channels that reach strangers collect first touches, because reaching a stranger is what they do. Channels that require an existing relationship collect last touches, because they can only speak to people who already gave you an address or a follow. Email cannot introduce anybody to your show, since you needed their address to send it, and it is the channel most often in front of them at the deadline.

That structural point has a sharp consequence. Any show that reports last touch alone will conclude, every single year, that its owned channels are producing the audience and its paid channels are underperforming. It will cut prospecting, the file will hold up for one edition on the strength of the existing database, and the first time share will fall the year after.

Which report goes to which meeting?

Here is the rule I would use, and it is simple enough to survive contact with a busy commercial team.

The first touch report goes to the audience planning meeting, the one held four to six months out that decides where the money goes. It answers the question that meeting is actually asking, which is where new people come from. Judge prospecting media, partnerships and content on it.

The last touch report goes to the campaign meeting in the final eight weeks, where the questions are about creative, offer and send schedule. It answers which message closed people against the deadline. Judge deadline emails, retargeting and price break communications on it.

Neither report goes to the board alone. What goes to the board is the registration count, the coverage rate behind it, and one line saying which channels found the audience and which converted it.

The failure mode I would guard against hardest is a single channel being judged on the wrong report. A prospecting campaign assessed on last touch will look like waste in every edition it ever runs, because prospecting rarely closes anybody. Berman's analysis in Marketing Science in 2018 makes the mechanism precise: the popular last touch method overincentivises ad exposures and lowers advertiser profits, while a Shapley value allocation reaches greater efficiency. The distortion is a property of the credit rule rather than a property of the campaign.

What each report gets wrong

First touch is generous to whatever your tracking happened to see first, which is not the same as what happened first.

A person who saw a trade press advertisement in March, heard about the show from a colleague in May and clicked a retargeting ad in June will have a first touch of paid social. The two touches that mattered are invisible, and no amount of window widening finds them. First touch also rewards cheap wide reach, since a low quality impression that produces one click on an unqualified visitor books the same first touch as an expensive, well targeted one.

Last touch is generous to the deadline. Anything running in the final fortnight collects credit for demand that nine months of work created, and the closer to the cut-off a channel operates, the better it looks. The extreme version of this is a discount code emailed 48 hours before close, which will out-perform every other line in the report and may have sold nothing at all that was not already sold.

Both reports share a blind spot: they say nothing about incremental effect. A registration credited to any channel might have happened anyway. That question needs a test rather than a report, and it belongs to the incrementality work rather than here.

What about registrations that only ever had one touch?

They are the quiet problem underneath both reports, and the size of the group decides how much either ranking is worth.

On the 12,000 row file above, suppose 3,900 registrations carry exactly one recorded touch. For those rows the first touch and the last touch are the same event, so they contribute identically to both rankings and they contribute nothing at all to the comparison between them. The real disagreement between the two reports is generated by the 8,100 rows with two or more touches, and any percentage point difference you are reading is a difference measured on two thirds of the file.

Single touch rows are also not a random third. They are the people who arrived through a forwarded link, an exhibitor guest code, a printed QR code at another event, or a device your tracking had never seen before. The channel labels on those rows lean towards direct and referral, which is why both reports credit those two more generously than they deserve.

The fix is to publish the count, every time. Twelve thousand registrations, 8,100 with multiple touches, 3,900 with one. Then the person reading the ranking knows how much journey there was to divide.

The version I would actually publish

One page, two rankings, side by side, on the same file with the same cut-off date and the same coverage figure printed underneath.

Then one derived column that earns its place: the ratio of last touch registrations to first touch registrations per channel. Email at 4,080 over 1,800 is 2.27. Paid social at 1,320 over 3,360 is 0.39. Paid search at 1,680 over 1,440 is 1.17. Organic search at 1,200 over 2,640 is 0.45.

Anything well above 1 is a closing channel. Anything well below 1 is a finding channel. Anything near 1, like paid search here, is doing both, which is usually a sign that the brand terms and the category terms inside it are behaving as two different channels and should be reported as two.

That ratio is the most useful single number I know of for a nine month sales cycle, and it costs one division per channel. It also survives the objection that neither model is true, because it uses the disagreement between the two models as the finding.

The limit

Both reports assume the file knows about the touch. On most registration files a large minority of rows carry one touch or none, and for those rows first touch and last touch are the same event, which drags both rankings towards whatever channel dominates the tail end of the journey.

Both also assume the journey belongs to one person. Shared inboxes, assistants registering delegates and colleagues forwarding a link all break that assumption, and they break it more often at the senior end of the audience.

Beyond that, the pair of reports is a description of what preceded a registration and never a claim about what caused it. If the question in the room is how to split credit across the middle of a long journey, that is a weighted split over the whole touch sequence rather than a choice between the two ends. If the question is how much weight the final fortnight deserves, an explicit decay parameter answers it better than either single touch view. And if the two reports disagree because touches are dropping out of the lookback, the window length is the thing to fix before anybody argues about models. The pillar page on acquisition and attribution holds the rest of that ground.

Run both reports on your last completed edition this week, put them on one page, and compute the last touch to first touch ratio per channel. Then look at the channel with the lowest ratio and check whether anybody has been judging it on the last touch number.