Effectiveness

If you are placing a bet on where your brand's growth will come from, penetration is the better bet. Growing brands usually win more buyers. Higher purchase frequency among the buyers you already have plays a part too, but it is usually the smaller change.

The evidence for this is strong. Research from the Ehrenberg-Bass Institute looked at more than a thousand rises and falls in brand share across dozens of UK packaged goods categories. When brands grew, their penetration usually changed much more than their loyalty, for big and small brands alike.

Part of the reason is the pattern known as Double Jeopardy, which Andrew Ehrenberg's research made famous. Brands in the same category tend to have similar levels of loyalty, and smaller brands have fewer buyers who also buy them a little less often. So the odds are against a plan that relies on loyalty well above the category norm. If the average buyer in your category buys their favourite brand 12 times a year, you might get yours to 14. Getting them to 24 is a long shot. Your category's loyalty figures will show you where the ceiling probably sits.

Campaigns point the same way. When Les Binet and Peter Field analysed the IPA Effectiveness Awards cases, campaigns with a penetration objective reported very large business effects more often than campaigns with a loyalty objective. Loyalty campaigns that did work mostly worked by bringing in new customers. Their report is Marketing in the Era of Accountability, sold by the IPA. Peter Field describes every brand as a leaky bucket. Customers keep leaving, mostly to competitors and sometimes for ordinary reasons like moving house, however well you treat them. So the bucket has to be refilled all the time.

The odds shift in some places. Some brands do hold more loyalty than their size would predict, and in the UK research these were mostly market leaders and supermarket own brands. The IPA cases also suggest a loyalty objective may be more worth trying for big brands, for cheaper brands, and in categories where people buy on emotion. The buying studies here come from packaged goods. The IPA cases found the same pattern in other categories, but if you sell something else, test it against your own data before you bet on it.

So what can you do with it? Set a penetration goal alongside any loyalty goal. Track penetration every year, because a market share figure on its own won't tell you whether you grew through more buyers or more frequent buying. Plan for reach, so your marketing gets to people who don't buy you yet as well as the ones who do. And if you are a big brand, a loyalty objective may be worth a test.

Charlie

When sales rise during a campaign, it is tempting to give the campaign the credit. But that is correlation: the two happened at the same time. Causation is how much of the rise the campaign actually produced. Always measure the second, and never report the first as if it were the second.

Lidl is a good example. Its sales in the UK grew by £2.56 billion between 2013 and 2017. In 2014 it went on national TV for the first time, and growth went from 9.5% a year before the ads to 15.5% after. That timing looks like proof, but on its own it is only correlation.

But Lidl's award paper doesn't stop at the timing. The team used a statistical model to split the growth by cause: the ads, the new stores, and everything else. The ads were given about a quarter, £627 million of the £2.56 billion.

By our sum, had Lidl kept growing at 9.5% a year, it would have added about £1.4 billion anyway. So about £1.1 billion sat above the old line. The ads were given £627 million in all, a bit over half of that. So even the growth above the old line wasn't all down to the ads.

One caution. The quarter comes from Lidl's own model, and the paper doesn't show how the model was built. Its headline figure, £2.7 billion over four years, is a different sum: all the sales the ads drove each year, not their share of the growth.

The case won Gold at the IPA Effectiveness Awards in 2018. The full paper has far more and is worth reading.

Sam

When you tell your boss a campaign worked, expect one question. Was it the campaign, or was it something else? A price cut, a promotion, a new shop stocking you, a good summer.

The best answer is a number. "We didn't change the price" asks the boss to trust you. "Our price was €3 before the campaign and €3 during it" lets them see for themselves.

Does it matter? In award papers it goes with winning. We compared 96 papers that won an effectiveness award with 87 that entered and weren't awarded. 78% of the awarded papers answered three or more other explanations with a number or a chart (75 of 96). 45% of the others did (39 of 87).

So before you present a result, do three things.

1. Write down what else could explain it.
2. Put a number beside each one.
3. If you have no number for one, say so.

Marie Curie's award paper shows the habit. It lists what else could explain its result and takes them one at a time. For the weather, it went to the Met Office's records.

Two cautions. Some answers we counted point to a model or a source and show no number. Leave those out and it is about two thirds against under four in ten. And this shows what awarded papers do more often. It doesn't show that this is why they were awarded.

Sam

If a job ad asks you to use AI, show the employer the time it saved you. Nearly half the ads we read want AI for speed.

For our report The AI Ask we read 43 Irish marketing and sales job ads that ask the person to use AI tools. 20 of them say the AI is for working faster or cutting manual work. Only 7 say it is for writing or making content.

I expected writing to come first. It didn't.

Dropbox is a good example. It is hiring a sales operations manager, and the ad wants AI "to reduce manual reporting effort while maintaining strong human judgment". So it wants the speed, and it wants a person who still makes the call.

Here is what to do.

1. If you are applying, pick one task you now do faster with AI. Say how long it took before and how long it takes now.
2. Say how you checked the result.
3. If you are hiring, say in the ad what you want done faster.

One caution. 43 ads is a small number, and these are the only two reasons I counted.

Sam

If a job ad asks for "AI tools" and names none, answer it with names. Say which tools you use, what you use them for, and how you check what comes back.

We read 636 Irish marketing and sales job ads that were live on 24 September 2026. They are part of The AI Ask, our quarterly count of what these ads ask people to do with AI.

Employers name their other tools. Salesforce is named in 51 of the 636 ads. Excel is named in 43. Canva is named in 22.

With AI it is different. 56 of the 636 ads ask the person to do something with AI.

So we looked in every ad for nine well-known AI tools, among them ChatGPT, Claude, Gemini and Copilot. 6 of the 636 ads name one. In 3 of the 6 the name is not something the ad asks of the applicant. That leaves 3 ads that ask the applicant for one of those tools. Osborne Recruitment lists ChatGPT. DocuSign asks for familiarity with Gemini and NotebookLM. MongoDB asks for a deep understanding of coding tools such as Claude Code and Cursor.

I don't read this as employers not caring. I think it is early. An employer can know it wants people who use AI well and still be working out which tools it needs. That is my view, and the ads can't prove it.

This is one day of ads, and 56 is a small number. We count again in December.

Here is what to do.

1. If you are applying and the ad says "AI tools", name yours. Give the tool, the task you use it for, and how you check the result.
2. If you are hiring and you know which AI tools your team uses, name them in the ad and say what they are for.
3. If you are hiring and still working that out, say so in the ad. Then ask candidates which tools they use and why.

Sam

Some of what a campaign does you can count. Some you can only estimate. Keep the two apart, and work out your return on the part you can count. Here is a case that does it.

In February 2011 an IBM computer called Watson played the American quiz show Jeopardy! against the two finest players the game had seen, and won. IBM sells to businesses, so no consumer was going to buy a Watson. The campaign around the show was there to bring in business buyers. It was later written up as a paper for the IPA Effectiveness Awards and won Silver in 2012. What follows is from that paper.

Some sales IBM could count. It records where each sales lead comes from. Leads that came in through Watson events and materials turned into $254 million of booked revenue in North America.

Other sales it couldn't count. Watson ran on IBM's POWER servers, and sales of those servers had been falling for a long time. After the show they grew, and by more than the Watson leads explain. Did the campaign make buyers keener on those servers, even where no lead recorded it? The paper thinks so, but it can't count it.

So it compares two regions. North America saw a lot of the campaign. In Western Europe IBM ran no Watson marketing and the show wasn't on television, though news of it travelled. Before the show, sales of those servers in the two regions moved closely together. After it, both grew. The paper takes the Watson leads out of North America's growth and the fall in the euro out of Europe's. The gap that is left is $68 million of revenue.

This is the part I'd copy. The paper doesn't say the campaign brought in that $68 million. It says $68 million is the most the campaign could have brought in, because the two regions differed in other ways too: "It's an upper limit as other factors in which W. Europe and NA differed will be involved". Then it leaves the $68 million out of its return altogether. The return it claims is profit of at least 2.4 times what the campaign cost, worked out on the $254 million alone.

One caution. The paper doesn't say how much of the $254 million IBM would have won anyway. It asks the question, calls it a valid one, and says that isn't how IBM judges a campaign. So we know that revenue came in through the campaign. We don't know it was extra.

The paper holds a good deal more and is worth reading in full. You can read the IPA's summary, and buy the paper, on a public page on the IPA's site.

Here is what to do in your next results paper or board report.

1. Split your result in two: what you can count, and what you think the campaign also caused.
2. For the second part, find a market like yours that saw less of the campaign, and compare the two. Call the gap the most it could be, and say what else could explain it.
3. Work out your return on the first part only. Say what you left out.

Sam

If you are writing up a campaign, for an award or for your own board, put one figure in before the results: what you would have sold if the campaign had never run. Here is why.

I've spent the week on award papers. We took the 183 papers we hold from five effectiveness award competitions between 2011 and 2018. 96 of them were awarded and 87 weren't. AI agents on our team read every paper from start to end against the same written list of questions.

I expected the insight to be the difference. I was wrong, and it took 183 papers to tell me.

99% of the awarded papers state an insight (95 of 96). So do 93% of the papers that weren't awarded (81 of 87). The insight is so important that almost nobody enters without one. Having one puts you level with everyone else.

So where is the gap? The three widest gaps we found are all about proof, and this is the one I'd start with. 83% of the awarded papers give a figure for what would have happened anyway (80 of 96). 53% of the others do (46 of 87).

The IPA's guide to working out payback calls that figure the base level of sales: "sales that would have been made even if the work hadn't run". Take the base away from what you actually sold, and what is left is what the campaign added. Without it, sales going up while a campaign runs is correlation, and nobody reading can tell how much of the rise was coming anyway.

These counts have limits, and the first is the lesson turned on ourselves. They are correlation too. They show what awarded papers do more often. They don't show that the base figure is why those papers were awarded.

We checked the count two ways. A second agent read 32 of the papers again without seeing the first agent's answers, and the two agreed on this figure for 29 of the 32. Awarded papers are also edited before they are printed, which could make them look more complete than the others, so we compared only the papers printed in the same form. The gap is smaller there and it is still there: 87% of those awarded papers give the figure and 64% of the others.

And we counted whether a paper states an insight, not how good it is.

Here is what to do before you write your next paper or your next board report.

1. Before the results, write one line: without this campaign we would have sold this much. Say how you worked it out.
2. If you have no model, extend your own sales trend into the months the campaign ran. The IPA's guide lists that method beside econometrics and regional analysis.
3. Lead with the difference between what you sold and that line. That difference is what the campaign added.

Sam

Sales going up while your campaign runs doesn't show that the campaign did it. To show that, you have to rule out everything else that could have caused the rise, one thing at a time. Marie Curie's paper on its 2010 campaign does it better than almost any I've read. It won Gold at the IPA Effectiveness Awards in 2011.

I've spent this week reading award papers. This one is a lesson in rigour, and the discipline in it is open to any team.

Marie Curie gives free nursing care to people who are terminally ill. Every March it runs the Great Daffodil Appeal, and just over a quarter of the money comes from volunteers with collecting tins. Most charity advertising asks people to give money. In January and February 2010 Marie Curie advertised, two thirds of it on radio, to ask for something else: an hour of your time, to collect. The whole campaign cost £184,151.

On the face of it, it worked. The charity had 13,774 collectors the year before and over 20,000 that year. That is correlation: two things that happened together. Causation is the harder claim, that one made the other happen. The paper sets out to show causation, and it starts by saying what it didn't have: "Due to the low investment in the campaign, we do not have any tracking research at our disposal."

So it did three things.

First, it looked at the route people came in by. Collectors normally sign up through their local branch. The campaign advertised a national phone number, a web address and a Facebook page, and the paper says every new recruit came in through one of those three. It shows the split as a chart with no numbers on it. Most came through the website. The paper calls only the phone line dedicated, and doesn't say whether the web address was a page of its own.

Second, it set the results against where the ads ran. The radio was heavier in some regions than in others. The paper puts the radio weight in each of six regions beside the new recruits there. The more radio a region had, the more new recruits it had. One region is out of line: Wales and the West had more radio than Scotland or the North and fewer recruits.

Third, it wrote down everything else that could explain the rise, and answered each one. There are six on its list. Here are two.

• The main appeal advertising ran in March, and collectors had to register by the end of February. So the bigger campaign couldn't have recruited them.
• Compared with the March before, March 2010 was colder, less sunny and equally wet. So a fine month didn't bring the collectors out.

That last answer came from the Met Office, and it was free.

The paper is just as careful with what the campaign paid back. Collector numbers were on a rising trend anyway, though they didn't rise every year. So it works the return out three ways, each against a different base, and uses the lowest of the three from there on: £2.45 for every £1 spent, after the cost of the campaign is taken off.

Read that figure with one thing in mind. It covers two years, and the second year is an assumption: the charity believes 46% of new collectors come back the next year. On the first year alone, by my own sum from the paper's figures, the return is about £1.39 for every £1.

One limit. This is one paper, and I can't tell you which part of it the judges liked. It also won the prize for Best New Learning, and the paper's own section on what it learned is about asking people to collect, not about the proof. So the idea may have counted for as much as the rigour.

The paper holds a good deal more than this: the other four causes, the media plan, the charts and the full working of the return. It is worth reading in full. You can read the IPA's summary, and buy the paper, on the IPA's site.

Here is what to take from it before your next campaign.

1. Give the campaign its own way in: a phone number, a web address or a code that nothing else uses. Then count what comes through it. Marie Curie's paper can't say how many came by each route, because it prints no count.
2. If you can, don't run the same weight everywhere, and keep a note of where it ran.
3. Write the list of everything else that could explain a rise. Find one fact for each. Start with the weather.

Sam