Fundraising 2011/Chapter Materials/Report WMFR

This version is a draft (but comprehensive and articulate enough to be published).

This report was written by Wikimédia France (WMFR) for the 2011-2012 fundraiser.

Wikimédia France and Wikimedia Foundation signed a fundraising agreement in November 2011. By this agreement, users that were located by their IP address as browsing from France and that read pages on Wikimedia projects were shown banners directing them to a landing page asking them to make a donation to Wikimédia France.

Wikimédia France raised 1 041 000 € from nearly 38 000 donors during the fundraiser. We used appeals (banners and landing pages) from 8 Wikimedians.

One great news is that we could reduce our dependance to the Jimmy banner/landing page and still had a large growth. We went from 35 days with a full Jimmy banner in the 2010-2011 fundraiser (and 18 days with 10% Jimmy) to 23 days with a full Jimmy and 7 days with Jimmy at 50%. This steep reduction in Jimmy days still meant our fundraising total grew 108%, our donor base grew 3-fold. Furthermore, we were able to share stories from several Wikimedians with readers, both receiving donations and explaining to readers why Wikimedia projects are so cool.

Wikimédia France used extensive testing to further improve its knowledge of online fundraising and to ask for donations more efficiently. Many aspects of the donation experience were covered:

layout of the donation forms,
graphical aspects of both banners and landing pages,
wording of the appeals,
proposed amounts.

We also improved and developped internal procedures to deal with the large influx of donors. We also strenghten our bond with big donors, successfuly trying to retain most of them. One more important source of pride is that we really improved our knowledge of donors.

Now, we know more on the demographics of the donors, and also on their expectations for Wikimedia.

All in all, the 2011 fundraising was the most successful for Wikimédia France. Not only in amounts but in experience and enhancement of the whole thing.

General considerations

Duration

WMFR ran some tests at the end of October (26th) and early November (4th and 10th). The fundraiser was kicked off for logged in users on November 8. On the 14th banners were shown for the logged out users. Banners for logged in users were taken off on the November 27. The final banner asking for donation was taken off on January 3, and a banner thanking donors was shown until the end of the week (a large number of donations came on week ends so we had to extend the thank you past the week end). The thank you banner was removed the 9th of January 2012.

Statistics are computed for a period from November 4 to January 15. The 2010-2011 fundraiser lasted from November 15, 2010 to January 6, 2011 (52 days). The 2011-2012 fundraiser lasted 50 days.

Logged-in campaign

The logged-in campaign started on November 8 and ended on November 27. It obviously brought less donations (200 donors for 7371€) that the logged-out campaign but our aim was different.

The logged-in campaign was an opportunity to communicate with editors and inform them of the achievements of Wikimédia France. It was also our first attempt to define and communicate on a common cultural basis among Wikimedians.

We started the logged-in campaign first because we wanted to show editors some special care: starting and ending early, not intruding on their visual experience of Wikimedia projects (there are always complaints on local village pumps that the banners are too intrusive or that banners are advertising, we tried to avoid those complaints). We also wanted to ask regular users if they were willing to share their Wikimedian story for the fundraiser (and the current banners/Landing page was an example to show them). Calls were made on several projects (Wikipédia, Commons, Wikisource, Wiktionary), none was successful.

On November 15, we set up a link from the logged-in landing page to our membership form.

Communication

We used three kind of communication. One for the general audience, one for the Wikimedians, and one for the larger Internet users.

For the first audience we sent one press release few days before the start of the campaign. It had some echo but not as much as we could hope. Nevertheless we didn't care that much since we're the media: people know about the ongoing fundraiser way more when they go read an article on Wikipédia than when the read their newspaper. We had the usual high number of requests for interviews (both during and shortly after the fundraiser) and we covered all of them. At the end of the fundraiser we also wrote a piece on our own blog.

We initially defined a larger strategy for the general audience, including contacting major news provider to figure on their home page, and influent Internet users. But we prefere to emphasize our efforts on improving our conversion funnel. Those communications experimentations will be done in the coming month, outside of the fundraising..

We reached out to Wikimedians through local pumps, mailing lists, and social medias (mainly Twitter and Facebook).

We also kept in touch with the larger Internet community via social medias and blogs. There were some controversies:

banners being equated to commercials,
Wikimedia movement being accused of not supporting free software and free movement enough,
concerns regarding the allocation of the money raised.

We tried to be present, explain how things actually worked and answer the concerns as much as possible (but obviously there's sometimes trolling and reason doesn't suffice).

We also answered donors and users that mailed us or wrote to us through our website, or even by postal mail. It is very important to us to show donors and users that we care. During the course of the fundraising we answered to more than 50 phone calls and around 350 donors by email.

Big donors

We expanded our program for major donors. Donors making a donation between 150 and 500 € (gold level) were thanked by a personal letter from Rémi Mathis, the president. Donors making a donation above 500 € (platinum level) furthermore received our printed annual report (2010) and the opportunity to figure in the 2011 report.

A very successful fundraiser: over 1 million euros collected

During the fundraiser, we raised 1 041 000 € from nearly 38 000 donors. The success of the fundraiser allowed us to test the limits of our procedures and infrastructure.

The technical infrastructure worked well. The server experienced no downtime at all. There were a couple of emails on OTRS complaining about buggy behaviors. These bugs were solved fast (donors were trying to help, providing snapshots and describing their donation route). Some donors mistakenly revealed their identity in our golden book of donors. Once noticed, we anonymized their donation within minutes.

Despite the steep increase we were still able to answer donors' questions within one workday. Some questions led to extensive research and really helped improve our expertise in tricky fiscal matters. Donors were also thanked fast : All our donors received confirmation for the gift with a thank you note within the last day of the fundraising. Big donors received a special treatment (see below).

Whilst it is not yet finished, we estimate that all donors who requested a tax document will get it before the end of March, well in time before the tax season is launched. Soon after the beginning of the fundraiser, we realized that our secretary, Élisabeth, would likely not be able to produce all tax receipts within a reasonable timeframe, due to a big hike in the number of donations (due to a voluntary decrease of default amount on the donation page, the average donation significantly decreased compared to 2010-2011 but the number of donors sharply increased). We consequently hired the services of another secretary to second Élisabeth. Louisa has been under a contractual agreement with Wikimédia France until mid-March, and the issuing of 2011 receipts has been finished mid-March.

Wikimédia France's fundraiser was volunteer-driven. A large part of the time of the volunteers was spent on negotiating with the Wikimedia Foundation.

Volunteers, both inside and outside Wikimédia France translated the appeals, banners, and other fundraising-related texts on meta. Although most of the appeals we used were written by members of Wikimédia France, we still used the work of translators for 2 appeals: Jimmy's and Susan's. One invaluable part of the work of the volunteers was the advice they provided to the fundraising team: volunteers come from various background and their expertise could shed light on specific questions, for instance on fiscal law, bank-related issues, press relations or user-interface.

Amount collected

From November 4 to January 15, we received approximately 1 041 000 € from nearly 38 000 donors. The sum is a 108% increase from last fundraiser and the number of donors increased 3.1 times.

Throughout the fundraiser the amount collected each day vary a great deal. But overall we found the same pattern as last year's fundraiser: strong start, slow decline until early december, a peak on the first week-end of december (probably due to the arrival of the november pay-check), low tide, dismal number on Christmas (people are away from their computer), donations flow back in until a big burst on New Year's eve to benefit from tax deductibility. This is the big picture, but as you can see from the daily chart (right), the big picture has discrepancies: even at low tide there are peak, mostly due to sundays.

At the start of the fundraiser we tried to predict, assuming the above-described pattern would be reproduced, how much we would get and if we needed to be agressive in the banner language. The prediction looked good so we sticked to a tepid invitation to read the landing page. One other thing that really startled us is that, for roughly the first 4 weeks of the logged-out fundraiser, the daily results matched almost perfectly the one from 2010-2011 by a 2.46 factor (green and yellow curves). The match was even better if one looked at the day-to-day growth in 2010-2011. The "magic" factor was not fitting for the remainder of the fundraiser, from the moment on we entered the low tidal phase.

The pattern for number of donations per hour was: strong numbers on the morning (10-11), slide during lunchtime, slow increase until the end of the workday (18), sliding (getting back from work, eating), small peak after dinner, and off to sleep. The amount in donations followed roughly the same pattern: strong peak between 10 and 12, lunch, slow increase, peak at 17-18h, slide, small peak after dinner, and off to sleep. The 10-12 peak is stronger in number of donations than in amount, meaning the average donation is lower by that time of the day. The only noticeable change between the weekdays and Sundays is that the evening peak is comparatively a bit higher on Sundays than that of weekdays.

Number of donations by hour
Amount of donations by hour

Types of donations

Donors had three ways to make a donation: online donation through our website https://dons.wikimedia.fr, checks, and wire transfers.

on-line donation. This method represents 92.2% in amount and 94.3% in number of donations,
checks. This method represent 7.0% in amount, and 5.2% in number of donations, and
wire transfers. This method represents 0.77% in amount, and 0.46% in number of donations.

Sidenote, we don't propose Paypal due to really high fees. We're studying the relevancy to propose it for the next fundraiser. We will also launch in the coming month the possibility for donors to subscribe to a monthly donation.

Origin of donations

The average donation from Wikimédia France's websites is way above the average donation from banners and landing pages (43.7 € vs 25.8 €, truncated average: 35.97 € vs 24.8 €).

The projects from which we obtain the most donations are in decreasing order:

Total number of donations	Average amount of donations	Total amount of donations	Project
27492	25.8642	711060	fr.wikipedia.org
3562	26.9587	96027	en.wikipedia.org
748	22.6832	16967	fr.wiktionary.org
147	28.2517	4153	de.wikipedia.org
104	23.0769	2400	fr.wikisource.org
85	24.2353	2060	es.wikipedia.org
82	25.1829	2065	it.wikipedia.org
69	24.9275	1720	commons.wikimedia.org

The table above gives the main referrers sort by decreasing amount.

Total amount of donations	Total number of donations	Average amount of donations	Project
99210	2712	36.58	http://fr.wikipedia.org/wiki/Wikipédia:Accueil_principal
7617	231	32.974	http://en.wikipedia.org/wiki/Main_Page
2995	149	20.1007	http://fr.wikipedia.org/wiki/Louis_Daguerre
2515	50	50.3	http://fr.wikipedia.org/wiki/Spécial:Recherche
2425	80	30.3125	http://fr.wikipedia.org/wiki/Wiki
1635	13	125.769	http://fr.wikipedia.org/wiki/Don
1030	2	515	http://fr.wikipedia.org/wiki/Cénobitisme
1020	3	340	http://fr.wikipedia.org/wiki/Modèle_standard_(physique_des_particules)
1000	1	1000	http://fr.wikipedia.org/wiki/Projet:Pyrénées
1000	1	1000	http://fr.wikipedia.org/wiki/Paul_Ricoeur
1000	1	1000	http://fr.wikipedia.org/wiki/Marie_Walewska
930	56	16.6071	http://fr.wikipedia.org/wiki/Stanislas_Lem
910	2	455	http://fr.wikipedia.org/wiki/Eminem

When preparing for the fundraiser we thought of including a Konami code (using JavaScript) into the page related to video games, as some sort of easter egg for donors. It was technically possible but very hard to put in place. We still think it would be a great idea to put some easter egg for our potential donors. That's for the fun part.

One more sensitive reason to have some targeted banners (i.e., banners adapted to the page you're visiting) is the following table of donation (in number and in amount) per page. Obviously there is a long tail of page leading to a couple of donations. There are also page leading to huge big donations (although we doubt it has anything to do with the subject, for instance cénobitisme). But the main point here is that several page get huge donor exposure: main pages (2712 of fr.wp, 231 for en.wp). They get a higher average donation 36.58€ for fr.wp's main page and 32.97€ on en.wp's main page. Notice also the 2515 donors from the search page. One last remark: Google doodle. We know that for some occasions Google adapts its search form to celebrate someone or something. This effect can be seen in the table above: Google made a tribute to Stanislas Lem, author of Solaris, on November 23, and also one for Louis Daguerre on November 18. There were also doodles for Mark Twain's 176th birthday on November 30 (41 donations for 862€), and for Diego Riviera's 125th birthday on December 8 (14 donations for 462€).

We're considering, for next year, to adapt our messaging to those dates to give donors a more contextual message.

Amount distribution

The distribution of donations from projects is:

5 €: 3.4% of donations
10 €: 45.9%
15 €: 2.3%
20 €: 4.7%
30 €: 26.7%
50 €: 10.6%
100 € and above: 4.6%.

The distribution of donations from Wikimédia France's website is:

5 €: 8.4% of donations
7 €: 8.9%
10 €: 8.5%
15 €: 3.7%
20 €: 6.8%
30 €: 29.8%
50 €: 18.5%
100 € and above: 9.1%.

Neither add up to 100% but these are the most given amounts. Besides the distribution varies a great deal between weekdays and week-ends, between early December and the last week of December, between banners and so on.

The donors on WMFR's websites were proposed default amounts of 7, 30, 50, 100, and a free-to-choose amount. Coming from the projects, the amounts varied: 10, 30, 50, 100, and free was chosen most often but we also tried other assortment of amounts, like 15, 35, 60, 100, and free amount, and 15, 30, 50, 100, and free amount. The minimum amount is 5 €.

Tax deduction

On average, 57.8% of online donors asked for a tax receipt. But it pretty much depends on the amount of the donation:

20.4% of donors with a 5 € donation want a receipt,
38.9% of donors with a donation ≤ 15 € want a receipt,
74.9% for donations between 15 and 50 € want a receipt,
and 85.1% for donations above 50€.

By default, the set-up is "no tax receipt" so these 57.8% explicitly chose to get a receipt.

When the amounts are split by intervals of length 10, we obtain the following curve:

The amounts above 100 € are bundled together and shown at point 100. The amount on the interval [x, x+10[ are shown at point x+5.

Overall, 690665 € of online donations are the object of a request for tax deduction (that is 71.8% in amount) and 270071 € are not (28.1%).

Of course, some people forget to ask for a tax deduction or make a mistake while filling the form. On the other hand, some people might ask for a tax deduction and not use it. We have no way of evaluating the number of people in those two categories. We also have people making donations by checks or wire transfers. The numbers are not fully known as yet.

One other interesting thing is that people making a donation from en.wp are less likely to ask for tax deductibility: 46.4% of en.wp donors ask for tax deduction (59.1% in amount). On fr.wp the corresponding figures are 58.4% and 72.1%.

Demographics

During the 2010-2011 fundraiser, 81.8% of donors were men and their average donation was 40.24€. Average donation for women was 35.46€. During the 2011-2012 fundraiser, 73.6% of donors were men (average donation: 28.36€). Women had an average donation of 23.41€. Some donors do not state their gender, nor is it possible to identify a gender (for instance on a joint bank account).

Last year, the Lila banner had a higher response rate for men, both in gender percentage and in average amount, compared to Jimmy's banner. The gender distribution was nevertheless close between the two banners and we could not draw any confident conclusion. This year:

When the Rémi banner ran at 100% on logged-out users, 73.48% of donations were made by men (76% in amount). Men average 34.27€ per donation while women average 30.0€.
When the Jimmy banner ran at 100% at the end of December, 74.19% of donations were made by men (77.46% in amount). Men average 31.11€ per donation while women average 26.02€.
When the Susan banner ran at 100%, 80.97% of donations were made by men (83.54% in amount). Men average 36.31€ per donation while women average 30.43€.
When the Charlotte-Benjamin banner ran at 100%, 79.6% of donations were made by men (77.2% in amount). Men average 35.43€ per donation while women average 40.91€ (including a high number of women making big donations. If those donations above 100€ are removed we have averages of 33.42€ vs 31.57€).

One factor mitigating the drawing of conclusion is that those banners were not shown at the same period. However there is a pattern: a woman in a banner/landing page increases the number of men making a donation (or symmetrically decreases the number of women making a donation, hard to say). Is it the picture or the writing of the appeal? How much of gender is involved in the choice to donate and the amount of the donation?

Unsurprisingly most donors give an address in France (99.0%). The other countries where donors live are, by decreasing number of donors: Switzerland (0.13%), Belgium (0.1%), Spain (0.09%), Italy, the United States, Monaco, Germany, United Kingdom, China, Luxembourg, Canada. Those donors are making a donation to WMFR either because geolocation by IP is not exact for bordering countries (most of the above-mentioned countries have a border with France), or because those people actually work in France, or travelled to France during the fundraiser.

We also built a large amount of statistics on the disparity of donation according to the demography of France. These statistics are of little use for an international audience. There is a great disparity between, Paris, and the greater Paris area on one side, and rural départements like Cantal or Corsica. The above map shows the average donation per département.

Big donors

We retained 59.3% of big donors from 2010-2011. Gold donors grew 68.2% in number (66.2% in amount) and platinum donors 56% (54.8% in amount). The average amount for gold donors is 310€ and the average for platinum donors is 768€. The global growth on donor is partly fueled by small donors. This explains why big donors don't grow that much in sheer number compared to the 211% overall growth. The effect of big donor cultivation can not be measured until the next fundraiser.

There are nevertheless some platinum donors not wishing to figure in the report, one reason is their wish to remain anonymous (no, Anonymous is not funding WMFR), another is that they feel their donation needs no more thanks than any other. Maybe we could set a number of donors (say 20) to thank in our yearly report, chose uniformly at random some donors, and ask them if they wish to figure on the report. Or the bottom 20, or the top 20. Maybe there's a way for us to show we're running things a bit differently, Wiki-style. From the contacts I had with platinum donors, they would feel ok about not figuring in the report.

We also envision new programs to thank some category of donors: those donors that make a donation every year (with a minimal threshold on the amount), and those making several donations throughout the year.

Specific results of logged-in campaign

The logged-in campaign obviously brought less donations than the general campaign (204 donors for 7431€) that the logged-out campaign but our aim was different.

A short methodological note: logged-in is only for online donations from Wikimedia projects (not from WMFR sites) since it is not possible to guess if the user has an account on a Wikimedia project.

We had a steady stream of membership applications throughout the FR (roughly 120 new membership requests). That's an impressive number. Another impressive fact is that most of those new members are not regular editors of Wikipedia or other Wikimedia projects (although they must have some account on a project).

204 donors came through the logged-in campaign, for 7431€ (an average of 36.42€ that is higher than that of the logged-out campaign by 10.63€). 185 via the banner (for 6946€), 14 via the donate link on the sidebar (for 420€), and 5 from unknown origin (for 65€). 187 donors coming from fr.wp (for 6981€).

We ran 3 banners and 2 landing pages: 2 Jimmy banners, one with Jimmy smiling in a black shirt with transparent background, one with Jimmy with a yellow jacket half smiling in the green. Both led to the same landing page. We also ran one Rémi banner, smiling with a brick background. We had little donations (so statistical evidences have limited confidence) but still the difference in average amount is impressive: Rémi had a 43.7€ average while Jimmy's had averages of 28.0€ and 36.89€. If one looks closer and removes the big donations (say above 100€), then averages are more comparable: between Rémi had 56 "standard" donations (i.e., donations of an amount smaller than 100€) for 1600€ (average: 28.57€), and the first Jimmy had 51 donations for 1310€ (average: 25.68€). When faced with the second Jimmy (Green Jimmy), Rémi did 34 "standard" donations for 1025€ (average: 30.14€) and Jimmy had 47 donations for 1471€ (average: 31.29€). The best Jimmy (Jimmy Green) had the upper hand on Rémi on average and in absolute number for "standard" donations, but Rémi still managed to get the better of Jimmy due to two big donors. One could say that those two big donors could have fallen on Jimmy's banner, hence saying Jimmy did better if we removed chance. One other way of seeing this is that big donors don't give 1000€ by an unbiased coin tossing, they give it a careful thought and the choice of big donors is not uniform, hence Rémi wins.

Advanced A/B testing and further data analysis

Appeals

We ran several appeals (6) from French Wikimedians (mostly Wikipedians, but also one appeal on Commons and one on Wikinews). The donors' response to the banners was a mixed bag we're still analyzing. It seems that donors favored senior citizens to younger people and people with fancy titles (founder, president) to "user/writer".

Amount distribution

The donors on WMFR's websites were proposed default amounts of 7, 30, 50, 100, and a free-to-choose amount. Coming from the projects, the amounts varied: 10, 30, 50, 100, and free was chosen most often but we also tried other assortment of amounts, like 15, 35, 60, 100, and free amount, and 15, 30, 50, 100, and free amount. The minimum amount is 5 €.

The free amount is used 10.8% of the time for donors from the Wikimedia projects.
When the proposed amounts are 10, 30, 50, 100, and free amount, 4.78% of donors chose a 20€ amount, 3.48% a 5€ amount and 1.28% a 15€ amount.
For donors from Wikimédia France's websites, the free amount is used 33.22% of the time. Most often chosen amounts are: 10€ (8.8%), 5€ (7.21%), 20 (7.21%), and 15€ (3.94%). The minimal proposed amount of 7 € (chosen by 8.37% of donors) may seem weird and "unround" to the potential donor, and therefore does not "unite" donors willing to make a modest donation (between 5 and 20 €). There is also a different profile of donors donating from WMFR's sites.

On November 18, we A/B tested two landing page (Jimbo's appeal), one (A) which proposed 10 € as minimal amount, and one (B) which proposed 15 € as minimal amount. Setting a higher minimal amount is sure to deter some potential donors to make a donation (10 € being in their budget, and 15 € being above). Increasing the minimal amount obviously increases the average. How many people donating 10 € in sample A would chose to donate 15 € if they were proposed this amount? How many would not make a donation? How many would use the free amount to fill the amount they want? Will the higher average donation offset the smaller number of donors? 383 donors made a donation through A for a total of 8885 € (average: 23.19 €), 317 donors made a donations through B for a total of 8380 € (average: 26.43 €). In the 10-30-50-100 setting, 10.2% of donors chose a different amount, they were unsatisfied with the proposed amounts (nearly 5% choosing to donation 20€). In the 15-30-50-100 setting, 7.5% of donors chose a different amount (3.4% choosing to donate 10€). Otherwise, not much could be said from the distribution.

Between December 9 and December 14 we tested two sets of amounts on Rémi's appeal: one (A) had amounts of 10, 30, 50, 100, and free amount, another (B) had amounts of 15, 35, 60, 100, and free amount. Before that the distribution of amounts from Rémi's appeal was smoother than that of Jimmy's appeal. Amount distribution for Rémi:

10€: 33.8% of donations,
20€: 5.3%,
30€: 33.5%,
50€: 17.2%,
100€: 6.6%.

So we figured that increasing the proposed amounts to the next roundish number could be interesting for the same reason as above. One assumes that donations above 100€ are not affected by the setting, so we actually compare the results for donations smaller than 100€. Setting A led to 530 donations for 13806 € (average: 26.0€). Setting B led to 471 donations for 13210 € (average: 28.0€). Setting A does slightly better than setting B in amount. Setting A has 11.4% of free amounts (5.2% of 20€, 2.0% of 15€, 1.5% of 5€) while setting B has 20.3% of free amounts (5.6% of 10€, 4.0% of 20€, 3.1% of 50€, 2.3% of 5€, and 2.1% of 30€).

Setting A:

5€: 1.5625
10€: 32.4653
15€: 2.08333
20€: 5.20833
30€: 30.2083
35€: 0
50€: 18.9236
60€: 0.347222
100€: 6.94444

Setting B:

5€: 2.32108
10€: 5.60928
15€: 35.5899
20€: 4.0619
30€: 2.12766
35€: 23.9845
50€: 3.09478
60€: 12.3791
100€: 7.73694

Let us focus on donors which donate 50€ in setting A (18.9% of donors). What would those donors do if they were in setting B? 65% would follow the amount hike and donate 60€, 15.8% would remain on the same amount (using the free amount), and 19% would go down to 35€ (their budget allows them to go to 50€ but not to 60€ so they go do the highest proposed amount that they can afford. They do not use the free amount). Overall, instead of donating an average of 50€, they would donate an average of 53.55€. Another question is: how many 50€-donors in setting A chose not to donate in setting B? For instance because the proposed amounts are not satisfactory (and they don't use the free amount). Of course the experiment was run on samples of size a bit less than 500 so one has to be wary of the margin of error.

One can repeat the reasoning for 10€ donors of setting A (32.4%). In setting B, 2.3% would go down to 5€ (their budget would not allow them to increase their donation to 15€), 17.2% would donate the same 10€ (using the free amount), and the rest would go up to 15€. Since 35.58% of donors make a 15€ donation, those do not only come from the 10€ donors from setting A, they also come from 30€ and 20€ donors from setting A (let's discard the other "transitional" amounts (22€, 24€…), since they are of little statistical significance). The average amount the 10€-donors of setting A would give in setting B is 13.91€.

Finally, 30€-35€ transition: that part is tricky because it is impossible from the data we have to know what the 20€-donors from setting A would do when confronted to setting B. Of the 4% of 20€ donors in setting B, how many are 20€ donors from setting A, and how many are 30€ donors from setting A that can not afford to go to 35€? We have two scenarios: one (S1) where all the 20€ donors go down to 15€, and one (S2) where all the 20€ donors from setting B are 20€ donors from setting A (meaning 78.0% of 20€ donors from setting A keep that amount in setting B) and the additional donors are 30€ ones from setting A. In S1, the 20€ donors that average 20€ per donation in setting A, average 15€ in setting B. And of the 30€ donors in setting A, 12.0% go to 15€, 13.4% to 20€, 7% remain at 30€, and 67.4% increase their donation to 35€. Than means 30€ donors in setting A make an average donation of 30.22€. Not much of an increase. Moreover, 20€ donors of setting A gave 5€ less. On 20 and 30€ donors in setting A, the average amount went from 28.53€ to 27.98€. In scenario 2, the maximum number of 20€ donors remain 20€ donors, the rest (21.9%) become 15€ donors. The average goes from 20€ to 18.9€. 30€ donors in setting A donate 15€ 6.27% of the time, 7.01% stick with 30€, and 72.2% increase their donation to 35€. Overall the average goes from 30€ to 30.47€. But if one looks at the 20 and 30€ donors, the average goes up from 28.53€ to 28.77€.

For the 2012-2013 fundraiser, we should test a 10-30-50-100-free amount setting against a 10-20-30-50-100-free amount.

Graphical aspects and wording

We tested some graphical aspects in the banners and landing page.

On November 4, we tested Jimmy banners during 2 hours. One banner had a picture of Jimmy flushed left, and the other one had Jimmy flushed right, the text being centered. The left Jimmy outperformed the right Jimmy. Clickthrough was 0.014 (14039 landing page views for 1000857 banner views) to 0.0126 (12698 for 1001675).

Our colleagues at WMDE told us (on IRC) that a picture of the appealer on the landing page was effective. So we gave it a try. We compared two landing pages for Rémi: one (A) with no picture, and one (B) with a picture of Rémi on the right (http://wikimediafoundation.org/w/index.php?title=Template:2011FR/appeal-remi-FR-1202&oldid=75712). The banner was the same and led uniformly to each of the two landing pages. Jimmy ran 50% on the other side for one part of the test. The landing page A led to 335 donations for 10788€ (average: 32.2€), while the landing page B led to 339 donations for 10347€ (average: 30.5€). If we remove the donations above 100€, landing page A had 331 donations for 9788€ (average: 29.5€), landing page B had 336 donations for 9647€ (average: 28.7€).

Next we compared a landing page (A) with no picture to one (B) with a picture of Rémi on the left (http://wikimediafoundation.org/w/index.php?title=Template:2011FR/appeal-remi-FR-1202&oldid=75793). Landing page A led to 325 donations for 10347€ (average: 31.8€), landing page B led to 314 donations for 9281€ (average: 29.5€). If we remove donations above 100€, landing page A led to 322 donations for 9647€ (average: 29.9€) and landing page B had 311 donations for 8681€ (average: 27.9€). So we stuck with no picture of the appealer in the appeal (landing page A). But this illustrates how cultural differences (between Germany and France) lead to different results.

On Wikinews, the banner was without any picture since the appealer didn't want his picture to appear. This would have been an interesting test if we had had some donations from Wikinews. Unfortunately we only had 5 donations for 70€.

We A/B tested wording for some of the appeals. We tried to keep the argumentation but with smaller paragraphs. We also tried shorter sentences, putting some faster rhythm in the text (thinking that maybe a faster rhythm would suit the younger generation). No change in the text led to any noticeable variation in the donors' response. Meaning, among others, that long appeals were ok.

Languages and localization

From the 2010-2011 fundraiser, we had a 228% hike for donors coming from fr.wp (and a 122% hike in amount). Donors coming from en.wp show a 170% increase (72.2% in amount). Does the lack of localized banner explain this gap (one could assume that since the donors live in the same country they are subject to the same economic situation)? Looking further: de.wp has seen a 116% increase on donors (45% in amount). No German banner/LP was shown during the 2010-2011 fundraiser, nor during the 2011-2012 one. So the language the donors read do not seem to be a major factor (to further back up this claim, we looked at es.wp: 136% growth in donors and 74.5% in amount. No Spanish banner/LP was ever shown). Besides depending on the day, users of en.wp could see English-only banners, mixed banners, or French-only banners. So we have to delve a bit deeper. The growth in readership from en.wp, de.wp, and fr.wp increases the number of potential donors. These figures are most likely not the same on the different projects explaining the gaps between these 3 Wikipedias. Another idea is to compare when English-only banners were shown on en.wp and French banners on fr.wp. We compare between the number and amount of donations for the 2010-2011 fundraiser (when English-only banners were shown on en.wp) and the early Jimbo banner on en.wp and fr.wp (from 14/11 to 29/11). We don't compare periods of the same duration so we adjusted the amount and number of donation figures to the pace of fr.wp. So on fr.wp the increase is still 122% in amount (228% in number of donations). On en.wp, the hike was 84% in amount (180% in number). If we compare between the 2010-2011 fundraiser and the period where en.wp users were shown only French banners, en.wp had a 47% growth in amount and 132% in number. And for the period where en.wp users were shown mixed banners, the growth is 90% in amount and 183% in number. We don't have enough data to compute valid growth figures from de.wp.

The growth of donations (both in amount and in number) from en.wp is much lower than that of fr.wp. One minor explanation is the lack of banners/Landing page in English for some periods. If at least one English banner had always been shown, the growth rate would have been of 85% (estimate) and we might have received an additional 2542 € (estimate). This explanation is minor: even if we consider the period where English banners were show, the growth is only of 84% in amount on en.wp compared to 122% for fr.wp (190% vs 228% in number). Having at least one English banner increases the number and amount of donation.

On the Elsässisch Wikipedia (als.wp, language spoken in Eastern France, well in Elsass :)) we ran a Jimmy banner/landing page translated in Elsässisch. We received no donation from als.wp (nor did we last year). We also translated an appeal into Breton but it was not shown.

Next year, we will try to translate banners in Arabic, Wolof, German, Spanish, Italian, Chinese, Breton, or Basque (languages that are spoken in France) and see whether it leads to some results. Remember also that our goal is not only to fundraise to sustain the global Wikimedia movement but also to reach out to users, future, present, and past. The Elsässisch banner, for instance, shows als.wp users that we care about their linguistic community and helps us to get the opportunity to get cultural content from the Elsässisch community or explain locally how WP works (although we didn't earn anything from it during the fundraiser itself). We might also try something new: bilingual banners switching between the project language and French (this bug has been observed on the Telugu Wikipedia, now we want it to become a feature :)).