<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Benjamin Billon on Medium]]></title>
        <description><![CDATA[Stories by Benjamin Billon on Medium]]></description>
        <link>https://medium.com/@benjamin.billon?source=rss-9dc98e61d32c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*5icZxncC6h17-FJYmIySew.jpeg</url>
            <title>Stories by Benjamin Billon on Medium</title>
            <link>https://medium.com/@benjamin.billon?source=rss-9dc98e61d32c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 13:18:58 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@benjamin.billon/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Carbon footprint of decaying emails]]></title>
            <link>https://splio.tech/carbon-footprint-of-decaying-emails-26b253aad4a7?source=rss-9dc98e61d32c------2</link>
            <guid isPermaLink="false">https://medium.com/p/26b253aad4a7</guid>
            <category><![CDATA[carbon-footprint]]></category>
            <category><![CDATA[email-marketing]]></category>
            <category><![CDATA[expiration-date]]></category>
            <category><![CDATA[change-of-habits]]></category>
            <category><![CDATA[cloud-storage]]></category>
            <dc:creator><![CDATA[Benjamin Billon]]></dc:creator>
            <pubDate>Fri, 15 Dec 2023 16:53:46 GMT</pubDate>
            <atom:updated>2023-12-15T16:53:46.088Z</atom:updated>
            <content:encoded><![CDATA[<p>In <a href="https://medium.com/@benjamin.billon/carbon-footprint-of-marketing-emails-d7ef1cbd39b9">a previous article</a>, I described my journey through email marketing carbon footprint calculation. The published article is not exactly what I intended to talk about, but I felt like it was relevant to share the whole background before coming up with a smaller, or at least less visible part of the footprint: the long-term storage of useless emails.</p><p>As mentioned in the previous article, a big difference between spams and marketing emails is that spams are either rejected during the SMTP transaction, or mostly junked or deleted shortly after delivery. In other words, they are not stored for long, as most mailbox providers have garbage collectors in place to remove from the junk and trash folders messages that have been sitting there for more than 30 days.</p><p>So let’s circle back to the 30 billion marketing emails sent daily. At 60KB each on average, they would account for 1800 TB of new data to store, daily. There also are the person-to-person emails, the mailing-lists with plenty of replies, the 10MB of pictures sent by this friend or family number after a nice week-end, the slide deck or the contract document with revision marking shared by this colleague to you (and a bunch of others), and frankly it doesn’t make a difference at this stage, as they’ll all end up as bytes in disks somewhere.</p><h3>Disks?</h3><p>To this day, cloud storage and datacenters in general still mainly rely on hard disk drives (HDD) instead of Solid-state drives (SDD), simply because it’s cheaper to store the same amount of data. Also, there’s much less data available about the footprint of SSD drives.</p><p>So based on what ADEME said (back in 2008), producing a hard drive weights 15.5 kgCO₂e (fun fact: 23 kgCO₂e per kg of hard drive), and a hard drive runs on average 7 years. We assume that the footprint is roughly the same to produce a 64GB or 10TB drive.</p><p>As per the <a href="https://www.cloudcarbonfootprint.org/docs/methodology/#storage">Cloud Carbon Footprint</a> project, the average consumption of HDD is 0.65 Watt-Hours per Terabyte-Hour.</p><p>We can summarize the carbon footprint of a 10TB hard drive (I picked 10TB as this seems to currently be the cheapest price per capacity):</p><ul><li><strong>15.5 kgCO₂e</strong> for production</li><li>0.65 Wh * 24h * 365 days * 7 years = <strong>39.9kWh</strong> of electricity over the life of the hard drive.</li><li>The average worldwide carbon intensity factor is <strong>460gCO₂/kWh</strong> (it’s 85gCO₂/kWh in France, 367gCO₂/kWh in the US, 503gCO₂/kWh in Australia …), so 39.9kWh of electricity * 0.460 = <strong>18.354kgCO₂e</strong> from electricity consumption during the drive’s life.</li><li>We have no data for end-of-life/recycling, but unfortunately recycling of electronic devices is still rare and <strong>most ends up in landfills in Asia</strong>.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*wwEoCsUd809mdPdA.jpg" /><figcaption>Garbage Is Spread and Compacted in Landfill Operation at Croton Dump 08/1973. Photographer: Blanche, Wil.</figcaption></figure><p>Then multiply this by 18, because we need 180 new TB every day, not just 10:</p><p>Hence: 18 * (15.5 + 18.354) = 609.37kgCO₂e.</p><p>Also, most cloud storage use redundancy to avoid losing data if a disk fails, and the most common technology for this is RAID 10, which basically means that to have 20 TB of redundant capacity, you need 4 disks of 10 TB.</p><p>So: 609.37 * 2 = 1.219tCO₂e.</p><p>Also, backups! Any serious hosting have data backups and data recovery procedures in place, in case something really bad happens. Many have multiple backups, but backups can also be compressed, so for the sake of simplicity, let’s just consider we need twice the storage space.</p><p>Now: 1.219 * 2 = 2.438tCO₂e.</p><p>By calculating this way, the carbon footprint of the storage of new emails is <strong>2.4tCO₂e per day</strong>, or roughly 890tCO₂e per year.</p><p>This result is however almost twice smaller (52.7%) than if we do the calculation using the footprint of 1GB of data stored on Cloud hosting (through wired network) for 10 years, as per the Negaoctet database (0.0253 kgCO₂e): 0.0253 x 180,000,000 = <strong>4.554tCO₂e per day</strong>, or 1663tCO₂e per year. The difference may be explained by more details in the calculation by the Negaoctet database: maybe they have better data than me on backups, maybe cloud hosting and email hosting differ, but they also may be including the footprint of network equipment, servers, coolers, building a datacenter, running a datacenter, cooling down the datacenter, etc. I don’t know if they are considering that a hard drive life duration is 7 years, and therefore it will be replaced over a 10 years period.</p><p>Let’s just consider that my calculation of <strong>2.4tCO₂e per day is optimistic</strong>.</p><h3>Deleting emails</h3><p>The obvious thought therefore is that, maybe, we should delete the useless emails sitting in our mailboxes. Let’s see how to do so, from a computer (I wish it would be this “easy” from a smartphone).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/483/1*YysjKok-nS9nRQOWs5dBhg.png" /></figure><p>I mentioned the volumes of emails with an average size of 60KB, that’s for marketing emails, but you may want to start deleting the <strong>heavy</strong> useless emails sitting in your mailbox.</p><ul><li>In Gmail, search for `larger:10MB` and change the value. And no, you can’t display the size of the message, and you can’t order by the size either.</li><li>In the Outlook email client and in Thunderbird you’d have to show the “Size” column, then click on it to sort by it.</li><li>On Outlook Web Access (on office.com), there’s no column “Size”, but you can filter/sort “By size”</li><li>On Yahoo!mail, I couldn’t find how to do so.</li><li>In Orange’s webmail, you can sort by size, and that would display the size each message.</li></ul><p>(This list is obviously not exhaustive, feel free to let me know if other MUA (Mail User Agent) should be added there)</p><p>So, now that you’re seeing the biggest emails you have, you’ll skim through them, deleting them one by one. That may take some time. Let’s say it takes you 1 hour to clean up one of your mailboxes from the heaviest useless emails, using your laptop to do so. As seen in my previous article, a laptop consumes <strong>5.91Wh</strong> per hour when used actively, so the action of deleting these emails would have a footprint of 0.00591kWh * 460gCO₂/kWh = 2.72gCO₂e.</p><p>Now let’s consider you deleted 1GB of useless emails. Congrats ! We can consider you avoided 52.7% of 25.3gCO₂e = 13.3gCO₂e over 10 years, so compared to the hour you spent doing it:</p><p>13.3 / (10*365.25*24) = 1.5*10^⁻⁴gCO₂e, or <strong>0.005578% of the footprint related to the time spent deleting them</strong>.</p><p>We could argue that only 10 minutes are needed to clean-up the heavy emails, so the footprint would only be 0,4531gCO₂e, while the footprint of hosting this GB during the same 10 minutes was 2.53*10^⁻⁵gCO₂e or …0.005581%. No real improvement here, and you only focused on the heavy emails, not on the thousands of small ones that, added up, may take a few more GBs in your mailbox.</p><h3>Toward Digital Sufficiency</h3><p>Does it mean you shouldn’t have deleted these emails? No, because the main difference lies in the long term storage (if not deleted, it sits there potentially forever), and the scale. Sure, your 1GB is nothing. However, Radicati estimated 4.2 billion email users in 2022. It is unclear if it is about 4.2B persons using emails, or 4.2B mailboxes, but either way, if there’s 4.2B GB of emails to be deleted from mailboxes … that’s not nothing anymore.</p><p>Then, once all 4.2B mailboxes have been cleaned from useless emails, what else can we do?</p><p>For a start, <em>not keeping</em> the new emails that are soon going to become useless.</p><p>Mailbox providers nowadays offer virtually unlimited space for storing emails and other files. Orange has a 10GB mailbox, Gmail has 17GB for emails and files, Microsoft <a href="http://outlook.com/">Outlook.com</a> offers 15GB for emails, Yahoo gives away 1TB.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/743/1*IelCdvuWJ93f6XgGZic4fA.png" /><figcaption><em>🧓 Old-timer note: Back in 2007, Yahoo (then 250 million global users) offered 1GB, while Microsoft (Live.com, a.k.a. Hotmail, then 228 million users) offered 2GB and Gmail (51 million users at the time) offered 2.8GB. When Yahoo Mail was launched in 1997, it included 4MB of storage. This was increased to 100 MB in 2004, 1 GB in 2005, and 1TB in 2013.</em></figcaption></figure><p>It is, for most of us, more than enough for all our needs, meaning that we don’t have to think whether or not we should store something online. It’s nice, it’s convenient, but it contradicts the principle of <strong>digital sufficiency</strong> we are (or should be) aiming at.</p><p>Obviously, Yahoo does not have 1TB x [<em>number of users</em>] data storage available right now. They have a certain amount of storage running, and when the available storage is running low, they’ll buy more disks, install new racks and plug them in.</p><h3>Deleting emails easily</h3><p>By uncluttering our mailboxes and getting rid of useless emails, we’ll decrease the frequency at which mailbox providers have to buy and run new disks. Despite the small part that email takes in the carbon footprint of humanity, this action can still have a positive impact, and is within everybody’s reach.</p><p>But as the “archive email” button, very well highlighted in most apps, managed to become the default action for many email users, we should shift this habit and use “delete” instead.</p><p>That is … if it was made available by said apps.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/595/1*6QHBzRP7bt7YKLDSMU1FPQ.png" /><figcaption>That slightly infuriates me.</figcaption></figure><p>Also, sometimes we don’t want to delete the email <em>just now</em>, because we would like to read it later, maybe. If we do, then we could delete it after reading it. If we don’t … it sits in your mailbox forever.</p><p>That’s where I believe the <a href="https://datatracker.ietf.org/doc/draft-billon-expires/">Expires: header</a> has a role to play. Senders can add such header in the email they send, specifying when the message will become valueless, and could or should be deleted. I’ll be honest, the header should mostly be used by brands in their marketing emails. As mentioned at the beginning of this article, we estimate there are 30 billion marketing emails sent daily. Most of them are about offers or product collections that will expire at some point. Other transactional emails, such as password reset, should contain a code or link with a short lifespan, and they could therefore also have an <em>Expires:</em> date in it.</p><p>Note that emails are not going to be deleted automatically, they now belong to the recipient. The sender only provided an indication, and the action of deleting them remains the responsibility of the recipient.</p><p>So that’s where we’re at now. <strong>Senders can add this header.</strong></p><p>The next step is for MUA editors (be it a webmail or an app) to see enough of this header, and offer to their users an easy way to regroup these “expired” emails, review and delete them.</p><p>I would also love to see the “Delete” button replace the “Archive” one, or at least both being displayed (come on Gmail, if Outlook does it, you can do it too).</p><p>Then I could write another article, with calculation of the carbon footprint when 1, 10, 20, 50, 75 or 90% of the marketing emails are being deleted immediately?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=26b253aad4a7" width="1" height="1" alt=""><hr><p><a href="https://splio.tech/carbon-footprint-of-decaying-emails-26b253aad4a7">Carbon footprint of decaying emails</a> was originally published in <a href="https://splio.tech/">Splio Product Engineering &amp; Design</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Moving PowerMTA to the Cloud]]></title>
            <link>https://splio.tech/moving-powermta-to-the-cloud-b065cda242e2?source=rss-9dc98e61d32c------2</link>
            <guid isPermaLink="false">https://medium.com/p/b065cda242e2</guid>
            <category><![CDATA[powermta]]></category>
            <category><![CDATA[migration]]></category>
            <category><![CDATA[cloud]]></category>
            <category><![CDATA[email]]></category>
            <category><![CDATA[aws]]></category>
            <dc:creator><![CDATA[Benjamin Billon]]></dc:creator>
            <pubDate>Thu, 19 Oct 2023 07:13:00 GMT</pubDate>
            <atom:updated>2023-10-19T07:13:00.762Z</atom:updated>
            <content:encoded><![CDATA[<p>A few years ago, it was decided at <a href="https://medium.com/u/8dcbe00601ba">Splio</a> that we were going to move from our on-premise datacenters to AWS. I remember it has been suggested to migrate our PowerMTA servers first, because that would be one thing off our list. Guess what: it’s one of the last things we ultimately moved to the cloud.</p><h3>Why, drawbacks, challenges and benefits</h3><p>Well, why move PowerMTA to the cloud wasn’t really a question for us, as weren’t going to move only half our infrastructure and maintain two separated costs for hosting. Although …</p><p>Sending high volumes of emails requires quite a lot of bandwidth, most of our traffic actually. On our on-prem datacenters, in France, we can have Gigabits of IP transit at affordable price. In the cloud, everything is charged, and it ain’t cheap. Also, we’re moving to AWS, which unwillingly helps spammers spam and the network’s reputation therefore isn’t great. What if announcing our IPs via AWS impacted our own reputation negatively? With all of this on table, we wondered about keeping an hybrid infrastructure, with the apps in the cloud and the MTA on-prem.</p><p>As we send email 24/7, we also had to find a way to migrate without disrupting our service.</p><p>And finally, one tangible benefit is the ease to scale to more powerful instances than what we were running, with more recent CPUs. We were already running on virtual machines and were therefore able to conveniently add vCPUs, RAM or storage if necessary, but only up to what was available with our hardware. With ̶P̶o̶r̶t̶2̶5̶ ̶S̶p̶a̶r̶k̶p̶o̶s̶t̶ MessageBird’s unfortunate decision to stop selling server-based licenses (to sell volume-based licenses instead), we couldn’t grow our MTAs horizontally (add more MTAs), and had to grow vertically instead (scale the servers up).</p><h3>We needed a plan</h3><p>Our MTA architecture until then was pretty basic: separated servers, each with different IPs assigned. That’s efficient, and easy to maintain one server without impacting the others.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/557/0*afQdFAK7lNdWa9mP" /></figure><p>However, other maintenances, such as moving IPs from one server to another, or rebooting a server for whatever reason were not seamless.</p><p>We dug up old projects we shelved years ago and imagined a few options to prepare our architecture to the migration. A feature made available since PowerMTA 5.0 was key to this: the support for PROXY protocol for outbound traffic. Although we initially thought about having a proxy both for inbound and outbound, we gave up this idea for a very specific reason: Orange (orange.fr/wanadoo.fr domains) only accepts 3 concurrent connections from a single outbound IP. Each PMTA instance could know it, but they wouldn’t know that the other instances know about it at the same time: several PowerMTA do not make a cluster.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*wODohH3bzNiLanJxj4knNQ.png" /></figure><p>In the above diagram, when sending emails to Orange, the Front Proxy would dispatch to the MTAs available, so several MTAs may handle trafic to Orange using the same outbound public IP. If each MTA is set to allow 3 concurrent connections to Orange per outbound IP, we could end up with 3 * [number of MTAs] concurrent connections, resulting in throttling by Orange.</p><p>Instead, we had to humble down a little bit, and give up the idea of having a automatically load-balanced use of the MTAs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*h4QsswGIRwgEfC9GomAczw.png" /></figure><p>Each submission engine could reach any of the MTAs directly. We just have to make sure we don’t send emails from the same outbound IPs using different MTAs.</p><p>Having a proxy in place was crucial to define the rest of the plan: all the IPs (thousands of them), originally set on each MTA server, will be moved into the proxy, allowing for much more flexibility to migrate the MTAs in the cloud. And as we would have all IPs in the proxy, instead of in several servers, migrating the IPs to the cloud (through the Bring Your Own IP or BYOIP feature) is also going to be easier.</p><p>So at this stage, the plan is defined as follow, with virtual-mta setup looking like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/511/1*EbWrdixV04kqUImoIj5vrQ.png" /><figcaption>Original configuration for a virtual-mta</figcaption></figure><ul><li>Setup the proxy on-prem (HAproxy in our case, in an active-backup setup with keepalived)</li><li>Move IPs from MTAs to proxy &amp; make MTAs use the proxy</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/754/1*40KJX4D5Otss_qfpdwP1yA.png" /><figcaption>Temporary configuration, 10.32.30.30 was the on-prem proxy</figcaption></figure><ul><li>Move MTAs to the cloud, still using the on-prem proxy</li><li>Setup a proxy in the cloud (NLB on AWS)</li><li>Move IP ranges from on-prem to the cloud &amp; make MTAs use the cloud proxy</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QO-AgUPXy8kDeyFMcv-ZCw.png" /><figcaption>Final configuration, using the NLB</figcaption></figure><p>We just had to run through it!</p><p>Note that while in the hybrid phase, when proxies (and IPs) are still on-prem while MTAs are in the cloud hosting, the bandwidth between both network will be charged and can be very expensive if not properly negotiated beforehand.</p><h3>A centralized PowerMTA configuration</h3><p>Another benefit of not having IPs tied up to each PMTA servers is that it allows us to have the exact same configuration for all the PMTA. All IPs can be used by all MTAs, making the migration of clients from on MTA to another totally seamless and transparent. The compromise when giving up the Front Proxy remains: we still have to manage the load of each server another way.</p><p>We also seized the opportunity to use gitlab to version our configuration and have CI (Continuous Integration) pipelines.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kOjtzzX7ocmZs7OHyW3x4w.png" /></figure><p>These pipelines have 4 steps:</p><ul><li><strong>Checkout</strong>, to download on the servers the commit version we want to use</li><li><strong>Deploy</strong>, to move or copy various configuration and other files in the right folders</li><li><strong>Test-config</strong>, to (guess what?) test that the config isn’t faulty. This uses PMTA’s API feature with the <em>http-api-key</em> directive.</li><li><strong>Reload</strong>, to finally apply the changes. Without this step and despite the previous ones, PMTA would still be running as before, unbothered. It also uses the <em>http-api-key</em> directive.</li></ul><h3>Stay alert</h3><p>Even with the best of plan, you should always expect the unexpected, and therefore setup monitoring and alerts at an early stage. Here’s one example why.</p><p>The specs of EC2 instances including CPU, RAM, “Network Performance” and prices are conveniently visible <a href="https://instances.vantage.sh/?region=eu-west-3">on Vantage.sh</a>. I’m using double-quotes because the notion of “Network Performance” doesn’t reflect the bandwidth capacity. If you choose big (8xlarge+) you should be safe, but these are very expensive (and probably overkill for a proxy). If you pick more reasonable instances, the “Network Performance” would most likely give you an “Up to” value. This value indicates the potentially reachable bandwidth, but is not a guarantee. These instances in fact work with a Burst bandwidth (the “Up to”) and a Baseline bandwidth that is burried deep into the EC2 documentation, for a <a href="https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/compute-optimized-instances.html">given category of instance types</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/792/1*zEp5JOVj3YcLBJ-KHsb_KQ.png" /></figure><p>The Baseline bandwidth is guaranteed. In the above example, with a c6in.large instance type, there are 3.125 Gbps available at all time, and in case you need more, you can go up to 25 Gbps … for some time.</p><p>This works with credits: as long as you don’t use the credits, they’ll increase up to a certain point. When you need more than the baseline bandwidth, you’ll start using the available credits, until there’s no credit left anymore, and when that happens, you’re back to the baseline.</p><p>Here’s an example when using a t3.micro instance, with a Baseline bandwidth of 0.064 Gbps, and a Burst bandwidth of 5 Gbps:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/630/1*VzCj249tjud2CMNA-N_bZg.png" /></figure><p>Would you guess at what point(s) there was no credit left, and the Baseline of 64Mbps has been enforced?</p><p>Two takeaways from this:</p><ul><li>pay a very close attention to the instance types to use</li><li>make sure you’re monitoring everything you setup</li></ul><h3>Conclusion</h3><p>All-in-all, having to move our infrastructure to the cloud forced us to modernize our architecture and is giving us more room to grow.</p><p>We also went from managing the PMTA configuration directly on the server with SSH, to allowing more collaborators to suggest changes in it: they can create a merge request with their changes, which could then be reviewed, tested and deployed in a few clicks.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b065cda242e2" width="1" height="1" alt=""><hr><p><a href="https://splio.tech/moving-powermta-to-the-cloud-b065cda242e2">Moving PowerMTA to the Cloud</a> was originally published in <a href="https://splio.tech/">Splio Product Engineering &amp; Design</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Carbon footprint of (marketing) emails]]></title>
            <link>https://splio.tech/carbon-footprint-of-marketing-emails-d7ef1cbd39b9?source=rss-9dc98e61d32c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d7ef1cbd39b9</guid>
            <category><![CDATA[email]]></category>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[carbon-footprint]]></category>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[environment]]></category>
            <dc:creator><![CDATA[Benjamin Billon]]></dc:creator>
            <pubDate>Mon, 02 Oct 2023 16:43:55 GMT</pubDate>
            <atom:updated>2023-12-18T08:52:13.576Z</atom:updated>
            <content:encoded><![CDATA[<p>Last April we at <a href="https://medium.com/u/8dcbe00601ba">Splio</a> published a quick article in our documentation to encourage our clients to use the <em>Expires:</em> header in their emails (see <a href="https://support.splio.com/hc/en-us/articles/10904976785682-Mission-Impossible-Your-marketing-email-will-self-destruct-in-x-days">the article here</a>). I provided a few numbers there, and voluntarily didn’t put more to avoid flooding the intended audience with those. Also to be honest, I didn’t have at the time as many details as I am now sharing here.</p><p>If you’re not familiar with the <em>Expires:</em> SMTP header, well, you can read the aforementioned article, or have a look at <a href="https://datatracker.ietf.org/doc/draft-billon-expires/">the related IETF draft</a>, but in short, it’s about providing emails with an expiration date so the recipient can easily sort through expired or useless emails in their mailbox, and delete them.</p><p>The article you’re reading right now aims at shedding some light at the numbers I found and the calculation I did. I’m not an expert at carbon footprint calculation, and it’s likely that my methodology is questionable. It necessarily contains many flaws such as using worldwide averages, focusing on carbon instead of global impact (this is not a proper life-cycle assessment), maybe not considering “per-capita” figures, or even mixing data sources from different studies. The environmental impact is an awfully complex topic and I had to take some shortcuts to keep the article bearable for most (and also keep my sanity). Also, since April I found more data to use for the calculation.</p><p>Therefore, the results cannot be interpreted as “the truth”. Anyone who got relatively close to a carbon footprint (or better/worse: a life-cycle) assessment could tell you the “real number” doesn’t exist.</p><p>I simply tried to, if not pave, at least strim a way through the email marketing digital carbon footprint tundra , and I hope my research could be helpful to, and improved by others.</p><h3>Why bother, though?</h3><figure><img alt="But why?" src="https://cdn-images-1.medium.com/max/581/1*evsRlXzetp7FdEOK2ynL1Q.gif" /></figure><p>The carbon footprint of emails has been calculated and discussed several times already but not to my entire satisfaction for various reasons, and as most of my days revolve around emails, the topic is important to me.</p><p>However, emails don’t matter. I hear that regularly, and it’s true to some extent. To illustrate how irrelevant emails are to the global environmental footprint, I tried to look at two things:</p><ul><li>The part of Digital in the global footprint was estimated <strong>around 4%</strong> in 2019 (sources: The Shift Project and <a href="http://greenit.fr/">GreenIT.fr</a>), and increasing.</li><li>The part of emails in the digital footprint:</li></ul><figure><img alt="No idea." src="https://cdn-images-1.medium.com/max/327/1*jJ8lYLEkIWGBtFdywKkMcw.png" /></figure><p>That’s what first tickled me.</p><h3>The global footprint of emails</h3><p>According to Radicati, the numbers of emails sent <strong>daily</strong> in 2022 was <strong>333.2 billion emails</strong>[1], worldwide. Based on what the mailbox providers (MBPs) I talked to (Yahoo, Comcast, Orange), 50% to 90% of incoming emails are rejected immediately as spams, and we estimate that 30 billion of emails sent daily are of marketing nature. The person-to-person emails are a thin line in this whole, as are the corporate email communications.</p><p>The Shift Project estimates that 80% of the Internet traffic is related to videos, generating 408MtCO₂e in 2018[2]. Cisco forecasted 82% for 2022, with 12% (or 61.2MtCO₂e) for “Web/Data”[3], in which Emails would fit. I would assume that “web” traffic would account for 90% of this category, leaving email with 6.12MtCO₂e or less.</p><p>Compared to the 40.6BtCO₂ emitted in 2022[4], Email accounted for 0.015% of it.</p><h3>The footprint of sending an email</h3><p><em>Many studies, many different results, and yet neither are (in)correct.</em></p><p>According to <a href="https://www.ademe.fr/en/frontpage/">ADEME</a> in 2011, sending an email with a weight of 1MB has a carbon footprint of 19g of CO₂ equivalent.</p><p>In 2019, <a href="https://theshiftproject.org/en/home/">The Shift Project</a> calculated that an email of 1MB, sent in Europe, would have a footprint of 0.03g of CO₂e.</p><p><a href="https://www.sami.eco/blog/empreinte-carbone-email">Another study</a> from late 2022 settled for 3.3gCO₂e per email of 1MB.</p><p>In the meantime, Mike Berners-Lee, researcher and professor at Lancaster University (and brother of Tim, inventor of the WorldWideWeb) makes a distinction between spam emails being rejected early, and long emails taking time to write and read, for a carbon footprint going from 0.03g to 26g CO₂e per message[5]. He also mentions that spam would only account for 2% of the footprint.</p><p>They all had a proper method to find their results, with the information they had available, but neither are relevant to what we’re interested in in this article, so here’s my contribution:</p><p>If we round up the spam at 300 billion per day, or 109 500B per year, and 98% of 6.12MtCO₂e is emitted by non-spam, then 12 118 billion legit emails are emitting 5 997 600tCO₂e per year, making each email emit <strong>0.5gCO₂e</strong> on average.</p><p>Ok here we are.</p><p>But at least one important variable is obviously missing: the size of the message. And this is just the tip of the iceberg.</p><h3>The footprint of marketing emails</h3><p>While most of the studies I found are considering 1MB for the size (or weight) of the message, this is not relevant for marketing emails.</p><p>Based on my experience, a rough average of the weight of marketing emails is 60KB. That includes the HTML part, the text part, (<a href="https://amp.dev/documentation/guides-and-tutorials/integrate/add-email">the AMP part</a> if you have one) and the envelope. That doesn’t include the pictures, which are hosted somewhere else and not transmitted when the email is sent (the body only contains the URL of the image, to be downloaded when the recipient opens the message).</p><p>If we arbitrarily consider 3.3gCO₂e per email of 1MB (a value somewhere between the min and the max of most studies), then we can estimate that the average marketing email (60KB) has a footprint of 0.198gCO₂e, right? Of course not, that would be too easy. But we’ll talk about it later.</p><p>[30 billion marketing emails per day] * 0.198gCO₂e = 5<strong> 940tCO₂e per day</strong>. Or 2.17MtCO₂e per year.</p><p>Note that at this stage, Berners-Lee estimation of spam accounting for 2% of the footprint doesn’t hold, as: [300B spams] * 0.03gCO₂e = 9 000tCO₂e per day. Or more than 60% of the footprint. Even if we take the average of 0.5gCO₂e per email calculated earlier, spam would account for 37.5% of the footprint, still far from 2%.</p><p>Would that mean that we shouldn’t calculate marketing emails’ carbon footprint this way?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/284/1*ioC9Ar7pgVWMny2AlogvDQ.gif" /><figcaption>Shocked.</figcaption></figure><p>But for the sake of the experiment, let’s continue down this road.</p><p>Based on the Shift Project’s and Cisco’s estimation of the digital footprint being 510MtCO₂e for 2022[2][3], we could consider marketing emails account for 0.42% of the carbon footprint of digital, and <strong>0.0053%</strong> of the footprint of humanity.</p><h3>Why it’s incorrect</h3><p><em>Or “how to calculate the footprint of a given email marketing campaign”.</em></p><p>As subtly hinted earlier, Mike Berners-Lee’s estimations go from 0.03g to 26g CO₂e per message depending on many variables such as, but not limited to: using Wifi or Mobile data, a laptop or a smartphone to write and read the emails, the number of recipients, the time spent crafting the content of the message, the time spent reading the message, the country the sender, the recipients’ servers and the readers are, the number and size of pictures in the message, the location of the hosting of said pictures …</p><p>He’s not the only one to go deep into details, I cannot not mention the tool <a href="https://impactco2.fr/usagenumerique/email">impactco2</a> developed by ADEME (and its <a href="https://github.com/incubateur-ademe/impactco2">source on github</a>), only in French unfortunately.</p><p>Obviously, all these variables just make the result more complicated to digest, or use. We’ll therefore pick a few that seem to be the most relevant for marketing emails, or simply for which we could get data for:</p><p><strong>Size of the message: </strong>the bigger it is, the more bandwidth it would use to be sent, but also the more space it would take to store.</p><p><strong>Number of recipients:</strong> there is not “one” message when sending to several recipients, we multiply the message by the number of recipients, each being stored in each recipient’s mailbox.</p><p><strong>Time spent preparing the design and the campaign (senders side) &amp; time spent reading the message (receivers side):</strong> because a big part of the carbon footprint comes from building the devices to use emails (computers, smartphones …), and the time spent dealing with emails is a prorata.</p><p>Note that it seems odd to include the impact of building the devices in the equation, because nobody buys a laptop <em>just</em>, or even <em>mostly</em> to deal with emails. But as you’ll see, even when the device is used only a fraction of its life for emails, it still isn’t negligible. Depending on how optimized is your campaign, it may even … be the major part of your footprint.</p><p>Then we use (possibly rough) assumptions:</p><ul><li>A laptop’s building carbon footprint equivalent is <strong>331kgCO₂e</strong>[6]</li><li>A laptop consumes <strong>5.91Wh</strong> per hour when used actively[7]</li><li>Transmitting 1GB over the network emits <strong>4.43gCO₂e</strong>[8]</li><li>Hosting 1GB in the cloud for 10 years emits <strong>25.3gCO₂e</strong>[9]</li><li>The average worldwide carbon intensity factor is <strong>460gCO₂/kWh</strong>[10] (it’s 85gCO₂/kWh in France, 367gCO₂/kWh in the US, 503gCO₂/kWh in Australia …)</li><li><strong>90%</strong> of the emails received are quickly deleted (this is a very, very high estimation)</li></ul><p>Finally, some parts of the equation are arbitrarily chosen:</p><ul><li>Designs weigh <strong>60KB</strong>, with <strong>5MB</strong> of pictures</li><li>There are <strong>150 000</strong> recipients</li><li><strong>20%</strong> of recipients will open the message, spending <strong>5 seconds</strong> on it</li><li>Everyone is using a laptop, which would last <strong>4 years</strong>. The calculation could be more complex if we consider openers can have various kind of devices (smartphones, tablets, etc.) so let’s keep it simple for the example.</li><li>Preparing a sending takes <strong>3h</strong> (creation of the design, setup of the campaign, etc.)</li></ul><p>Here’s the summary:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LlmcVK6JT6V7okTggJbGlw.png" /></figure><p>Leading us to estimate a carbon footprint of <strong>1270.35gCO₂e</strong> for this campaign. Here are a few more results, with a few changes in the variables:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/661/1*n-Wpzl7ZsOb7bS3_ksd5Aw.png" /></figure><p>Feel free to use <a href="https://onedrive.live.com/view.aspx?resid=9C227735736E803F!24775&amp;ithint=file%2cxlsx&amp;wdo=2&amp;authkey=!AM0t-SrcQKhZJrU">this little calculator</a> where you can easily change the variables for what makes more sense to your case. If you improve it, please let me know 😉</p><p>But again, and I can’t stress this enough: I hope this article helps you get a grasp at what your email marketing carbon footprint may roughly be, and that it could give you a reference to improve. However, it doesn’t replace a proper carbon footprint assessment, made with the help of experts.</p><h3>Advices to reduce the footprint</h3><ul><li>Extend your devices life. My old Thinkpad T510i recently died after 13 years. Who could beat that? (I’m even planning on replacing the fan to keep it even longer)</li><li>Limit your images’ size, choose the right format, and more generally, follow the <a href="https://www.badsender.com/en/guides/eco-design-email/">eco-design principles</a></li><li>Purge your designs from hosting after some time. Nobody <em>needs</em> to re-read a marketing email of 2 years ago and see all pictures as they were.</li><li>Add an <a href="https://datatracker.ietf.org/doc/draft-billon-expires/">Expire header</a></li><li>Consider climate shadow: if the email you send is about flying far away (high carbon footprint), or is encouraging to walk or ride a bike, the impact of reaching out to the recipients could be dramatically different (although awfully complicated to calculate).</li></ul><h3>Next article on this topic</h3><p><a href="https://medium.com/@benjamin.billon/carbon-footprint-of-decaying-emails-26b253aad4a7">https://medium.com/@benjamin.billon/carbon-footprint-of-decaying-emails-26b253aad4a7</a></p><h3>Sources</h3><ul><li>[1] “Total Worldwide Emails Sent/Received Per Day (B) in 2022: 333.2” <a href="https://www.radicati.com/?p=17936">https://www.radicati.com/?p=17936</a></li><li>[2] “Online video: 60% of online data flows […] this generates 306 MtCO2” (p. 4) <a href="https://theshiftproject.org/wp-content/uploads/2019/07/2019-02.pdf">https://theshiftproject.org/wp-content/uploads/2019/07/2019-02.pdf</a></li><li>[3] Cisco Visual Networking Index: Forecast and Trends, 2017–2022, Trend 4: Applications traffic growth, <a href="https://www.cisco.com/c/en/us/solutions/executive-perspectives/annual-internet-report/index.html">https://www.cisco.com/c/en/us/solutions/executive-perspectives/annual-internet-report/index.html</a></li><li>[4] “Total CO2 emissions remain high at a projected 40.6 billion tonnes of CO2 (GtCO2) in 2022” <a href="https://globalcarbonbudget.org/wp-content/uploads/Key-messages.pdf">https://globalcarbonbudget.org/wp-content/uploads/Key-messages.pdf</a></li><li>[5] Mike Berners-Lee in <em>The Carbon Footprint of Everything</em>, 2020, ISBN 978–1788163811</li><li>[6] “the average CO2 emissions during production of a new laptop is 331kgs” <a href="https://circularcomputing.com/news/carbon-footprint-laptop/">https://circularcomputing.com/news/carbon-footprint-laptop/</a></li><li>[7] “On average, laptops use […] 5.91W in Short Idle mode.” <a href="https://ecocostsavings.com/how-many-watts-does-a-laptop-use/">https://ecocostsavings.com/how-many-watts-does-a-laptop-use/</a></li><li>[8] “Footprint of 1GB of data on wired network: 0,00443 kg CO2 eq.” from the Negaoctet database</li><li>[9] “Footprint of 1GB of data stored on Cloud hosting (through wired network) for 10 years: 0,0253 kg CO2 eq.” from the Negaoctet database</li><li>[10] “The global average carbon intensity of electricity generation is currently 460 grammes of carbon dioxide per kilowatt‐hour (g CO2/kWh)” (p.45) <a href="https://iea.blob.core.windows.net/assets/830fe099-5530-48f2-a7c1-11f35d510983/WorldEnergyOutlook2022.pdf">https://iea.blob.core.windows.net/assets/830fe099-5530-48f2-a7c1-11f35d510983/WorldEnergyOutlook2022.pdf</a></li><li><a href="https://ademe.typepad.fr/files/acv_ntic_synthese_courrier_electronique.pdf">https://ademe.typepad.fr/files/acv_ntic_synthese_courrier_electronique.pdf</a></li><li><a href="https://presse.ademe.fr/files/acv_ntic_synthese_resultats.pdf">https://presse.ademe.fr/files/acv_ntic_synthese_resultats.pdf</a></li><li><a href="https://www.badsender.com/en/2023/04/07/carbon-footprint-emailing-mediapart/">https://www.badsender.com/en/2023/04/07/carbon-footprint-emailing-mediapart/</a></li><li><a href="https://www.wholegraindigital.com/blog/website-energy-consumption/">https://www.wholegraindigital.com/blog/website-energy-consumption/</a></li><li><a href="https://www.greenit.fr/wp-content/uploads/2021/12/EU-Study-LCA-7-DEC-EN.pdf#page=42">https://www.greenit.fr/wp-content/uploads/2021/12/EU-Study-LCA-7-DEC-EN.pdf#page=42</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7ef1cbd39b9" width="1" height="1" alt=""><hr><p><a href="https://splio.tech/carbon-footprint-of-marketing-emails-d7ef1cbd39b9">Carbon footprint of (marketing) emails</a> was originally published in <a href="https://splio.tech/">Splio Product Engineering &amp; Design</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>