Software engineers hang out on Twitter.

I know this anecdotally and by feel because I spent most of my career as a software engineer and I’ve had a Twitter account for more than a decade. But you can confirm this somewhat more objectively as well.

For instance, a Google search for “programmers to follow on Twitter” does get actual search volume (per Keywords Everywhere). When I plugged in other vocations, like lawyer, doctor, and teacher, no search volume registered.

While my initial Twitter presence was largely to interact professionally and promote a hobby blog, about seven years ago I went into business for myself as a consultant. So social media started to become a lead generation channel for services and any products I offered. Twitter was no exception.

I dutifully promoted content and offerings on Twitter and LinkedIn because that’s just “Marketing Your Business 101.” Best practices and all that. I imagine that a lot of startup founders and indies, like me, do this by rote.

Asking the Question: Should Brands Market to Developers on Twitter?

But for reasons I won’t bore you with here, I wound up shifting from writing software to starting a developer marketing business about four years ago. And with this business, we take a kind of Moneyball/Freakonomics-style approach to content campaigns. We don’t take on work unless we can model out, at least in the abstract, ROI on the content we create.

That recently brought me full circle to ask what I should have asked all those years ago: is Twitter a worthwhile marketing channel for reaching engineers?

Common sense and anecdotal experience say yes. But nagging doubts have been creeping in as I study successful influencers’ use of the platform.

It’s not that I doubt that they reach people and build relationships. There’s no doubt about that. It’s more that I think their love of the Twitter game causes them to lose sight of how much labor (and thus cost) they sink into the platform to get those results.

And that’s fine for influencers in the space. But it can translate to an attractive nuisance for brands. And these days, I’m in the business of helping brands’ marketing departments avoid wasting money on attractive nuisances.

So let’s take a data-driven look at Twitter, using the data that I have available to me: my tweets and my followers. All of this content and people skew heavily programmer.

Setting the Stage: Caveats and Methodology

Before diving into the research that I did, I’d like to clarify a few things.

The Degree of Rigor Involved Here

First, please understand that I’m not about to lay out a PhD thesis for defense here. Between my years as a salaried software engineer and starting a developer marketing business, I earned my living as a specialized management consultant in the software world. And one of the things I brought to bear to help clients was drawn from the excellent book How to Measure Anything.

Anything can be measured. If a thing can be observed in any way at all, it lends it self to some type of measurement method. No matter how “fuzzy” the measurement is, it’s still a measurement if it tells you more than you knew before. And those very things most likely to be seen as immeasurable are, virtually always, solved by relatively simple measurement methods.

So the purpose of gathering this data and making inferences isn’t to have an absolutely airtight case. Rather, it’s to take our conception of this channel from, “I dunno, other people do it, so let’s try it” to having hypotheses against which to experiment.

Looking Only At Building and Marketing to a Twitter Following

The way that most developer influencers use Twitter is fairly straightforward. They build an audience and then speak to that audience. Any virality or hashtag success along the way is a bonus.

What I’m going to look at here today is data that assumes developer marketing brands are interested in recreating developer influencer success in this fashion. In other words, I’m going to look at cost and return on something like a dev tools brand building a following and then sharing content with that following.

This means that I’m not looking at sponsored tweets, gaming the algorithm, purchasing followers, or targeting hashtags. Those are all potentially viable approaches and could change the holistic ROI of Twitter as a platform. But today, I just want to address the ‘traditional’ Twitter campaign of building a following and sharing content with those followers.

Methodology: How to Gather Twitter Data

The last bit of housekeeping that I’ll offer is the methodology of how one extracts data from Twitter and what data I extracted.

To get data out of Twitter, you can actually use native Twitter functionality to download it. They’ll send you a zipped archive within 24 hours. If you don’t fancy the wait or want more granular control, Twitter also offers an API.

In my case, however, I did neither of these things. Hit Subscribe actually has a relationship with Panoply, who can instantly ingest data from a wide variety of sources into a relational format.

This saved me the headache of wrangling APIs and exports. Instead, it just put everything into a relational structure. So I jumped right to the good stuff, like this:

This gave me access to all of the data from my Twitter account. From there, I used information about my followers (largely programmers) and tweets (historically, largely programming-oriented).

So this is the data that I will use for the analysis. It goes almost without saying that, though representative, this is a small sample size. To build our initial experimentation framework into a robust model, we’ll need to draw a lot more data as we go forward.

But I think what I have here is interesting for the sake of conversation.

Looking at Some Raw Numbers and Findings

With that out of the way, let’s dive into some data. When we’re looking to model ROI for a content channel, we need to understand two main variables:

1. Overall cost
2. Outcomes (revenue) generated using that cost

In the digital marketing world, tying activities like tweeting to top-line revenue becomes tricky because of the butterfly-effect-like explosion of variables. So we can take a handy shortcut and reason in a couple of shorthand ways:

1. Think about a metric like impressions or clicks to a site and assume a placeholder valuation, like dollar value of a site visitor or impression.
2. Compare the cost per impression or click to what someone would have to pay for advertising. Thus we can think of something as a “good” channel if you can earn impressions and clicks for less cost than you’d incur with the average ad campaign.

Because the first option there is so widely variable from business to business, I’m going to stick with the second one. Let’s back of the napkin the cost of impressions, as compared to advertising platforms.

Our campaign to market through Twitter thus has two flavors of cost, if you will:

1. The cost to build a following
2. The cost to create content marketing to that following

Let’s now use my data to start reasoning about this.

Approximating Labor With Tweets Per Follower

The most straightforward way I can think to look at the cost of building a following is to assume that creating content on the platform is what results in that following. In other words, what is the ratio of tweets to followers?

Looking at my (at the time of writing) 3,892 followers, I added a column for each that calculated their ratio of tweets to followers. Then I took the average of that column. That value was 15.4. In other words, the mean number of tweets per follower among my followers was 15.4.

Now this reasoning is obviously imperfect. There are clearly other ways people attract followers than by tweeting. Famously, James Comey switched his Twitter handle, became verified, and instantly attracted hundreds of thousands of followers.

But we have neither the luxury of understanding off-platform sources of followers, nor do we have access to data that would inform it. So I’ll make things less noisy by using the median rather than the mean. This lets us eliminate the outliers of both Comeys and people who are awful at Twitter, shrieking into the void of their zero followers with many, many tweets.

Indeed, using the median flattens the data considerably. The median tweets-to-followers ratio among my followers is a more reasonable 4.1 tweets per follower.

Follower Attraction Is a Positive Feedback Loop

As I was looking at my data here, I started to wonder whether Twitter accounts became more efficient at attracting followers as they grew in size. So I segmented the Twitter users into four non-mutually exclusive groups:

1. People with less than 1,000 followers.
2. Those with more than 1,000 followers.
3. Those with more than 10,000 followers.
4. And finally, those with more than 100,000 followers.

Taking the median of these smaller segments revealed a clear trend.

This appeared to confirm my sneaking suspicion. The path to 100K Twitter followers isn’t simply banging out 400K tweets. As followings grow, attracting incremental followers—for reasons requiring more study to ascertain—becomes easier. Some possible explanations that come to mind include fame outside of the platform and the social proof of a large following attracting followers more easily.

For our purposes of the outline of an experimental framework, though, this is enough. We can see that a positive feedback loop exists on the platform.

Brands Are More Efficient Than Individuals

Next up, I wanted to distinguish the data for brands, specifically. Truth be told, my assumption was that brands would need to work harder and tweet more to attract followers. After all, brands are often self-promotional and fairly boring in their Twitter presence.

But, interestingly, that wasn’t the case.

I went through and manually identified all brands among my followers. I then ran them through the same analysis and saw the same trend. The only difference was that brands appear to be more than twice as “efficient.”

Overall, the ratio of tweets to followers for brands was about 1.3, and here is the trend for the follower buckets.

Now we’re starting to get to a point where we can add in some cost projections for building a brand following. We can see how much content creation on the platform has corresponded with what kind of following size. So we need to project out the cost of creating that content.

But before we do that, there’s one last piece of raw data I’d like to gather. This will tell us essentially how and when to start “being promotional.”

Tweets Without Links Engage Way Better Than Tweets With Links

Over the years, one of my main use cases for Twitter and social media tended to be promoting content I had created in other channels. This generally took the form of a “proper” post, wherein I didn’t just dump a link and say, “here.” Instead, I took time to write a blurb teaser about the content, included some hashtags, and linked to the content.

I wanted to see how this content stacked up against simply creating content for the platform with no links.

Did looking to call people to other content or offers put a drag on engagement?

To approximate this on the back of our thematic napkin, I looked at likes and retweets as the tells for engagement. I would have loved to include and perhaps even weighed replies to the tweets, but I did not have this data.

In this case, my calculations were simple. I looked at the overall engagement, as measured by retweets plus favorites per tweet. And then I segmented my tweets by whether they included a link or not.

• My average tweet received 4.1 engagements.
• My average tweet with a link in it received 3.7 engagements.
• And my average tweet without a link in it received 9.54 engagements.

In this case, I did want to stick with the mean, rather than using the median. I’m not interested in smoothing out outliers because the outliers represent mass engagement, which is, after all, what we’re after.

Building a Campaign Cost Model

With this data, we now have enough for the beginnings of a cost and ROI model. We can see what it takes to build a following, which is the main hurdle. From there, we’re talking about the cost and efficacy of creating promotional content.

Cost of Producing Twitter Content

First, let’s make a ballpark assumption for the cost of producing a tweet. I’m going to call it 15 minutes to produce the average tweet and estimate the labor cost of doing this at $25 per hour.

My reasoning here is that while, sure, you bang out some tweet replies like “thanks!” in a few seconds, you put a good bit more time into promotional and other strategic pieces of content. So average it out to 15 minutes.

And as for the cost, sure, anyone could tweet and you probably could get people to do it for minimum wage. But presumably you want some level of marketing savvy here. You’re also decently likely to outsource this at times—and for a higher cost than having a full-time employee do it.

But when building our models, we leave these figures as variables for the sake of discussion. That way, if someone says, “Pff, $25 is ridiculous—we pay $10,” we can simply plug $10 in instead and look at the projected impact on ROI.

The Cost of Building a Following

With that in mind, what is the cost of building a following?

Well, based on the figures above for brands, you’ll tweet about 2,000 times on the way to building 1,000 followers, and then about 9,000 more times on the way to building that to 10,000 followers. That’s a total of 11,000 tweets.

Using our cost numbers from above, 11,000 tweets means 2,750 labor hours, at a cost of $68,750. And that feels like it squares. If we assume that’s two years of an entry-level social media specialist’s salary, that means we’d expect an entry-level marketer to be able to build a brand’s following to 10,000 people over the course of a couple of years.

It’s also worth mentioning that I’m considering follower-building disjoint from promoting content and linking people off of the platform. In other words, you’re not going to build a following with nothing but self-promotional tweets. You have to roll up your sleeves and create platform-specific content to build this following (or become director of the FBI).

Modeling ROI for the Campaign

How do we tie this all together for the sake of starting to model ROI?

Well, the last missing piece of the puzzle is the actual click-through rate (CTR), which lets us talk in terms of cost per click (CPC). You can find different figures by googling around, but let’s go with the one from HubSpot, of 1.64%. (It’d be interesting to unpack the “why” of CTR decreasing as a function of follower count, but given that I don’t understand the reasoning behind that, I’ll just use their broad average.)

Let’s say that we want to build a following of 10,000 people and then promote content to that following. Here’s our cost and return structure.

• Building that following costs $68,750.
• Once you start promoting content, you pay about $6 per 10K impressions and 164 clicks. That’s a $0.60 cost per mille/thousand (CPM), and a $0.04 CPC.

So once you’ve sunk almost $70K into the platform, you earn extremely good returns. This article pegs a typical CPM for advertising on Google at $2.80, and your Twitter following is likely to be way better segmented than Google searchers. And this article puts the CPC in tech at $3.80 for search ads and $0.51 for display ads.

But obviously we need to factor in that investment in the platform when looking at it for marketing.

Let’s do that by assuming that, after you’ve built your following, you create 10, 100, 1,000, and 10,000 promotional tweets. Further, let’s assume that you should create about three non-promotional tweets for every promotional tweet to avoid losing followers. Here’s what that looks like, in terms of CPM and CPC.

How to Interpret the Values and Model

Throughout this post, I’ve talked about creating campaign models through a general lens of reducing uncertainty. That’s been fairly abstract so far, but you can see how we approach this more concretely through the spreadsheet screenshot above.

Anything requiring more research or data becomes a variable in the model. We can then add confidence factors to the variables (or eventually consider them constants) as we go along.

For instance, if you wanted to argue that you could build 10K followers with way fewer tweets, we could just adjust the “following tweets” cell accordingly and observe the effect on ROI. Same for something like “target CPC” and “target CPM” where we’re taking figures from the lower end of the range and thus creating pessimistic ROI projections. If those costs through Google are actually substantially higher through your niche, relative ROI to Google advertising becomes much rosier.

Contrarily, this model currently assumes that 100% of your followers see each tweet, which is certainly too high a percentage. People’s sporadic consumption of Twitter and the algorithm itself likely reduce impressions by a fairly significant factor that will require more research to properly model.

And this entire model currently only looks at ROI against comparable advertising spend. As we flesh this out, we’ll also look at the value of Twitter itself serving as a lead qualification platform and brand awareness. And of course, we’ll look at larger and more representative data sets than “Erik’s followers.”

But even this early model dramatically reduces ambiguity and presents a fairly clear picture.

All in or All Out: Don’t Go Through the Motions

At Hit Subscribe, most of the campaigns that we help clients with are organic traffic campaigns. And the reason for this is that organic traffic to your site is just about the most blue-chip content marketing play there is, from an ROI perspective. But it’s also a long play that requires significant time and investment before you start yielding (generally prodigious) returns.

And it turns out developer marketing Twitter is actually pretty similar.

Even if I ratchet the cost in the model down to $1 per tweet, we don’t see positive returns without significant promotional investment.

And I’d argue that this is too rosy because if you cheap out on tweet creation and just put any old thing into Hootsuite, I’d argue that the number of tweets you’ll need to get to 10K followers will spike to well over 11K. There doesn’t appear to be free, or even cheap, lunch here.

We’re going to keep building and refining this model, of course. But based on what I’ve got here alone, I’d say that developer marketing brands have two clear, disjoint options for a Twitter strategy:

1. Optimize for rock bottom cost and have a perfunctory Twitter presence just so that you’re not completely absent from the channel.
2. Treat Twitter as a legitimate lead generation channel, but understand that the investment in the platform will have to be substantial before it yields a return.

And I think the real, qualitative lesson here is that building this as a channel not only takes time and substantial investment but also talent. Influencer engineers on the channel, often through a labor of love, pour tons of effort into it, and, in so doing, they get good at it.

(I am not one of these — I’m pretty bad at Twitter, personally.)

So if you want to recreate their success, it’ll take time, money, talent, and someone willing and ready to treat it as a first-class content creation platform, not just as a repository for announcements and promotion.

Appendix: A List of Tools

If you’re interested in the various tools that I used in this modeling exercise, here they are:

• The browser plugin, keywords everywhere.
• Downloading your Twitter data.
• Panoply for ingesting data.
• DBeaver, the SQL workbench I screenshotted.
• Google sheets for the calculations and graphs.
• If you’re interested in the calculations that I screenshotted in the model, please reach out. I’m happy to share a copy/template version of it, but I’m not going to go to the bother of making it public unless folks specifically ask.