It is a truth universally acknowledged that a software product in possession of a good value proposition must be used by its creator. Fair enough, but why do tech companies insist on making such a big deal about it?
In this post I will focus on the general question of what dogfooding in tech accomplishes today. Why do marketers love dogfooding; what does dogfooding signal to users; when is dogfooding persuasive; do developers care about dogfooding; and, possibly most importantly, how is dogfooding even defined? Answering these questions will certainly interest practitioners and engineering leadership forced to endure declarations of value based on dogfooding.
That is, of course, unless the vendor has transformed their dogfood into champagne.
Dogfood > Champagne (fight me)
When I worked at NewRelic we used to call it "Drinking our own Champagne" because we don't make dogfood, we make champagne. 😛
— Amjith (@amjithr) February 12, 2021
Whether a creator-as-user prefers to characterize their process as “eating our own dogfood” or “drinking our own champagne” depends on brand preference: preferences that are useful to unpack. Although they are considered synonyms meaning “the practice of using one’s own products or services,” each resonates very differently, and can even draw censure when misapplied.
As a skeptic of marketing goo, I actually prefer the term dogfood over champagne owing to the disgust it suggests. Bear with me on this one. Dogfood is toxic (sometimes literally). Since the catastrophic train derailment in East Palestine, Ohio, in my mind the term invokes photo ops featuring Ohio’s Governor Mike DeWine, Representative Bill Johnson, and Michael Regan, head of the Environmental Protection Agency (EPA), holding glasses and red solo cups filled with the local tap water in order to demonstrate its safety by consuming it themselves. The method of persuasion these Ohio politicians use is well studied. As Chip and Dan Heath argue in Made to Stick (2007), nothing provokes greater emotional effect or memorability than experience: “what’s more likely to stick with someone: hearing about someone who fooled someone else, or being fooled yourself?” (163). By documenting a group of educated and powerful Ohio men consuming East Palestine’s tap water, the hoi polloi knows it must be safe.
DRINKING EAST PALESTINE WATER: with EPA Administrator Michael Regan & Ohio Governor Mike DeWine (2/21) pic.twitter.com/8fnYYkKZgQ
— Howard Mortman (@HowardMortman) February 22, 2023
Of course, the National Cancer Institute states it is “Repeated exposure to vinyl chloride,” and not a single glass of tap water or spending a few days in a polluted area that “is associated with increased risk of liver, brain, and lung cancer as well as leukemia lymphoma.” DeWine and his associates’ political theater is effectively meaningless to the health of those involved because they don’t actually live and work in the community. These facts do nothing to diminish their tactic’s effectiveness. Audiences on social media and the nightly news don’t have the patience to follow East Palestine’s residents for the years needed to track the consequences of this chemical spill. Despite its questionable truth claim, dogfooding’s parallel testimonial form aligns well with tried-and-true means for convincing skeptics.
When human beings choose to eat kibble intended for animals we know it must be not only safe but also tasty (according to Regan: “It’s pretty good water”). Drinking champagne is not intended to come from a comparable place of skepticism and disgust. Champagne is a luxury, and it doesn’t require consumers to stick out their necks. But following the post-Covid economic downturn of 2022, skepticism and disgust is often precisely what adopting a new product signifies.
At RedMonk we have noticed that many of the B2B companies we speak with are endeavoring to tighten their belts, which makes dogfood have a different degree of urgency in today’s cost cutting climate. Buyers are skeptical about experimenting with new products requiring significant overhead. They are wary of undertaking the risk of non-urgent digital transformation, no matter how seductive the features, because they might be poisonous.
The purpose of this post is not to engage with the particular boondoggle of champagne versus dogfood. But what the champagne/ dogfood debate does well is to set up the larger conversation around what exactly it is (rhetorically, procedurally, culturally) that dogfooding accomplishes.
Dogfooding is not CX
Dogfooding’s merits have been questioned since its inception in the 1970s or 80s. We’ve all heard stories of employee backlash when companies require their employees to use their products. Microsoft is notorious for implementing these internal policies, and have received pushback for dogfooding their Zune, Microsoft Exchange, the Windows NT operating system, and Microsoft Teams. But Microsoft is far from an outlier. Most large tech companies, and nearly every FAANG, engage in some form of dogfooding. Last year, for instance, Meta drew criticism for requiring employees to use Horizon Worlds.
this is rough. fb reduced to forcing employees to spend time in its flagship metaverse app https://t.co/EG83xZOUkW pic.twitter.com/vcCgFwuKeP
— Max Chafkin (@chafkin) October 7, 2022
At a higher level, complaints against dogfooding tend to stem from its melding of incompatible roles: product designers somehow must also serve as typical users. This amalgam is dubious to say the least. Businesses possess asymmetric information about the roadmap, backlog, bugs, ideal use cases, and ways to fit square pegs into round holes that a customer will never have. In 2013, Alan Downie wrote “The dangers of eating dogfood” to address his concerns about the impossibility of a vendor-customer:
the fact is, you’ll never be your own customer. You may use your product in a way that is similar to your customers, but that’s where the similarity ends.
By professing to serve as both creator and consumer, companies that dogfood corrupt the user pipeline. Whereas the actual customer journey allows companies to identify and address pain points, by sidelining the onboarding process product designers lose out on the necessary discovery and frustration that occurs for non-experts. In order to authentically test a product and assess the customer experience, companies need to recruit users with zero preconceptions and disinterested motives.
So what gives? Are software companies in the era of AI and cloud still dogfooding for the same reasons they always have? The answer seems to be a resounding “Yes.”
Despite its well known problems, marketing for cloud products is awash in testimonial-style claims that vendors are confident in the strength of their offerings based on the evidence of their own teams. According to a 2018 CRN article, Jarrod Levitan, who worked at the time as the chief growth officer at Onica, an AWS partner: “Amazon is famous for eating their own dog food. Developing an application for themselves then productizing it for the world to use on their AWS Cloud.” This summer, The New Stack published an interview with Karishma Irani, head of product at LaunchDarkly, titled “What LaunchDarkly Learned from ‘Eating Its Own Dog Food’” as part of their Makers series. According to HQO’s 2022 “Drinking Our Own Champagne” blog post, “This isn’t the first time we’ve used our technology at our own offices, but this latest development means that we’re now using it in a way that allows us [to] better meet the needs of our growing workforce.” From where I’m standing, tech’s dogfood mound shows no signs of abating.
Developer Experience
One of the things dogfooding seems to do well is beta testing. It is useful for ensuring security and quality, and developers and marketers agree that this is worth bragging about.
Cloudflare leans particularly heavily on the idea of dogfooding in the context of developer experience. Omer Yoachimik, a Senior Product Manager at Cloudflare specializing in DDoS Protection & Security Reporting, positions dogfooding as essential to security in his post “Who DDoS’d Austin?” (2019). Yoachimik describes how SREs at Cloudflare used a red team DDoS attack to test “our newest cloud service, Magic Transit, which both protects and accelerates our customers’ entire network infrastructure.” Cloudflare also joins a number of other cloud players in using dogfooding as a means of product development. In a 2020 post for the Cloudflare blog, Evan Johnson, then a Senior Director of Security Engineering and Product Security at Cloudflare, accessed anxieties about remote work by titling the post announcing their Cloudflare for Teams program’s launch “Dogfooding from Home.” Johnson assures potential buyers:
But It’s Not Just for Us
With the massive transition to a remote work model for many organizations, Cloudflare Access can make you more confident in the security of your internal applications — while also driving increased productivity in your remote employees.
Historically, companies with versioned software dogfood unstable versions prior to official release. It is important for the latest versions to be tried and tested internally to iron out all the wrinkles. The software engineers at IBM Red Hat take this approach. Illustratively, in a 2022 Hacker News conversation touting Fedora’s benefits, contributor hotpotamus wrote:
At Red Hat, they give you a laptop with RHEL on it, then the other engineers tell you to blow it away and install Fedora, so there’s a lot dogfooding going on. A buddy who works there got the latest Dell XPS (for more or less personal use) and was able to get the sound card driver merged into the next kernel release by knowing who to ask. I’m more of an Ubuntu user going back for over a decade, but he’s got me looking at Fedora.
I like this comment because it points to the overlap between dogfooding and practitioner currency. There is a cache among developers associated with being an early user of a popular technology, whether that’s an open source project or else an unstable beta version that isn’t ready for general availability.
In this way dogfooding overlaps with the pipeline to open source as many companies build internal tools that they then decide to license and release as FOSS. Of course, in this case bragging about these to external users is secondary and intended to support the primary, revenue producing product.
Upskilling
An additional usage of dogfooding overlaps with education and upskilling. Insisting on internal usage of technical tools can be especially difficult when not every person on the team is technical. Pierce Ujjainwalla, Co-Founder & CEO of Knak, speaks about this challenge in his “Drinking Our Own Champagne” (2022):
Beyond using Knak as a key tool in our martech stack, we’ve taken things a step further to get everyone comfortable with the tool. Four times a year, we get everyone together (with teams gathered on Zoom and in person) and we spend half a day using our own product to build emails and landing pages for a mock campaign.
Because this type of dogfooding/champagneing is so important to Knack’s business strategy, leadership has made the decision to invest time to ensure that all team members understand how to use the product in-and-out. Interestingly, Ujjainwalla’s definition of champagne strays perhaps furthest from the definition used by developers. Indeed, this expanded meaning pushes dogfooding’s connotations into the realm of team upskilling.
The AI boom has crystalized an overlap between dogfooding and the skills gap. In a space where everyone is a pioneer, dogfooding is assisting companies with establishing conventions and best practices. At a recent AWS partner panel on the subject of AI, and specifically AI use cases, Accenture shared how they use their services internally to upskill their teams. According to Arnab Charkroborty, a Senior Managing Director focused on Data and AI at Accenture:
Accenture should be its own best credential.
We are all learning together with AI. By iterating and trying out their own AI services internally Accenture is better positioned to enable customer success. The reason I like this quote is that dogfooding is so often discussed from a product perspective rather than a services perspective. Accenture’s use of AI to develop their own workforce is a pursuasive example of the latter.
Wrapping Up
Was this post just an opportunity to post classic puppy gifs? Maybe.
Although the story of dogfooding’s beginnings is contested, it continues to be a popular testimonial-style appeal to audiences and is therefore worth assessing.
In a sense, dogfooding represents a lazy marketing approach. Instead of rounding up external customers to vouch for a product or service’s virtues, these companies select potential reviewers by looking down their own line of cubicles. Companies are certain to get positive soundbites from these users, and there is no incentive to announce the agony that may have preceded favorable experiences. Moreover, by virtue of owning a product, a company can’t truly be in the position of the customer.
I have argued that dogfooding may be misleading and unhelpful when it comes to establishing product-market fit and simulating the customer journey, but it still has merit. A lot of this has to do with dogfooding’s definitional squishiness. Dogfooding should probably be abandoned in the domain of CX, but in terms of grassroots feature development it has been, and in all likelihood will continue to be, consistently effective.
Disclaimer: IBM, LaunchDarkly, Microsoft, Red Hat, AWS, and Cloudflare are RedMonk clients.
No Comments