Case Studies

Economies of Scale

By Kenny Pyatt, Founder & CEO of DevOcho

Note: This case study was not written by AI, it was written by me.

In this case study we are going to explore two themes. 1) When is custom software the better option than buying a SaaS offering and 2) when it is actually cheaper to build custom. In the spirit of modern day social media posts, “I think the results will surprise you for the second one”.

1. First, let’s discuss when it is better to build custom.
2. When is custom software the “better” choice?

Better is typically easy to identify. In more than 20 years of building software, I’ve often seen the case that it is 100% better to build custom when a business has unique processes in their industry. They do something that truly differentiates their company from their competitors. Normally the existing SaaS software doesn’t perfectly fit this unique process so they have to make a choice. Change the process to be like everyone else or build something that automates exactly what they are doing to bring efficiency, scale, and improve the customer experience. In this case, it’s almost always better to build than standardize. You have something that is truly unique and better serves your customers. Why would you change to match your competitors' inferior service?

Custom software is also beneficial when companies are growing insanely fast. Being able to build the software iteratively provides a huge benefit of helping the company organize to help them keep up with the growth trajectory as they scale. The trick here is to know the right timing. Automate too soon and the processes are changing too frequently to gain benefit, automate too late and you’ve missed the window and the growth will be slowed while you try to catch up. There is no perfect rule here, but I have found that bringing in super smart engineers often helps you organize that amazing process earlier which is often the major benefit of hiring an agency in the first place. Smart people asking the right questions will help point out wasteful steps, and do a good job of simplifying the complex.

I have a great example of this from my past. I used to work for the largest home restoration company in the USA. They held an annual convention where thousands of franchisees and their key employees would attend. They had some very unique and specific requirements for attending (you had to be a current employee or owner, your franchise had to be current on royalties, etc.) as well as some very specific financial incentives (they would return a portion of your royalties for attending). There were more but these are public knowledge so I feel comfortable sharing them. There was nothing off-the-shelf that could have accommodated those specific requirements (or the many others). We built custom software with our in-house team and then used an offshore team to maintain it. It’s a unique advantage and it provides an amazing cost savings for the company every year. They automated their annual conventions via custom software and it works beautifully.

When is SaaS cheaper?

Better is easily explained, but normally custom is not cheaper, right? I thought custom software was always more expensive, right?!? Well it’s a bit nuanced. I have two thoughts here:

1. When it’s cheaper simply due to scale (with an AI advantage)
2. When it’s cheaper because it is revenue generating or cost saving

Let’s start with the scale argument.

I recently ran across an anonymous poster on Reddit that said his company was paying $1.9 million annually for 5,000 users to use a popular chat application. $1.9 million, every year, to let employees chat! A quick Google search showed that Slack’s annual revenue exceeded $1.5 billion dollars. Amazing.

In this case they are paying for the “enterprise features” which include the ability to manage users, protect company secrets, and audit communications to meet compliance requirements. They can’t use WhatsApp because Corporate can’t see what is being said and they can’t lock people out of the conversation. In some industries the ability to monitor communications is actually required by law (think investment and finance). So Slack has been able to build a billion dollar business on the simple idea that companies need chat and they need the enterprise features, so they will pay for them.

Also, building a chat application internally doesn’t really offer much in the way of a competitive advantage right? It’s like building an Office Suite, or an Email client. Those are things every company needs but not things that would make sense for you to build when there are so many other more important priorities. I 100% agree with this focus on the items with the highest ROI.

But if you find that you're in charge of a large team but your budget is constrained due to expensive commodities, there is actually an option to free up some of that spend. When you reach a certain scale, building commodity software actually ends up being cheaper for you. Let me explain.

You probably saw the famous prediction from Microsoft CEO, Satya Nadella, that SaaS is dead (or will soon be dead) thanks to AI. I only partially agree with him, and I’ll discuss his prediction a bit more at the end of this case study, but I think there is a strong nugget of truth here that is very interesting. Thanks to the speed and efficiency increases brought about by Software teams who are building with AI, there is a compelling case for software to become more and more of a commodity (with lower prices). If the same team can build at 20% to 50% faster thanks to the modern AI copilots, naturally the amount of software will grow at an even faster (exponential) rate. The same number of developers should be able to build ~35% more software. If there is software that has largely been “defined” in the industry like email clients, slack applications, etc. but they are still priced at enterprise rates, those industries are ripe for disruption in my opinion. The barrier to entry is going to shrink.

That would create the environment to make custom software competitive with Enterprise SaaS options. Meaning it could be cheaper to hire the team and build it yourself once you reach a certain size. And if your business has even slightly different needs than the software that is available, the case for “build it yourself” becomes very compelling. This is why some large organizations have a “tools” team that manages internal applications.

We recently built a chat application. My initial thought was to not do this. Chat applications are fairly standardized at this point and they are also cheap to license per user. It will be a low six figure job for us to build the core features. What would the ROI on that look like? Well it turns out, as I mentioned at the beginning of this case study, that “enterprise” pricing is actually still amazingly expensive when you look at the cost for thousands of employees. The quotes were coming in between $12/month and $32/month per user for a company with 1,000 employees. These quotes included the usual enterprise requirements (SSO, 2FA, audit logging, etc.) plus access to AI enhancement options. I also reviewed the open source options and found them to be very lacking when it came to features, licensing, and/or stability.

This is when I realized that for large companies, the math actually made sense to build something custom, I was surprised. Considering, you get to define the requirements to fit your business, it is often even better because you customize it to meet your specific cultural and compliance needs. Even when I consider the ongoing maintenance costs of keeping a software team engaged, for certain scales it is still cheaper to build the chat software and own it than it is to pay for a SaaS option. Let’s look at the numbers.

Assuming the middle price of $24/month/user would be an average price range:
1,000 employees is $288,000/year
2,000 employees is $576,000/year
3,000 employees is $864,000/year

So what would it look like for a company to hire DevOcho to build a custom chat application a the same caliber of Slack or Microsoft Teams? A quick estimate, based on the features from those two applications (minus phone and video calling), was that we could build the software in 16 to 20 two-week sprints (24 to 28 weeks) for a cost of less than $350,000. Both Mobile and Web applications. We would need to keep a small operating team around to manage it once it was built to make security updates and continue to add small features. In this case, a team with a DevOps engineer, two Developers, QA, and Scrum Master could “keep the lights on”. The ongoing maintenance costs would be about 30% of the monthly cost that a company with 1,000 employees would have been paying. So for that company with 5,000 users we can create a super secure, internal chat application that provides better access for auditors, and better fits the culture of the company’s communication style. And with the savings they can hire DevOcho to build other custom applications further creating more efficiency and saving them additional money. It’s a beautiful cycle of efficiency and growth that only comes because they have a larger scale.

We can build something custom for you

We can build amazing software. Not just chat applications. But if you are in the market for an internal chat tool, with the basic enterprise features of software like Slack, WhatsApp, Teams, Discord, etc. we actually built one and have open sourced it. We are calling it “D8 Chat” because we aren’t super creative. It has basic functionality that you’d expect. And we are more than happy to customize and white label it for your specific needs.

Satya’s predictions on the death of SaaS

As far as the death of SaaS, I don’t agree fully with what Satya is saying. His prediction is that AI Agents will talk directly to backends (databases) and they will make the updates directly which will make user interfaces unnecessary. It’s a bold claim. I personally think there are situations where clicking in a user interface is still faster than talking. And with current AI, which often makes mistakes, it’s dangerous to give it direct access to a database. His long term prediction is that the databases will also go away because the AI will have sufficient memory to remember everything that is going on. This sounds cool, but I don’t think we reach this point in the next 15 to 20 years. Since the typical shelf life of software is 5 to 7 years, I believe we will go through 2 to 3 generations of software before we start to see any AI displace the backend databases. I guess time will tell but I’m 100% sure that you shouldn’t do it with the AI as it stands now.