Introduction

When ChatGPT was released in November 2022, many businesses faced what felt like an existential threat. “AI can do everything; my software business is at risk.” As an investor who managed a tech mutual fund and now building AI/ML systems, I want to address these misconceptions and explore AI’s impact on the software world, specifically in the software-as-a-services (SaaS) industry.

In this post, I will provide a framework with key metrics to consider when evaluating the quality of a software bussiness. This will help businesses understand which types of software are at risk and which are likely to thrive in the AI era. By focusing on high-value, hard-to-replicate software solutions, you can make informed decisions and leverage AI to your advantage.

Key Metrics for Software Businesses

Software is not just code; it is the layer between people and workflows. Before analyzing how AI might impact a software business, we need to first define some important metrics for measuring the quality of a SaaS businesses.

Time to Buy
This measures how quickly a customer can complete a purchase, from the start of a buying process to signing the contract. For some, this might mean reducing the number of clicks to purchase, while for others, it involves the full procurement and compliance cycle.

Cost to Implement
How quickly can a customer start using the software after making a purchase. This may also including migration costs if replacing an existing solution. This could range from minutes in product-led growth SaaS tools to several quarters for enterprise softwares.

Breaking this down further, the Cost to Implement can be relatively simple for green field projects or a complex endeaver for migrating existing software. For example, a typical migration includes several phrases: planning, data migration, system migration, user onboarding, rollout, and stabilization. Switching from OpenAI API to Anthropic API for a small volume internal chatbot? That’s just a few lines of code. Switching from on-prem Oracle to cloud-based Snowflake? Good luck.

Time to Implement
This is the time it take to implement the software, not the cost of implementation or migration. Even with more resources, certain critical path dependencies in the implementation process can’t always be sped up. This leads to the next metric, Time to Utility.

Time to Utility
This is the time between when a customer starts using the software and when they start to derive value from it. Some services provides near-instant value while others have delayed returns due to learning curves or adoption rates. It’s important to note that utility is not always easy to quantify in dollar terms. For instance, Github Copilot allows developers to use AI for coding immediately, but framing its ROI can be difficult, and teh full value might not be apparent immediately.

Average Contract Value
This refers to the pricing of the average contract, which can varies widely. Contracts might be structured as flat per-user license, consumption-based license, or a base + consumption hybrid license. OpenAI API is a classic consumption-based license and charges users per input and output tokens. The licensing cost ranges widely as well. For example, Github Copilot for retail users is $4 per month, while Bloomberg Terminal is ~$2,500 per month.

Churn
From time to time, business lose customers, either because they don’t find value in the software or switch to lower-cost options. Sometimes, businesses shut down or even new owners simply wants to “shake things up” a little.

Data Collection/Network Effects
What data does the application capture that can create a data flywheel or data moat, further optimizing the application. For example, companies can track user behaviors on their website and use that data to increase engagement. Alternatively, what kind of network effect does the service have? LinkedIn and Facebook, for instance, let users invite their contacts, building a social graph that makes users more likely to stick around and less likely to churn.

These are the key metrics to consider when evaluating the quality of a software business.

Displacement Risks by AI

Now, let’s return to the idea that “AI can do everything and my software business is at risk” and assess whether this holds true.

For softwares with long time-to-buy, high cost to implement, long time to implement, and short time to utility, AI is unlikely to pose much of a risk. However, applications with instant buy-in, quick implementation, and low switching costs are vulnerable to AI-powered alternatives that offer similar functionality at a lower price. Ironically, Large Language Models (LLMs) APIs suffers from the latter with fast setup time and low cost to implement and short time to implement.

While some mega applications like ChatGPT and Github Copilot have been extraordinarily successful, there are still not a whole lot of AI success stories outside of the startups being showered with venture capital fundings. One very specific reason is that Utility is usually very hard to calculate, streteching time to utility to quarters or longer. Github Copilot? Yes, it makes developers happy, helps developer writes code, but that’s only has minor impact ot make developers more productive.

Case Study: DocuSign

Let’s use DocuSign, a popular document-signing SaaS tool, to illustrate these metrics, assuming a small business customer:

  • Time to Buy: Nearly instant, measured in clicks.

  • Cost to Implement: None, as the web application can be used directly. However, this means switching to an alternative like Pandoc or Acrobat Sign is also low cost.

  • Time to Implement: Immediate. Small business users can start using the SaaS app right away.

  • Time to Utility: Near-instant. Once signed up, users can send or sign documents immediately, saving time and effort compared to traditional methods like printing, signing, faxing/emailing, or meeting in person.

  • Average Contract Value: $10 per month for personal use.

  • Churn: Depends on the quality of the small business; if the bussiness is doing well, the number of documents signed might cross the threshold and bump the user into another pricing tier. But if the business is not doing well, the user might even cancel the service.

  • Network Effects: DocuSign enables multi-party signatures, creating opportunities to convert other users if they aren’t already using Docusign. This lowers customer acquisition costs in the inbound marketing funnel. While converting a free user to a paid user can still be challenging, at least it captures mindshare.

What This Means for Businesses

Distribution
With AI reducing barrier to entry for SaaS businesses, distribution becomes even more critical. Justin Kan, of Justin.tv and Twitch fame, once quoted a common Silicon Valley saying: “First-time founders focus on product, second-time founder focus on distribution.”

alt text

Optimize
Focus on optimizing these metrics to deepen your moat. Long sales cycle? Shorten it. High implementation costs? Lower them. Low contract value? Add new features customer will pay for.

Network Effects
Build network effects using data-as-a-product, such as marketing funnels and product analytics, to collect unique data that competitors will find hard to replicate. This can create a data flywheel effect, improving customer acquisition, reactivation, and retention.

@article{
    leehanchung,
    author = {Lee, Hanchung},
    title = {Navigating Through Time: The Evolution of UX in AI Systems},
    year = {2024},
    month = {08},
    howpublished = {\url{https://leehanchung.github.io}},
    url = {https://leehanchung.github.io/blogs/2024/08/30/ai-impact-software/}
}