Amiel Kollek is a Software Engineer at Lindus Health who works extensively with AI to automate manual processes and improve the capabilities of our eClinical technology, Citrus™.

Why do even the best SaaS solutions leave users wanting more? Some tools are incredibly powerful, but you need a consultant to use them. Others are straightforward but don’t fit your use case. Or maybe it’s almost perfect, but it’s just missing that one feature you need.

It’s not a coincidence, it’s because of something I call the ‘SaaS Trilemma’: SaaS can't be powerful, flexible, and also simple. It has to pick two.

Powerful, Flexible, or Simple: Pick Two

Software often has to make a compromise between power (depth of functionality), flexibility (adaptability to different use cases), and simplicity (ease of understanding/use) because, simply put, it’s really hard to tackle all three.

Let’s examine some examples to see the SaaS trilemma in practice:

• Salesforce: Powerful and Flexible – Salesforce is very comprehensive and has everything you’d need. But it has a steep learning curve and often requires special expertise to properly navigate.
• Wix: Flexible and Simple – Wix makes it super easy to set up a wide range of websites. However, any more advanced application is difficult to support.
• Slack: Simple and Powerful – Slack’s intuitive design and structure make it a great way to communicate in a team. It is not very configurable though and assumes everyone will be using it in the same way.

These products aren’t bad, but they are victims of the SaaS Trilemma. And maybe that’s OK. Depending on the product, there might not be any need to avoid the Trilemma (I see no need for Slack to start being extra configurable, for example).

But for a lot of software, the Trilemma is a real problem. Why is that?

Fundamental Tension

The SaaS Trilemma: SaaS can't be powerful, flexible, and also simple. It has to pick two.

The SaaS trilemma stems from the tension between power, flexibility, and simplicity. Powerful software offers deep functionality, but making that functionality available often sacrifices simplicity. Flexibility enables multiple use cases but increases configuration complexity, making the user experience harder. Simplicity is achieved by hiding complexity through opinionated design, which improves usability but limits capability. While power and flexibility can coexist, each new feature requires more effort to maintain configurability, meaning that teams with limited resources who are building for both face practical constraints.

‍

In practice, this often leads teams to a choice. Whether it’s done in the open, or without anyone realizing, each team building SaaS must pick two from power, flexibility and simplicity.

The Zapier Exception

So are we going to be forever stuck with imperfect SaaS? AI offers a glimmer of hope to break out of the constraints of the Trilemma for teams who know how to leverage it!

A great example of this is Zapier. Zapier sets up automations and integrations between different tools. It’s powerful (the tools it integrates with have lots of functionality), flexible (there are loads of integrations available) and building new integrations with Zapier’s copilot means you just write a few lines of English to get what you need (simple!). How did Zapier manage to somewhat sidestep the Trilemma? The secret lies in its copilot and use of AI.

Breaking the SaaS Trilemma in Clinical Trials with AI

Using AI to support English as an interface through Large Language Models (LLMs) means you have a huge advantage when it comes to the SaaS Trilemma. Spoken language is powerful, flexible, and your users have spent their whole life using it — so it's natural, they have no trouble with the complexity!

Language isn’t a perfect interface, however. Software engineers are familiar with the complaint that what they built isn't what was asked for. The issue is that instructions in English are often vague, ambiguous, or inconsistent in non-obvious ways.

Even if language was a perfect interface, AI isn’t a magic bullet. Tasks as large as setting up a clinical trial are still quite difficult for AI, and without guidance and constraints there’s a high risk of hallucination. More importantly, attempting to replace SaaS entirely with an LLM is kind of like asking AI to reinvent the wheel. Why give up on loads of existing functionality? 

AI and its users need constraints to guide their work in order to be effective, especially for complex, domain-specific tasks that LLMs might not grasp. Software, particularly in the powerful and flexible category, has been built up over years to provide exactly those constraints. It has embedded within it loads of domain specific expertise. AI can unlock this functionality while keeping tools intuitive and manageable.

This isn’t to say you can sprinkle some AI over SaaS to make it simple. As always, there is no free lunch. This does, however, present a path to developing simpler interfaces for powerful and flexible tools, if you can build agents that understand your tool and your domain.

Example: How We Build With AI at Lindus Health

In the past, I’ve spoken about how the blind application of AI can’t fix a broken system. That has long been our belief at Lindus Health, and we’ve invested heavily in building better systems (both technical and organizational) to address the core problems in the clinical trials industry. One example is Citrus™, our in-house eClinical platform for executing and managing our clinical trials. 

A challenge we’ve faced is that when building a comprehensive tool for a complex domain, the result tends to be powerful and flexible (i.e. not simple). We can create curated experiences for specific users and use cases, but the full functionality of our platform is complex. That’s where AI comes in.

Consider a concrete example in clinical trials. Simply put, a clinical trial consists of a series of activities to capture essential data (like a blood test or a questionnaire). Since trials have complex workflows, Citrus™ supports a powerful and flexible range of logic for when these activities should become available. Previously, the UX for configuring an activity that becomes available “4 days after the participant is enrolled, but only if they answered yes to Question A and no to Question B” was complex. With AI, however, a trial designer can just type out a description of a rule as above, and Citrus™ will automatically configure the activity and allow the designer to instantly test it.

The key is building an AI system which deeply understands what our platform is capable of and agents which understand how to configure it. We’ve found Claude particularly effective at this due to its superior ability to understand code. Equipped with these agents, and combined with good user experience, we can sidestep large parts of the SaaS Trilemma and enable powerful and flexible clinical trial configurations while maintaining a simple and easy to use interface for trial designers.

With this AI enabled flow, we’re able to expose the full power and flexibility of our platform while keeping the interface simple enough that technical specialists only need to review the results and occasionally step in for especially complex cases. This means reduced risk of human error, faster feedback loops, and freed up technical resources.

Clinical Trial Software and the Next Generation of SaaS

The breaking down of the SaaS Trilemma presents a potential turning point for how software enables us to be more effective, especially in complex domains, and clinical trials is a prime example.

There is a massive treasure trove of power and flexibility that has hitherto been locked away behind a barrier of steep learning curves and difficult UX, and it seems like AI might hold the key.

In my next blog, I’ll discuss why most clinical trial software has ended up broadly in the powerful and flexible category, what the implications of AI are for this domain, and more detail on how we at Lindus Health apply AI to safely accelerate our trials. Stay tuned!