3 min read

The AI Tax: Why Most SaaS AI Features Will Cost You Later

Imir Kujtimi

Chief Product and Technology Officer

The AI Tax: Why Most SaaS AI Features Will Cost You Later

What We Did Instead

In late 2024, we started our platform transformation: a phased evolutionary modernization. Not a rewrite from scratch. Something harder: replacing the foundation while 1,000+ customers run their businesses on the existing product every day.

We built a new architecture with a clean, unified data model, running in parallel with the existing platform and replacing the legacy layer module by module. The old system does not go dark until the new version is proven.

You do not get a clean slate in software. You earn it incrementally.

We committed to a deliberate sequence: data foundation first, then analytics, then AI. Not because we could not move faster. Because moving faster without the right foundation is exactly how the AI tax accumulates.

The Kind of Answer That Is Correct Twice

Large language models are powerful for synthesis, interpretation, and problem solving. But they are poorly suited for what enterprises depend on daily: answers that are correct twice.

This is not just a problem for data platforms. It applies equally to any business application where AI needs to reason about projects, time, revenue, and people, and produce the same correct answer every time you ask. A PSA is exactly that kind of application. When a delivery leader asks "what is the margin on this project right now?" the answer cannot be approximately right. It has to be precisely right, and it has to be the same answer regardless of who asks or when.

For that, the model needs a foundation it can trust. Without governed data, consistent definitions, and a unified data model, AI is guessing with confidence. And confident guesses are more dangerous than no answer at all.

What the AI Tax Looks Like

When you skip the foundation and bolt AI onto a legacy architecture, you get something that demos well and degrades in production. Incorrect suggestions. Inconsistent outputs. Support tickets nobody can diagnose because the issue lives three layers below the feature.

And eventually, your customers pay for the cleanup. Slower releases because the engineering team is patching around bad data. More bugs because the AI is reasoning on inconsistent inputs. A price increase tied to technical debt they never asked for and cannot see.

That is the AI tax. It does not show up on day one. It shows up in year two, when the shortcuts start compounding.

Where We Are Now

The foundation is far enough along that the next layer can sit on top of it. A unified data model where a project, resource booking, and financial posting mean the same thing across the platform. Real-time data flows. An API surface designed to be consumed by humans and agents, not retrofitted after the fact. One platform converging, with the legacy layer phasing out module by module.

Most of this is invisible to customers right now. It will not stay that way. Everything downstream gets faster and more reliable when the foundation is solid.

We are now in the analytics phase. Project Reporting is coming. Project Portfolio with embedded analytics as well. This is not cosmetic. It is the first layer of genuine insight built on top of clean, governed data. When the data model is consistent, the reports mean what they say. The numbers agree with each other.

AI comes next. And the order matters.

How We Use AI Today

We are already practicing this distinction internally. We have launched our AI in Operations Playbook, a framework for how TimeLog uses AI across the company today. It organizes every AI initiative into one of three categories: personal productivity, reusable team tools, or automated workflows with proven value. Crucially, these are not steps you climb in order. An initiative can enter at any level, including straight to full automation, as long as the value is clear and the foundation supports it.

Whatever the level, the test is the same: does it accelerate the work or compete with it? That discipline is exactly what prevents the AI tax from accumulating. AI on operations, while the product foundation matures.

That is the discipline that keeps AI working for the business, not the other way around.

What This Means for Our Customers

When the analytics layer is solid, AI features in TimeLog will not be bolted on. They will be a natural output of the architecture, running on governed data, consistent definitions, and a platform built from the start to be consumed by both humans and agents.

That is the only version of AI in a PSA we are willing to deliver to our customers. Not the fastest version. Not the flashiest. The one that gives you the kind of answer that is correct twice.