We Spent a Decade Cleaning Data. Now We Have to Clean Our Processes.

Process design, not data volume, is what separates AI projects that deliver from AI projects that stall.

In the era of business intelligence and machine learning, data ruled. Rightfully so. You have probably heard every version of it:

You need more data. Your data needs to be consistent. We need to build a data lake. Garbage in, garbage out.

To be clear, we love a clean, large dataset as much as the next AI services firm. It is how we build confidence in a strategy, spot patterns, and develop models, knowledge bases, and reports worth trusting. Data will always have a place in business, and nothing here argues otherwise.

But we keep running into the same pattern. Data gets conflated with the thing that actually determines whether an AI project succeeds.

Process, process, process

One of our founding quotes introduces this best, and I never pass up a chance to repeat it:

Businesses do not have technology problems. They have business problems accelerated by technology.

So we always ask clients the same question: when was the last time you sat down and reviewed a process end to end?

The answers are consistently interesting, and they converge on one word. Evolution.

The business grew. It expanded. People retired, new people came on, and the core processes evolved around them. The result looks a lot like a twenty-year-old codebase. Much of what remains is redundant, antiquated, or built for an environment that no longer exists. Nobody designed it that way. It simply accumulated.

Plenty of firms want to jump straight into AI, and they are right to want that. It is a force-multiplying technology that is disrupting markets. But the real differentiation we see is not in who deploys first. It is in the firms that make the underlying process as efficient as possible, then scale it with AI and automation.

Force-multiplying a broken process just gets you to the wrong answer faster.

1. Document the process

The most overlooked solution on the list.

Documenting a process forces you to fully understand it and see through the noise. It is unglamorous work, and it is usually the first time anyone has looked at the whole thing at once rather than the piece they own.

Once it is written down, you get a view from above: what is actually happening, where the handoffs break, and where AI would realistically sit. Most teams find at least one step nobody can justify anymore.

But documenting is where the work starts, not where it ends.

2. Pressure-test the procedure

You can have a documented process and still have the wrong one.

Things that were true five years ago may no longer hold. The underlying tools may be outdated, either in raw capability or in their ability to connect to anything you want to do with AI. And the output itself may never have been thought through, let alone optimized for a model to consume.

Take a portfolio review meeting. What are the key metrics we actually want to extract? Which figures do we care about most? Do associates understand how to cut through fluff and drill into ambiguity?

That question defines every step upstream of it. If the team is not extracting the right information in the first place, thirty agents reviewing their output will not surface insight that was never captured.

It also opens up more than the report. We often start these conversations focused on the deliverable that lands on an MD's desk and end up somewhere more useful: tools that augment junior employees directly, shaping how they think as they walk into a call or collect information, rather than cleaning up behind them afterward.

3. Build accountability

If your process is documented and clearly defined, you are in a good place. There is one piece left.

Definition and documentation do not mean people will follow. That is natural and not uncommon. We can strive for perfection, but still need to prepare for imperfection.

The last piece is making sure the things that move the needle are actually being followed. That requires visibility, and visibility into a process is as important as the process itself. A step nobody can observe is a step nobody is accountable for.

What we recommend

Before you build anything, make sure the underlying process is there. You and your implementation partner should be able to answer three questions clearly:

1. What is currently happening? Documented end to end, not described from memory.

2. What do we want to happen? Including what the output needs to look like for a model, and a person, to use it well.

3. How do we know if it is not happening? Named owner, visible signal.

If you cannot answer all three, you are not ready to build. You are ready to look.

This thinking is not glamorous, and it will not make anyone's demo look better. But it sets the foundation for every AI project that comes after it, and the time pays dividends. Once the process is sound, you get to force-multiply the things you already know work for your business.

BaseForge Advisors is an AI and data consultancy serving middle-market private equity firms and their portfolio companies. Contact us | Take our AI Readiness Assessment