From Incomplete Data to Reliable Forecasts: What “Good Enough” Data Looks Like

For many mid-market leaders, forecasting feels like an impossible balancing act.

On one side is the pressure to make decisions quickly in a competitive market. On the other is the reality that the data needed to support those decisions is often incomplete, inconsistent, or difficult to trust. The result is a familiar cycle: leaders delay action while waiting for better information, only to discover that the perfect dataset never arrives.

The assumption is that better forecasts require perfect data. In reality, the organizations that forecast most effectively understand a different principle: forecasting is not about perfection. It is about confidence, context, and knowing when additional data will no longer meaningfully improve the decision.

The Cost of Waiting

One of the most common challenges organizations face is determining whether they have enough information to move forward.

According to Altum’s Brendan Polke, that conversation should begin with a simple question: what is the cost of waiting?

Organizations often focus on the gaps in their data while overlooking the business consequences of delaying decisions. Every week spent chasing additional information carries a cost, whether in lost revenue, missed opportunities, slower product launches, or competitive disadvantage.

As Polke explains, “even if we only hit 75% of our goal, that’s way better than hitting 100% after the competitor goes to market.”

The challenge is not determining whether the data is perfect. The challenge is determining whether the next round of data collection will materially improve the outcome.

There is always an inflection point where the effort required to gather, clean, and validate additional information begins to outweigh the value that information provides. While there is no universal percentage that defines “good enough,” experienced leaders recognize when they are spending more time refining forecasts than improving them.

In many cases, the pursuit of certainty becomes the very thing preventing progress.

Forecasting Starts with Trust

Data quality discussions often focus on technical issues such as missing records, inconsistent definitions, or outdated information. While those problems matter, forecasting challenges frequently stem from something more fundamental: trust.

Leaders rarely reject forecasts because they dislike the numbers. They reject forecasts because they do not trust how those numbers were produced.

“Trust in data is very similar to trust in people,” Polke notes. When a report consistently produces results that align with expectations and business reality, confidence grows. When reports generate unexpected or contradictory outputs, trust erodes quickly.

That is why successful forecasting initiatives often begin by validating known outcomes.

Rather than immediately introducing new models, organizations can build confidence by recreating historical reports using new tools and methodologies. When leaders can see familiar inputs producing expected outputs, confidence in the underlying model begins to grow.

This becomes particularly important as organizations introduce AI-driven forecasting capabilities. Many executives do not need to understand every technical detail behind the model. They simply need confidence that the system behaves logically and consistently.

The goal is not to eliminate the black box entirely. The goal is to demonstrate that the black box produces outcomes that align with reality.

Defining “Good Enough”

The definition of good enough data varies across industries and business models, but there are common characteristics.

First, the data must support day-to-day operations. If employees cannot reliably perform core business processes, forecasting will always suffer.

Second, organizations must establish consistency around their most important business drivers. For companies that manage inventory, sales, or supply chains, that often means ensuring visibility into the products, services, or activities that generate the vast majority of revenue.

Polke frames this through an 80/20 lens. If organizations can achieve accurate, reliable data for the products and activities that drive 85% to 90% of the business, they often have enough information to support planning and forecasting decisions.

The bigger challenge is ensuring that information is structured correctly.

Forecasting problems often start with something surprisingly simple: organizations don’t always track information at the level they need. A retailer may know how many pallets of inventory arrived but not how many individual products are available to sell. Other companies may use different definitions for the same data point across departments. When the underlying information isn’t organized consistently, forecasting becomes less about predicting the future and more about trying to reconcile competing versions of the present.

Among common data quality issues, inconsistent definitions are often the most damaging because they undermine confidence in every downstream metric.

Without a shared understanding of what is being measured, forecasting becomes guesswork regardless of how much data is available.

More Data Is Not Always Better

One of the most persistent misconceptions in forecasting is that more data automatically produces better forecasts.

In many cases, organizations already have the information they need, but they struggle to separate what matters from what doesn’t. When companies try to track everything, forecasting models can become cluttered with data that adds complexity without improving accuracy.

The most effective organizations focus on the metrics that directly influence business performance and use those as the foundation for planning and forecasting.

More information can be valuable. More noise rarely is.

Why AI Changes the Equation

Historically, forecasting models were rigid. Missing data could break the model entirely. Introducing new variables often required extensive technical work and significant time investment.

AI is changing that dynamic.

Modern tools can test scenarios, identify patterns, backfill certain gaps, and evaluate multiple forecasting approaches in a fraction of the time previously required. What once demanded specialized programming expertise can now be accomplished through effective prompting, model training, and iterative testing.

That does not eliminate the need for quality data. It does, however, make it possible to extract meaningful value from imperfect data much sooner.

For mid-market organizations, that shift is significant. The conversation is no longer whether forecasting can happen before data is perfect. The conversation is how quickly organizations can build enough confidence to begin.

The companies that answer that question successfully are often the ones that move faster, adapt sooner, and make better decisions while competitors are still waiting for perfect information that may never arrive.