Baseline: definition and value
Definition
A baseline is the measured state before a change. It records how long a process takes, how high an error rate is or what a step costs before AI is introduced. Without this starting value, you cannot later tell whether an improvement truly comes from the AI or is merely chance.
Why the Term Matters
AI value is often claimed but rarely evidenced because the before value is missing. A baseline closes that gap and makes impact provable. As a neutral advisor we treat it as mandatory: without a starting value you can judge neither the ROI nor a failure cleanly. Defining the baseline early lets you decide later on facts instead of impressions.
When It Is Relevant
- Before a pilot, value should later be evidenced measurably.
- An investment decision needs a dependable before-and-after comparison.
- Several stakeholders dispute the actual starting state.
- An AI system should be monitored against a reference value over time.
When Not
When a change is purely qualitative and no one needs to prove value, a formal baseline is overkill. And when the available data is too messy to form an honest starting value, the data situation should be clarified first rather than manufacturing false precision.
Industries That Use It
- Production: record lead time and scrap before an optimisation to evidence impact.
- Customer service: measure handling time and resolution rate before an assistant is introduced.
- Sales: set conversion rates as a reference before AI-supported prioritisation starts.
- Logistics: capture on-time delivery and error rate as a starting value.
Typical Use Cases
- Record process times and error rates before a pilot.
- Measure cost per case to compute the ROI honestly later.
- Define a reference value for ongoing monitoring.
- Build before-and-after comparisons for an investment decision.
- Align team expectations on a shared factual basis.
Decision Guide
| Signal | Baseline fits | Alternative |
|---|---|---|
| Value should be evidenced | yes, measure the starting value | without a before value, impact stays unclear |
| Purely qualitative change | rather not | manage without a formal baseline |
| Data too messy | clarify the data first | avoid false precision |
| Ongoing monitoring planned | set a reference value | deviations stay unassessable |
Example
A service team wants to test an AI assistant. Before the pilot we record handling time and resolution rate as a baseline so the later assessment rests on a real before value. This describes the method, not an asserted client metric.
Service Connection
We define the baseline in the AI readiness check and the AI potential analysis before a pilot starts in AI implementation.
Sources
This framing describes common practice in impact measurement. It is general orientation and does not replace a project-specific definition of metrics, which must be made case by case.