How do you measure the return on an enterprise knowledge base?

Why “how many people did it replace” is a poor measure

It cannot be verified and it cannot be attributed. If customer service headcount did not fall and sales did not close more deals after launch, you have no way to tell whether the knowledge base delivered value — which does not mean it failed, it means the wrong measure was chosen.

The right sequence is to measure the process first, and the outcome after.

Three process measures

1. Hit rate. Of the real questions front-line staff ask, what proportion gets a usable answer. This is directly testable: collect 30 real questions, ask them, and record how many are answered. Below half means the knowledge set does not match the questions — not that the AI is failing.

2. Adoption rate. Whether the answers received actually get used — sent to a customer, written into a proposal, passed on to a colleague. A hit is not the same as useful; adoption is what shows the answer was good enough to use.

3. Lookup time. For a given document, how long it takes on average to go from “dig through chat, dig through the shared drive, ask a colleague” to “just ask”. It is easy to measure and easy to feel: most companies measure a drop from several minutes to the order of a dozen seconds.

Once all three move, look at outcomes

At that point you can look at whether new staff get up to speed faster, whether repetitive questions stop interrupting experienced colleagues one by one, and whether customer service responds faster. Those three can be converted into staff hours, or left unconverted — a consistent and sustained direction of travel is enough to justify continuing to invest.

What investment looks like

A small business building a usable knowledge base typically spends in the low thousands of CNY on the first build (one-off construction and deployment), with annual maintenance in the same range on a yearly basis. The exact figure moves with the volume of material, the number of access channels and the degree of self-hosting. Compared with employing someone full-time to keep documents in order, that is a difference of an order of magnitude — but only on the condition that the business owner is genuinely involved in reconciling the wording. Otherwise it is a cheap asset that sits unused.

One warning sign

If a month after launch the front line still instinctively asks a colleague first and the knowledge base is only used because it is required, that is not an ROI problem — it is a content quality problem. Go back to hit rate: does the question list fail to cover the real situations, and is wording that should have been updated still sitting at the old version.

The one-pager for the boss

One page is enough: this month's value for the three process measures, transcripts of three real questions and their answers, and one record of a wrong answer that has since been fixed. A real question-and-answer transcript is more persuasive than any percentage.