Accuracy
What is Accuracy?
Accuracy, as a product strategy, means customers choose a product because its core output is right more often than a rival's: a security tool that catches more threats, a test that reads more samples correctly, an estimate closer to the real price. The gap has to be measured in a trial, benchmark or independent test.
How Accuracy works
Where correctness is the reason to buy, a measured lead gives buyers something concrete to act on. It can win a head-to-head evaluation, persuade an insurer to pay for a test, or let users stop double-checking the output.
Accuracy decides markets less often than it seems, because buyers also weigh speed, price, workflow and switching costs. In 2017 an NSS Labs test scored Cylance above CrowdStrike on detection, and CrowdStrike still won the market. A trial showing a product beats usual care is evidence that it works. It says nothing about how the product compares with a rival.
Accuracy vs. Reliability and Speed
| What it is | How to tell it apart | Example | |
|---|---|---|---|
| Accuracy | Customers choose it because its core output is correct more often than a rival's | Look for a measured comparison of right-or-wrong output against a named rival | NSS Labs scoring Cylance against CrowdStrike on detection |
| Reliability | The product fails, crashes or drops work less often than the alternative | Asks whether it worked every time; a working product can still give wrong answers | Teams leaving HipChat for Slack after failed message delivery |
| Speed | Faster than the alternative at the customer's core task | Measured in time to finish the task; the result's correctness is a separate question | Reviewers finding Spotify faster and more responsive than Rdio |
How to tell if a company has it
- An output that can be wrongIts correctness is the point of the product: a detection, test result, estimate, match or prediction.
- A measured comparison with a named rivalA trial, benchmark, independent test or formal evaluation scoring both products on the same cases. A comparison with usual care doesn't count.
- The difference changed a decisionThe lead shows up in adoption, reimbursement, a buyer's choice or a head-to-head evaluation the company won.
- Same-period tests point the same wayIndependent tests from the same years don't favor the rival. Cylance led one detection test and still lost the market.
Examples
- NSS Labs' 2017 test scored Cylance above CrowdStrike on detection. CrowdStrike still won the market.
- Two listing-accuracy studies found Zillow and Trulia about even.
Who names it
| Author | What they call it / where it appears |
|---|---|
| Husain & Shankar + | Evals: measuring whether an AI product's output is correct |