How to measure content performance in training: metrics that tell you what to fix

Content performance and training effectiveness

Content performance asks whether one piece of content does its job, such as a forklift pre-check video or a one-page job aid. Training effectiveness asks whether a whole program changed behavior and business results, and answering it needs manager observation and business data from the weeks and months after training. Our guide to measuring training effectiveness covers that program-level method.

This guide stays at the level of the asset, where each finding names something specific to change, such as a segment to re-record or a question to rewrite. A program can show good results overall and still contain a module that most learners misunderstand, and program-level numbers will not tell you which module that is.

Learning content performance metrics

Choose the metrics in each layer that match the asset's job. The last column says what a weak result points to, so each number leads to something you can change.

Learning content performance metrics, with formulas and what a weak result points to

Layer	Metric	How to calculate it	A weak result points to
Reach	Open rate	Learners who opened the asset ÷ learners it was assigned to × 100	Delivery: the link, the assignment, or where the asset sits in the course
Attention	Watched percentage or scroll depth	Average share of the asset each learner consumed	Length, or an opening that does not state the task
Attention	Drop-off point	The moment where many learners' furthest point stops	The segment just before that moment
Attention	Replays	Sections played more than once	A step that moves too fast, or a step learners use as a reference
Participation	Participation rate	Questions answered ÷ questions shown × 100	Learners letting the content play without working through it
Understanding	First-attempt correct rate	Learners right on the first try ÷ learners who answered × 100, for each question	The explanation before the question, or the question itself
Understanding	Most-chosen wrong answer	Share of learners who picked each wrong option	The specific misconception the content leaves in place
Use	Resource clicks	Learners who opened the linked job aid ÷ learners who saw the link × 100	A resource shown at the wrong moment, or one learners do not need
Use	Task result	Errors, rework, or help tickets on the task the asset teaches, before and after release	Content that teaches recall when the job needs a decision
Reaction	Relevance rating	Average answer to "How useful was this for your work?" on a 1 to 5 scale	Content aimed at the wrong audience or the wrong problem

For definitions of the watch metrics and figures by video length, see our guide to video engagement metrics. For the participation rate and how to calculate it at each moment in a video, see what engagement rate is and how to work it out.

How can I measure the effectiveness of interactive training content using actionable content analytics?

Interactive content records what each learner answers and chooses, so you can measure understanding inside the asset instead of inferring it from completion. Set up the measurement before the first learner watches:

1. Write the job of the asset in one sentence, naming the decision or skill it supports. For example: "Warehouse staff choose the right lifting equipment for a load."
2. Place one question or scenario choice at the moment each objective is taught, and make it test the decision the job requires. A question that asks which equipment to use for a heavy pallet in a narrow aisle tests the job. A question that asks for the equipment's name tests recall.
3. Identify each learner by email or employee ID, so you can follow up with individuals and later join answers to work data. In Mindstamp, use lead capture, a personalized link, or ?email= at the end of the link, as our help article on tracking a training course shows.
4. Agree on decision rules with the content owner before launch: for each metric, the result that triggers a change and the change you will make. The decision table further down gives examples.
5. After the first group of learners finishes, read the question results first, then the drop-off, then the work data.
6. Change one thing, and compare the same question for the next group of learners.

In the demo above, learners finish the video but miss the selection scenario. Total library views cannot show which equipment cases learners get wrong. Answer accuracy for each case can, and it tells the team which part of the lesson to revise.

In Mindstamp, a question pauses the video and can require an answer before playback continues, and each answer is stored against the learner. The reports give you watched percentage, participation, and correctness for each view, an answer breakdown for each question, and drop-off by section on supported plans.

Read each question like a test item

An in-video question measures two things at once: the learner and the content before it. Item analysis, the method testing offices use to review exam questions, helps you separate the two. The University of Washington's guide to understanding item analyses describes three checks that also work for knowledge checks in training content.

Difficulty: the share of learners who get it right

For a question with one correct answer, item difficulty is the percentage of learners who answer it correctly, so a higher value means an easier question. The University of Washington's scoring software labels a question easy at 85% or above, moderate from 51% to 84%, and hard at 50% or below, and the guide calls that classification arbitrary. For a four-option multiple-choice question, it cites an ideal difficulty of 74% for separating learners who know the material from those who do not.

Use first attempts. If learners can retry until they answer correctly, the later attempts hide the first miss. When most learners miss a question, check the question before the content: confirm that the wording is clear and that the keyed answer is correct, then review the segment before it.

Discrimination: whether strong learners get it right more often

A useful question is answered correctly more often by learners who do well on the rest of the module. The University of Washington's software reports this as a correlation between each question and the total score on all other questions, and labels it good above 0.30, fair from 0.10 to 0.30, and poor below 0.10. A negative value can mean the question is mis-keyed. Without statistics software, use the older hand method the guide mentions: split learners into a high-scoring group and a low-scoring group by overall score, and compare their correct rates on the question. If both groups do about as well, the question does not separate learners who know the material from those who do not.

The same guide notes that discrimination runs lower when a test covers a wide range of content. A short knowledge check spread across several concepts is that kind of test, so read the question again before you act on a low value.

Distractors: which wrong answer learners choose

The guide notes that frequently chosen wrong answers may indicate common misconceptions. If most of the learners who miss the equipment question pick the same wrong option, you know what the lesson failed to rule out. If learners who score well overall pick a particular wrong answer, the question may be ambiguous.

Where to find the data in Mindstamp

The Questions report shows the answer breakdown for each question. The Interactions export lists every answer with its date and whether it was correct, so the earliest answer for each learner and question is the first attempt. The report score counts every attempt when Repeat Until Correct is on, so a wrong answer followed by a right one scores 50% for that question (passing scores and retries). To read a whole course, export a completion report with one row per learner and a score column for each video, as described in reporting across multiple videos.

The University of Washington guide also calls item analysis data tentative: the number of learners and chance both affect it. With 20 learners, one answer moves a question's difficulty by 5 percentage points. Record the results for each group of learners and act on a pattern that repeats.

Decide what to keep, fix, or retire

Review your content as a list with one row per asset, and record a decision for each. The table lists common patterns and the decision each one supports.

Content performance patterns and the decision each one supports

What you see	Decision	First change to make
Few assigned learners open the asset	Fix delivery	Check the link, the assignment, and where the asset sits in the course before you change the content
Learners finish, but most miss one question on the first try	Revise	Check the question, then rewrite or re-record the segment before it
Many learners stop at the same moment	Revise	Shorten or clarify the segment just before the drop
Nearly everyone answers the early checks correctly, group after group	Shorten	Let learners who answer correctly skip ahead with branching, and keep the full path for the rest
Learners replay one step and open its job aid	Add a job aid	Turn the step into a one-page aid or a short reference clip linked at that moment
Scores are high, but errors on the task do not fall	Redesign	Replace recall questions with scenario choices that match the decision at work
Few learners open it after launch, and the process it teaches has changed	Retire or merge	Archive it, or move the current steps into another asset

Fix first the assets that many learners use for an important task and that show the weakest result. For example, a safety procedure that 400 people take each quarter, with a 45% first-attempt correct rate on its key question, comes before an optional module with 12 learners. Both figures are illustrative.

How to measure training effectiveness beyond completion rates

Add measures that a learner cannot meet by letting the content play. At the content level, those are first-attempt correctness on questions placed where each concept is taught, the drop-off point, and resource use. At the program level, add behavior on the job and a business result, each compared with a baseline you record before training. To estimate what low completion and comprehension cost in learner time, enter your own figures in the training ROI calculator.

If you also measure marketing content

Marketing content has a different job, so it needs different metrics. Reach Influencers' 2025 overview of digital marketing performance metrics sorts them into three tiers. The first covers audience growth and visibility (website traffic, impressions, and reach). The second covers engagement and conversion, with click-through rate, cost per lead, and lead quality. The third covers revenue, with customer acquisition cost and return on ad spend. For articles and landing pages, see our guide to content engagement metrics. To connect content to revenue, see how to measure content marketing ROI.

Common questions

How do I assess the engagement level of training content?

Calculate an engagement rate at each moment that asks learners to act: the learners who answered or clicked, divided by the learners who reached that moment. Read it next to the drop-off point and first-attempt correctness. Engagement shows that learners took part, and correctness shows whether the content worked. Our guide to user engagement metrics covers engagement across a learning platform rather than a single asset.

What metrics should you track for training and documentation videos?

Track what matches the video's job. A training video teaches a decision or a procedure, so read first-attempt correctness on the question that tests it, the drop-off point, and replays. People open a documentation or how-to video to find one step, so a low watched percentage is normal; check where they start and leave, whether they open the linked article, and whether help tickets on that topic fall. Our guide to video analytics for training content lists the metrics for each type of training video. Chapters help viewers reach the step they came for.

How often should I review content performance?

Review new content after the first group of learners finishes it. After that, review it whenever the process it teaches changes or the work metric it supports moves, and set a regular review date when you publish anything learners must take every year.

Can my LMS report content performance?

An LMS records what the course sends it. A Mindstamp SCORM package, in SCORM 1.2, SCORM 2004 3rd edition, or SCORM 2004 4th edition, sends completion and a score to the LMS on supported plans, and passes the LMS learner name and ID to Mindstamp. By default, the score in the LMS counts the first answer to each question. Each answer, click, and path stays in Mindstamp reports, which is where you read question results and drop-off. See LMS compatibility for the other ways to deliver a video in an LMS.

To start, pick the training asset that the most learners take for an important task. Write its job in one sentence, put one question at the moment it teaches the key decision, and read the first-attempt results after the next group of learners finishes.

Sources and further reading

• University of Washington Institutional Assessment and Evaluation, understanding item analyses: item difficulty, item discrimination, and distractor review.
• Reach Influencers, digital marketing performance metrics (2025): a three-tier model for marketing metrics.