Search for emotional intelligence strategies and you will find the same five headings on every page: self-awareness, self-management, empathy, social skills, growth mindset. Underneath them sit tips like “reflect daily” and “listen actively,” and underneath those sit statistics that mostly cannot be traced to anything.
The underlying question is more interesting than the tips, and it has a real answer. Emotional intelligence training has been meta-analyzed. So have the tests. The two results point in awkwardly different directions, and knowing that changes what a sensible program looks like.
The short answer
Emotional intelligence can be trained: a meta-analysis of 58 studies found a moderate positive effect on emotional intelligence scores. But self-report emotional intelligence tests measure mostly personality and self-belief, and predict job performance at close to zero once those are controlled. So train specific behaviors, and do not buy an assessment to prove it worked.
Is emotional intelligence teachable?
Yes, with a caveat worth understanding before you spend a budget.
Mattingly and Kraiger (Human Resource Management Review, 2019) pulled together 58 published and unpublished studies of emotional intelligence training programs, using pre-post or treatment-control designs, and found a moderate positive effect regardless of design. The effect held across participant gender and across both types of measure, ability-based and mixed-model. That is a more solid answer than most workplace skills can offer.
The caveat is in the same paper. Effect sizes were larger in published studies than in dissertations, which is the standard signature of publication bias: studies that found an effect got printed, some that did not never left the university. It does not erase the finding. It means the true effect is probably at the smaller end of what the headline suggests.
Read plainly: training moves emotional intelligence scores. Whether it moves behavior at work is a separate question that the training literature does not fully answer, which is why the measurement section below matters more than the strategies section.
Why the 90 percent statistic does not belong in your business case
Nearly every article on this topic opens with the claim that 90 percent of top performers have high emotional intelligence. It comes from TalentSmart, a company that sells emotional intelligence assessments and training, drawn from its own commercial testing database. It has not been published in a peer-reviewed journal, the underlying data has never been released for independent analysis, and the vendor has an obvious interest in the result.
That does not make it false. It makes it unusable in a business case, because the first person who checks the source will find a marketing page, and everything you said after it will inherit the doubt.
There is a second and more uncomfortable finding. Joseph, Jin, Newman and O’Boyle (Journal of Applied Psychology, 2015) meta-analyzed what self-report emotional intelligence tests actually measure. The answer was a combination of ability EI, self-efficacy, self-rated performance, conscientiousness, emotional stability, extraversion and general mental ability, with a multiple correlation of .79. In other words, the questionnaire is largely re-measuring personality and self-confidence under a new name.
Their correlation between mixed EI and supervisor-rated job performance was .29, well below the .47 that had been circulating. Once they controlled for those underlying constructs, the relationship fell to essentially zero, a beta of negative .02.
The practical consequence is narrow and important: run the training, skip the questionnaire as a proof of value. A pre and post score on a self-report instrument mostly tells you whether people feel better about themselves, which is not nothing and is not what you were asked to demonstrate.
Which parts of emotional intelligence are actually trainable
Daniel Goleman’s four domains remain the most useful map, so long as they are treated as categories rather than as evidence. What follows is each domain reduced to the part that can be practiced, observed and corrected. Everything else in the popular version is disposition, and disposition does not move in a workshop.
How to improve self-awareness, which needs other people
Self-awareness is the one domain you cannot train alone, which is inconvenient because journaling and daily reflection are the two things every article recommends. Reflection improves the story you tell yourself about your behavior. It does nothing about the gap between that story and what colleagues experienced, and that gap is the entire problem.
What works is structured external feedback on specific incidents. Emotional awareness improves fastest when someone else supplies the data. Not an annual 360 with 42 rating scales. Three questions to two colleagues after a meeting that went badly: what did you see me do, what effect did it have on the room, what would you have done instead. Written down, monthly, reviewed against your own account of the same meeting.
The measure: how often your prediction of how you came across matches what other people report.
How to improve self-management under pressure
Self-management is the most trainable domain because the trigger is identifiable and the intervention is mechanical. Pick the two situations where your responses cost you something: a specific person’s tone in review meetings, a certain kind of escalation email, being interrupted in front of the team.
Then attach a rule to each, not an intention. “I do not reply to that category of email for two hours” is a rule. “I will stay calm” is a wish. Rules survive the moment they are needed. Intentions are made in a calm state and fail in a hot one.
The measure: count the number of times the rule was invoked, monthly. It is a behavior count and does not depend on anyone’s judgment.
Social awareness, and the limits of reading the room
This is where most training oversells. Accurate reading of other people’s emotional states is partly a stable individual difference, and training moves it least. The teachable part is not perception, it is verification: asking rather than assuming.
Replace “I could tell she was frustrated” with a habit of checking. One sentence, out loud: “You went quiet after that point, did I read that wrong or did something land badly.” That sentence is trainable in an afternoon. Accurate mind reading is not trainable at all, and programs that promise it are selling perception as if it were procedure.
The measure: in reviewed conversations, how often an assumption about someone’s state was checked rather than acted on.
Relationship management, which is mostly scripted moves
The highest-value domain and the most concrete. Difficult conversations, feedback, disagreement and repair all have known structures, and people who seem naturally good at them are usually running a sequence they learned somewhere.
Give people the sequence explicitly: describe the specific behavior, state the effect, ask for their account, agree the next step. Then have them run it on a real situation they are currently avoiding, with a manager observing. This overlaps heavily with structured workplace communication protocols, which is not a coincidence. Most of what gets labeled emotional intelligence at work is a communication sequence someone has practiced.
The measure: whether the conversation happened at all. Avoidance is the failure mode, and it is easy to count.
A 12-week emotional intelligence improvement plan
Most people searching for an emotional intelligence development plan get a list of tips instead. Here is an actual plan. It assumes one person, one manager, and no budget.
| Weeks | Focus | What you actually do | Evidence at the end |
|---|---|---|---|
| 1 to 2 | Baseline, not assessment | Write down the three recent situations that went badly. Ask two colleagues the three feedback questions about one of them. | A written gap between your account and theirs |
| 3 to 4 | Self-management rules | Name your two reliable triggers. Write one mechanical rule for each. Tell your manager what they are. | Two rules, in writing, with a count started |
| 5 to 6 | Verification habit | Use the check sentence at least twice a week instead of assuming a colleague’s state. | A log of assumptions checked, and how often you were wrong |
| 7 to 9 | The conversation you are avoiding | Run the four-step sequence on one real situation, with your manager observing or debriefing after. | The conversation happened, plus the manager’s notes |
| 10 to 11 | Repeat under load | Same sequence, harder situation, no rehearsal. This is where the skill either holds or does not. | Second observation, compared with the first |
| 12 | Review against week 1 | Re-ask the same two colleagues the same three questions about a recent meeting. | Change in the gap, which is the only score worth having |
Twelve weeks is not arbitrary. It is long enough for the week 10 test to sit far from the week 7 rehearsal, and short enough that people finish it. Programs that run a year without an observation point produce attendance, not change.
How to measure it without buying an assessment
Given what the 2015 meta-analysis found about self-report instruments, the sensible measures are behavioral and boring.
- The prediction gap. How close a person’s account of an interaction is to what others report. It shrinks or it does not.
- Conversation completion. How many avoided conversations were actually held. Managers usually know this number without being asked.
- Rule invocation counts. Self-reported but hard to fake usefully, because there is nothing to gain from inflating it.
- Second-order signals with a long lag. Voluntary turnover on a team, grievance volume, and whether people bring problems early. These move slowly and are affected by ten other things, so treat them as context and never as proof.
If your organization does run assessments for development conversations rather than for proving results, keep them inside the process rather than in a separate tool. Assessment features are useful for capturing manager observations against a rubric, which is a different job from scoring a personality questionnaire.
Where this approach falls short
- Publication bias is real in this literature. The training effect is likely smaller than the headline meta-analytic figure, and anyone citing that figure without the caveat is not reading carefully.
- Training moves scores. The evidence that it moves job outcomes is thinner, because the studies mostly measure emotional intelligence rather than performance. Do not promise the second when you can only demonstrate the first.
- Behavior does not survive a hostile environment. If raising a problem gets someone punished, no amount of relationship management training will produce candor. That is a management problem wearing a training costume.
- Individual differences persist. Some people will be better at reading others after training and still worse at it than an untrained colleague. Programs that imply everyone reaches the same level set themselves up to disappoint.
- None of this is clinical. Emotional intelligence training is a workplace skills intervention. It is not treatment, and a person in real distress needs support, not a development plan.
Running the program without adding administrative load
The failure mode for this kind of program is not content, it is that the observation points never happen. Week 7 and week 10 in the table above each require a manager in the room, and those are the two things that quietly drop off a calendar.
That is the scheduling problem a training management system for enterprise L&D teams exists to solve: recurring sessions, named observers, and a record that shows which observation happened and which was skipped. Where the practice sessions are live, which they should be for anything involving a difficult conversation, instructor-led training management handles cohorts, facilitators and rooms without a spreadsheet.
If the program sits inside a wider leadership development track, keep the emotional intelligence component on the same record as the rest, so a manager sees one history per person rather than four. The same applies when this is one strand of a broader skill development program: separate systems produce separate truths, and the one that gets updated is whichever is easiest.
Reporting should be tied to the observation record, not to completion. Training management reporting is worth configuring around “did the week 10 observation happen and what did the observer write,” because that is the question a sponsor will ask in month four.
Frequently asked questions
How to be emotionally intelligent at work, in practice?
How to improve self-awareness in emotional intelligence?
How to improve self-management in emotional intelligence?
Can you build emotional intelligence as an adult?
Are emotional intelligence assessments worth using?
How is emotional intelligence different from interpersonal skills?
Does emotional intelligence training improve resilience?
Where to start
Take the last three interactions at work that went worse than you expected. Write your account of each, then ask one person who was present for theirs. The size of the difference is your starting point, and it is a better baseline than any score you could buy.
If the constraint is running this across a lot of managers and keeping the observation points from quietly disappearing, book a walkthrough and we will show you how the sessions, observers and records sit together.
