Before you trust a number on a slide, ask: are people answering consistently, and is the survey actually measuring what we think it is? Those two questions are the whole job of reliability and validity, and they're the difference between a finding you can defend and a guess wearing a percentage sign.
The trouble with unreliable or invalid data isn't that it looks suspicious. It looks fine. It moves through the deck, into the strategy, into next quarter's budget. Only later does anyone realize the conclusion was built on noise.
An HR team announces a six-point lift in employee engagement. The CEO praises the wellness initiative. Six months later, the best engineers leave anyway.
The survey was measuring how comfortable people felt being honest, not how engaged they actually were.
A marketing team reports a 14-point NPS jump after a campaign. Leadership doubles the budget. Next quarter NPS drops 19 points, with no campaign change.
The instrument bounced because the question wording was tweaked between waves and the sample mix was different. Nothing actually changed in the brand.
A program submits course-evaluation data to its accreditor. The reviewer asks a single question: "What's the reliability of this instrument?" Nobody knows.
The submission goes back for another year of data collection, this time with proper psychometric reporting.
Two ideas. Each one answers a different question about whether your numbers can carry the weight you're about to put on them.
= consistency.
A reliable survey gives you the same answer when nothing real has changed. If you ask the same person the same questions twice in a calm week, the answers should look alike. If two questions are supposed to measure the same thing, they should agree.
= measuring the right thing.
A valid survey actually measures what it claims to measure. A "leadership effectiveness" score should reflect leadership effectiveness, not how nice the manager seems on a good Tuesday, or how scared the team is to be honest.
Each target shows where the answers land when you ask the same question many times. The bullseye is the truth.
You don't need to memorize the names. The platform runs these in the background and tells you, in plain language, where your data is strong and where it isn't.
Are answers consistent enough to trust?
If five questions are supposed to measure the same idea, do they actually agree with each other? A score above 0.70 says yes; below 0.50 says you're measuring different things you've been treating as one.
When the same person answers a week later (and nothing real has changed), do they answer roughly the same? If not, the instrument is reading mood, not the thing you wanted to measure.
When two reviewers grade the same open-ended response, do they agree? If not, your "themes" are really one analyst's opinion dressed up as data.
Are we measuring the right thing?
Do the questions actually cover the topic, not just the easy parts of it? A "wellbeing" survey that only asks about sleep is missing most of wellbeing.
Are we really measuring "trust in leadership," or just "I had a good week"? We test this by checking that scores move with things they should move with, and don't move with things they shouldn't.
Do survey scores predict the outcome you actually care about, like turnover, repurchase, or course completion? If a high score doesn't connect to a real-world result, the measure isn't doing its job.
Reliability and validity aren't abstract. They're the difference between a finding that survives the next meeting and one that quietly falls apart.
Reviewers don't ask "is this surprising?" They ask "is this measurable?" If you can't report Cronbach's α, McDonald's ω, and convergent validity evidence, the manuscript stalls.
Bad data here looks like: a scale with α = 0.41 that you only catch at the analysis stage, after 18 months of fieldwork.
A tracker is only useful if the number means the same thing this quarter as it did last quarter. Different wording, different sample, different mode, and "we're up 12 points" turns into "we changed the ruler."
Bad data here looks like: a brand tracker that bounces 18 points wave to wave, until leadership stops reading it altogether.
If a pulse score swings 1.2 points between Tuesday and Thursday, no manager will act on it. If a 9/10 engagement reading comes from a team that just lost three engineers, you're measuring social desirability, not engagement.
Bad data here looks like: a "we're doing great" report two weeks before a wave of resignations.
Accreditation reviewers will ask one question about every instrument: what's its reliability? If you can't answer, the cycle restarts. And section-level evaluations swing wildly when the instrument is noisy, making instructor decisions feel arbitrary.
Bad data here looks like: the same instructor scoring 4.7 in one section and 3.2 in another, with no way to tell what's signal and what's noise.
ReliCheck does the reliability and validity checks in the background, and shows you, in plain language, where your data is strong and where it isn't.