New webinar on 3rd March 2026 | 9:00 am PT
In this webinar, discover how Sopact Sense revolutionizes data collection and analysis.
A DEI dashboard that connects demographics to outcomes by segment. Close the Representation Ceiling that traditional dashboards can't reach.
Training evaluation survey questions
This page is a working bank of training evaluation survey questions, organized by what each one decides. Reaction, learning, behavior, and results, with paired open-ended prompts and identical wording across pre and post waves. Every question below earns its place because it points at a decision you can act on. Examples come from workforce training, healthcare continuing education, and sales enablement cohorts. No prior background needed.
The four-level question pathway
Donald Kirkpatrick's four levels are the global standard for training evaluation. They are also the right way to organize a question bank. Each level asks a different question, calls for a different scale, and runs at a different moment. The underneath thread is the architecture that lets the four levels behave as one cascade rather than four disconnected events.
A bank without these four threads is a list of items. A bank with them is a cascade. The architectural thread is what lets four levels behave as one report.
The four-level model is from Donald and James Kirkpatrick, Kirkpatrick's Four Levels of Training Evaluation (ATD Press, 2016). The architectural thread underneath is the pattern Sopact's evaluation customers use most often: each level connects to the next through one persistent learner ID, set at intake.
Definitions
Five short definitions covering the questions readers come to this page with. Each one matches a search query a real evaluator typed last week. Read top-to-bottom for the full architecture, or jump to the one you need.
Training evaluation survey questions are the specific items asked of a learner before, during, or after a training program in order to decide whether the program produced the change it was designed to produce. A training evaluation questionnaire is the container that holds them; the unit is a question, not a survey. A survey is a container; a question is what does the work.
Most published banks list questions by survey section: opening, content, instructor, takeaway. The reorganization that matters is by decision. A reaction question feeds a decision about session design. A behavior question feeds a decision about program continuation. If a question does not feed a decision, it is asking learners to fill in time.
A pre and post training questionnaire is a pair of surveys with identical wording and identical scales, run before and after the training, on the same learner identified by a persistent ID. The pair measures change for each person, not group averages. A 5-point self-rated confidence item that runs as a 1 at week zero and a 4 at week six is a confidence delta of 3 for that learner. That number is what funders and boards now ask for.
The pair fails when the wording shifts between waves, when the scale changes, or when the same person becomes Sarah Johnson at intake and S. Johnson at follow-up because two tools split the record. The architecture comes before the wording.
A post-training survey runs at one of three moments, and each moment asks a different question. End-of-session asks reaction: did the training feel useful and relevant. Immediately after asks learning: did knowledge or skill increase. Thirty to ninety days after asks behavior: is the skill showing up in the work. The same instrument cannot serve all three; the time gap between the training and the question is part of the question.
Kirkpatrick level questions are the survey items that map to one of four measurement levels named by Donald Kirkpatrick in 1959 and codified by him and James Kirkpatrick in subsequent editions. Level 1 (reaction) measures whether learners found the training useful. Level 2 (learning) measures whether knowledge or skill was gained. Level 3 (behavior) measures whether work practice changed. Level 4 (results) measures whether organizational outcomes shifted.
The numbering is sequential because the cascade is causal: a learner who does not find the training useful (L1) is unlikely to learn (L2); a learner who learns (L2) but never applies it on the job (L3) cannot move organizational outcomes (L4). A bank that asks Level 1 questions only is asking the easiest one.
A behavior-anchored question asks about a specific work moment in a defined time window, not about a feeling. "In the last two weeks, how many client conversations used the framework from the training?" anchors a behavior. "Do you feel confident applying the framework?" does not. Confidence reads on a Likert scale; behavior reads on a count.
Behavior-anchored questions are the ones funders accept as Level 3 evidence. They are also the hardest to write because they require knowing the work. Pair every behavior-anchored count with one open-ended prompt asking for a specific example, so a low count has an explanation.
Related but different
A reaction-only survey, almost always at end-of-session. Training feedback questions tend to ask about pacing, materials, and instructor clarity. Useful for course design, insufficient for funder reporting. The "feedback" framing signals scope: feelings about the session, not change in the work.
A measurement of what the learner knows, scored against a rubric. Training assessment questions are knowledge-check or scenario-application items, not opinion items. A Level 2 instrument, not a Level 1 one. Pre-post pairs run as identical assessments at both waves; the delta is the learning measure.
A multi-level evaluation, usually covering Levels 1 through 4 with system-data linkage. The "effectiveness" framing implies behavior and results, not reaction. A six-question feedback form is not an effectiveness study.
A survey that runs before training is designed, asking what the gap is. Different from pre-training: pre-training measures the same learner who will sit in the session; needs assessment surveys the population deciding what session to run.
Six design principles
These are the rules that separate a question bank that produces evidence from one that produces a number nobody can explain. Each rule comes from a failure mode evaluators see in cohort after cohort. Apply all six and the bank survives a funder review.
01 · DECISION
If you cannot act on the result, do not ask the question.
Each item should point at a decision you will make: keep this session, change that activity, send a 90-day survey to this cohort, escalate this manager observation. Items that do not feed a decision lengthen the survey, raise drop-off, and dilute the items that matter.
Why it matters: a 30-question survey with 22 ornamental items reads to learners as procedural and to funders as unfocused.
02 · PAIRING
A scale counts. A prompt explains. The pair is the unit.
Likert scales tell you 78 percent of learners rated the session a 4 or 5. The paired open-ended prompt tells you why. Without the count, you cannot see scale. Without the explanation, the count cannot be acted on. Every closed-ended scale on the page below has a paired open-ended counterpart.
Why it matters: a 4.3 average with 200 blank explanation fields is a number, not a finding.
03 · IDENTITY
Every wave shares one ID. Set at intake. Carries to follow-up.
The same learner answers pre, post, and follow-up. Without one shared ID, matching is a manual reconciliation in a spreadsheet that costs three weeks of analyst time per cohort. With one ID, the four-level cascade compiles in hours.
Why it matters: Sarah Johnson at intake and S. Johnson at follow-up are two records in any tool that cannot bind them on collection.
04 · WORDING
Same item. Same scale. Same response set. No rewrites between waves.
A pre item that asked about confidence on a 1 to 5 scale must run as the same item on the same scale at post. A reworded item is a new item: the delta becomes uninterpretable. The instinct to "improve" the wording at post should be resisted; improvements roll into the next cohort, not the current one.
Why it matters: two slightly different items measure two slightly different things, and the comparison is no longer fair.
05 · LEVEL
Reaction reads on a Likert. Knowledge on a rubric. Behavior on a count.
A reaction question on a 5-point Likert reads well. A behavior question on the same scale does not: "How often do you apply the framework?" with response options "always" through "never" produces a feeling about frequency, not a count of moments. Match the response format to what the level can actually produce.
Why it matters: a Level 3 question with a Level 1 scale looks like behavior data and is in fact a feeling.
06 · STOPPING
Three to five items per level. Twelve to sixteen total in a pre/post pair.
Long surveys are not stronger surveys. Drop-off rises. Open-ended fields go blank. A focused bank of three to five items per Kirkpatrick level you measure produces more usable evidence than a 30-item bank that nobody finishes. The question to ask of every additional item: which decision does this feed.
Why it matters: completion rate above 80 percent on a 12-item bank beats a 35-item bank with 40 percent completion.
Six choices that decide the bank
The matrix below is a question-bank design tool. Each row is a choice the evaluator makes, almost always at intake. The "broken way" column is not a strawman: it is the workflow most teams fall into when the choice is left implicit. The "working way" is the alternative the principles in the section above ask for.
Bank is grouped by survey section: opening, content, instructor, takeaway. Each item reads in isolation. No item points at a decision the team will act on.
Bank is grouped by the decision each item feeds. Items that do not feed a decision are removed before launch.
Whether the evaluator can act on the report or only present it.
Likert scales only, plus one optional comment box at the end. Open-ended responses arrive blank or as "n/a." Numbers without explanation reach the funder.
Every closed-ended scale is paired with one open-ended counterpart placed inline. Theme extraction runs on the paired field at analysis.
Whether a low score has an explanation attached or sits as a number alone.
Pre and post run as separate forms with no shared identifier. Matching learners across waves becomes a manual reconciliation. Twenty percent of records drop in cleanup.
Persistent unique learner ID set at intake. The same ID carries through pre, post, 30-day, and 90-day waves. Matching is automatic.
Whether the four-level cascade is a default output or a heroic effort.
The post survey rewords items because someone on the team flagged them as unclear during the cohort. Pre and post are no longer comparable; the delta becomes uninterpretable.
Verbatim wording locked at design. Rewording proposals roll forward into the next cohort, not the current one. Item numbering preserved across waves.
Whether the delta is a real change or an artifact of two different questions.
Every item runs on a 5-point Likert because the template did. Behavior items return a feeling about frequency. Knowledge items return a self-report of knowledge, not a measurement of it.
Reaction on Likert. Knowledge on a scenario item with a rubric. Behavior on a frequency count of specific work moments in a defined window. Results linked to system data.
Whether each level produces the kind of evidence its name implies.
Bank grows to thirty or forty items because each stakeholder added their item. Drop-off climbs above 50 percent. Open-ended fields late in the form arrive blank.
Three to five closed-ended items per Kirkpatrick level you measure, each paired with one open-ended prompt. Twelve to sixteen items total in a pre/post pair.
Whether completion rate stays above 80 percent or the bank loses the cohort late.
Compounding effect
The first row controls the next five. A bank organized by decision will not tolerate ornamental items, will not skip the paired open-ended prompt, will not run pre and post on different IDs. A bank organized by survey section invites every later mistake. Decide row one first.
Worked example
A pharma sales enablement cohort runs a five-week training on a new product launch. Two hundred forty reps across three regions. The medical affairs team needs to know whether reps absorbed the clinical material and whether discovery calls in the field reflect the framework. The bank below is what the team built. The architecture is what made the report renewal-ready.
"We had three weeks between session five and the first quarterly review with medical affairs. The old approach would have been a 35-question SurveyMonkey form at end-of-session, plus a follow-up email at 90 days that nobody answered. We rebuilt the bank around the four levels. End-of-session was seven items. Knowledge check was eight items, run identical at intake. The 30-day behavior survey was four items: a frequency count of discovery calls using the framework, a paired open-ended for one specific call that went well, a manager-observation prompt sent separately, and a confidence item. Every learner had one ID set at intake. The dashboard for medical affairs compiled in an afternoon."
Sales enablement lead, pharma launch cohort, 30-day post wave
Sopact Sense produces
One record per rep, per wave
Pre, post, 30-day, 90-day all share the persistent learner ID. The dashboard reads as one row of four columns.
Themes extracted automatically
Intelligent Column reads the paired open-ended prompts and surfaces themes. No three-week manual coding cycle.
Manager observation joined in
The manager survey runs as a separate form on the same rep ID. Self-report and observation appear side by side.
Level 4 link to CRM data
Discovery-call counts from the CRM join on rep ID. Quantitative behavior data has a system source, not a self-report only.
Why traditional tools fail
Pre and post are separate forms
No shared ID. Matching reps across waves is a spreadsheet exercise. Twenty percent of records drop in cleanup.
Open-ended responses arrive as raw text
Coding takes weeks. By the time themes are visible, the next quarterly review is over and the cohort moved on.
Manager survey lives in another tool
Joining manager observation to rep self-report becomes a manual reconciliation that costs an analyst day per cohort.
Level 4 stays a goal
CRM data and survey data live in separate tools with separate IDs. The link to revenue or call quality cannot be made on schedule.
The cohort hit a 91 percent completion rate on the 30-day survey because the bank was twelve items, not thirty-five. The medical affairs review compiled the four-level cascade in an afternoon because the four waves shared one rep ID. The renewal conversation with the executive sponsor showed Level 3 behavior counts paired with rep voice and manager observation, all on one slide. The architecture is what produced the slide. The slide is not a visualization trick; it is the natural output when the question bank is built right at intake.
Three program contexts
The architecture is the same. The cohort size, wave timing, and Level 4 outcome differ. Three contexts below show what changes and what does not when the bank runs in different programs.
01
Single cohort, six-month program. Outcome 90 to 180 days post-program.
Two hundred to four hundred participants per cohort. The funder reports on placement rate at 90 days and retention at 180 days. The pre-training bank captures baseline confidence on five role-relevant tasks and one open-ended prompt about what the participant hopes the training will change.
What breaks: the post-training survey runs as a separate Google Form. Participant names diverge between intake (legal name) and follow-up (preferred name or nickname). At 90-day, the cohort coordinator spends six days matching records by email. Twelve percent of the 90-day responses cannot be matched to baseline.
What works: a persistent participant ID set at intake, used across all four waves. The pre-post knowledge check uses identical wording. The 90-day behavior survey asks about specific work moments at the new job ("In the last two weeks, how many times did you ...") with one paired open-ended prompt for context. Funder report compiles in two days from one record per participant.
A specific shape
Twelve items in the pre/post pair. Three Level 1 reaction items at end-of-week-one. Four Level 2 knowledge items pre and post. Three Level 3 behavior items at 90-day. One Level 4 outcome item linked to the partner-employer's payroll record on participant ID.
02
Modular CME, repeating cohorts. Outcome at 60 days.
A health system runs a continuing-education series for clinicians on a new clinical guideline. Three to five hundred clinicians per quarterly cohort, six modules each. Each module needs its own Level 1 reaction set; the series as a whole needs Level 2 and Level 3.
What breaks: Level 1 reaction surveys at the end of every module flood the inbox. Drop-off climbs from module three onward. The Level 2 knowledge check at series end has no clean link to module-level reaction, so when knowledge gain is uneven, the team cannot tell which module contributed.
What works: Level 1 collapses to three items per module, run inline at module close. One paired open-ended prompt per module asks for the one clinical situation the clinician will apply this in. Level 2 runs as a single scenario rubric pre and post. Level 3 at 60 days asks about the count of patient encounters where the guideline applied, with a paired prompt for one specific case. All waves share clinician ID.
A specific shape
Three Level 1 items per module, six modules. Eight Level 2 items as a scenario rubric. Four Level 3 items at 60-day. The series-level dashboard shows reaction by module, learning delta per clinician, and behavior frequency at 60-day on one screen.
03
Quarterly enablement, multi-region. Outcome linked to CRM data at 30 to 90 days.
A B2B sales organization runs quarterly enablement on a new product framework. One to three hundred reps per cohort. Executive sponsor wants Level 4 evidence: revenue, deal velocity, or close rate. The data lives in the CRM and the LMS.
What breaks: the post-training reaction survey lives in the LMS. The 30-day behavior survey runs in a separate tool. Manager observation is a third instrument. The CRM data is owned by sales operations, joined on rep employee ID, which differs from the LMS user ID. Joining the four sources for the executive readout takes two weeks; by then the next enablement is in flight.
What works: persistent rep ID set at enablement intake, used across post-training reaction, 30-day behavior survey, manager observation, and the CRM join. Behavior survey asks frequency counts of framework-aligned discovery calls in the last two weeks, paired with an open-ended prompt for one specific call. Executive readout compiles in an afternoon and shows the four-level cascade per rep.
A specific shape
Seven Level 1 items at end-of-session. Eight Level 2 items pre and post. Four Level 3 items at 30-day, plus a manager-observation prompt on six behaviors. Level 4 metric joined from CRM on rep ID, weekly. Executive review every two weeks, no analyst time required for compilation.
A note on tools
Each of these tools collects responses well. SurveyMonkey and Google Forms are familiar, fast, and have everything a single-wave reaction survey needs. Qualtrics adds branching logic, panels, and enterprise governance. LMS native surveys live where the training does. The architectural gap they share is identity: none of them, by default, set a persistent learner ID at intake that survives a name change between waves and joins to system data on the same key.
Sopact Sense binds the persistent learner ID to the form at intake, carries it through every wave, and lets paired closed-ended and open-ended responses live in one record per learner. Intelligent Column extracts themes from open-ended fields automatically. The four-level Kirkpatrick cascade compiles as a default output, not a heroic reconciliation effort. The tools above remain useful where their strengths fit; the join across waves and across system data is what Sopact Sense is for.
FAQ
Fourteen questions evaluators ask when designing a training evaluation bank, in the order most teams hit them. Answers mirror the JSON-LD on the page so search results and the page agree.
Q.01
Training evaluation survey questions are the specific items asked of learners before, during, or after a training program to decide whether the program produced the change it set out to produce. Good banks are organized by the decision each question feeds, not by survey section. They include closed-ended scales for counts, open-ended counterparts for explanation, and a persistent learner ID so pre and post answers connect to the same person across waves.
Q.02
Pick the level you can act on. End-of-session: ask about relevance, pace, and one thing the learner will try at work. Immediately after: ask a short knowledge check that mirrors the pre-training version verbatim. Thirty to ninety days after: ask about specific work moments where the skill applied, paired with one open-ended prompt for an example. Stop adding questions once each one points to a decision you will make with the result.
Q.03
Start with the decision the question feeds. Pair every closed-ended scale with one open-ended counterpart, so a low score has an explanation. Use verbatim wording in pre and post pairs. Anchor behavior questions to specific work moments and a defined time window. Identify every learner with a persistent ID so the same person's answers connect across waves. Match the scale to the level: reaction reads well on a 5-point Likert; behavior reads better as a frequency count.
Q.04
A pre and post training questionnaire uses identical wording and identical scales at both waves. Knowledge example: "On a scale of 1 to 5, how would you rate your understanding of [topic]?" Confidence example: "How confident are you handling [specific situation] today?" Behavior intent example: "In the next month, how often do you expect to use [skill] in your work?" The post-training version reuses the same items, plus a paired open-ended prompt asking what changed and why.
Q.05
Level 1 asks whether learners found the training useful and relevant. Sample items: relevance to current role on a 5-point scale; pace of delivery; clarity of materials; one open-ended prompt asking what one thing the learner will try at work. Avoid asking about overall satisfaction in isolation: a high score with no follow-up cannot tell you why the training landed or what to change for the next cohort.
Q.06
Level 2 measures knowledge or skill gained. Pair a pre-training knowledge check with an identical post-training version. Use scenario items, not recall items, where possible. A scenario item asks the learner what they would do in a defined situation, scored against a rubric. Match the scale across waves so the same learner's pre and post scores can be compared as a delta, not as group averages.
Q.07
Level 3 asks whether work practice changed. The strongest items count specific work moments in a defined window: "In the last two weeks, how many client conversations used the framework from the training?" Pair a manager-observation prompt with the self-report. Send the survey 30 to 90 days after the training, when behavior has had time to stabilize but is still recoverable.
Q.08
Level 4 measures organizational outcomes the training was meant to move: revenue per rep, time-to-resolution, retention, error rate, customer satisfaction. The data usually lives in the LMS, the CRM, or HR systems, not in a survey. The survey's job at Level 4 is to capture context that the system data cannot: a paired open-ended prompt asking the learner what changed in their work that the numbers reflect.
Q.09
Thirty to ninety days. Earlier than 30 days, behavior has not had time to stabilize: the learner reports intentions, not practice. Later than 90 days, the link between training and behavior weakens, and other work events crowd in. Many programs run a 30-day pulse and a 90-day deeper survey. Both must use the same persistent learner ID so the two waves connect to the same person.
Q.10
Three to five closed-ended items per Kirkpatrick level you measure, each paired with one open-ended prompt. A reaction-only survey runs five to seven items. A pre and post pair covering reaction, learning, and behavior runs twelve to sixteen items per wave. Going longer raises drop-off without adding usable evidence. Stop adding questions once each one points to a decision you will act on with the result.
Q.11
Use both, paired. A pure Likert scale training survey counts: 78 percent of learners rated the session a 4 or 5 for relevance. Open-ended responses explain: the same learners said the framework was useful, but the role-play was rushed. Without the count, you cannot see scale. Without the explanation, you cannot act on the count. Every closed-ended scale on the page below has at least one paired open-ended counterpart for that reason.
Q.12
For a sales enablement cohort: a Level 2 knowledge check on the framework taught; a Level 3 frequency count of client conversations using the framework in the last two weeks; a paired open-ended prompt asking for one specific situation where the framework changed the conversation; a Level 4 link to a CRM-sourced metric (deal velocity or close rate) using the same learner ID. The pharma cohort worked example below shows the pattern in detail.
Q.13
Both tools collect responses. The gap they leave is connection across waves. Neither tool assigns a persistent learner ID at intake that survives a name change between pre and post, so matching pre to post becomes a manual reconciliation in a spreadsheet. Both tools also separate closed-ended counts from open-ended responses across exports. Sopact Sense binds them at collection so the two are one record per learner.
Q.14
Sopact Sense assigns a persistent unique learner ID at intake. Pre, post, 30-day follow-up, and 90-day follow-up share the same ID. Closed-ended scales and paired open-ended prompts live in one form, exported as one record per learner per wave. Intelligent Column extracts themes from the open-ended prompts automatically. The four-level Kirkpatrick cascade compiles in hours, not weeks, because the underlying records connect by default.
Related guides
Six guides that connect to question-bank design at a deeper or adjacent layer. Read the cluster siblings first if you came here looking for the methods or the framework; read the others when the question becomes how to collect, link, and analyze.
Cluster sibling · methods
Training evaluation: 7 methods to measure training
The methods question. Kirkpatrick, Phillips, CIRO, Brinkerhoff matched to your funder's actual question.
→Cluster sibling · framework
Kirkpatrick model: all 4 levels of training evaluation
The framework question. The four levels in depth, plus the architectural fix for the Cascade Break.
→Sister guide
Pre and post surveys: design and matched analysis
The pair as a bridge. Identical wording, persistent ID, three-moment architecture.
→Sister guide
Longitudinal survey design
Multi-wave designs. When 30-day and 90-day pulses become a longitudinal study, and what the architecture has to hold.
→Sister guide
Open-ended questions: design and analysis
The paired prompt, in depth. How to write the open-ended counterpart that earns its place next to a Likert.
→Sister guide
Mixed-methods research
The methodology pillar. Why counts and explanations have to be designed together, not joined later.
→A working session
An hour with Unmesh, working from a bank you have already drafted or one you are about to design. We map your items to Kirkpatrick levels, identify the orphaned questions, and show what the four-level dashboard looks like when one persistent learner ID runs through every wave. No procurement decision required.
Format
60 minutes, video. Working from your draft bank or a sample we share.
What to bring
A current question bank, a funder report you need to produce, or a cohort you are about to design for.
What you leave with
A mapped bank by Kirkpatrick level, the orphaned items flagged, and a sample dashboard view with your structure.
Enterprise Analysis: Measuring Progress Toward Inclusive Workplace Culture
Women and underrepresented minorities in director+ roles increased 27% after implementing sponsorship programs and transparent promotion criteria.
Employee Resource Groups (ERGs) and monthly pulse surveys increased belonging sentiment from 68% to 82%, particularly among remote workers and new hires.
Salary analysis revealed and closed gender and ethnicity pay gaps. Transparent salary bands and annual audits ensure ongoing equity across all departments.
| Group | Entry-Level | Mid-Level | Senior | Executive |
|---|---|---|---|---|
| Women | 52% | 46% | 38% | 29% |
| People of Color | 48% | 41% | 35% | 27% |
| LGBTQ+ | 14% | 12% | 11% | 8% |
| People with Disabilities | 8% | 6% | 5% | 3% |
Create targeted retention programs for diverse mid-level managers. Implement skip-level mentoring and transparent succession planning to accelerate advancement.
Require all people managers to complete bias interruption and inclusive leadership training. Track behavioral change through 360 feedback and team belonging scores.
Establish core collaboration hours that respect global time zones. Rotate meeting times quarterly and create asynchronous decision-making processes for more inclusive participation.
Audit all tools, physical spaces, and processes for accessibility. Partner with disability advocates to implement accommodations proactively rather than reactively.
TechCorp has made measurable progress toward diversity, equity, and inclusion goals through transparent metrics, continuous feedback, and targeted interventions. Representation in leadership increased 27%, belonging scores rose 14 points, and retention of diverse talent reached 91%. However, data reveals persistent challenges: diverse talent advancement slows at mid-level, microaggressions continue despite training, and meeting culture excludes some voices. The path forward requires addressing pipeline leakage through sponsorship expansion, reimagining inclusive leadership expectations, and creating genuinely accessible and flexible work structures. With Sopact Sense's Intelligent Suite, DEI becomes a continuous learning system—measuring impact in real time, surfacing barriers as they emerge, and connecting employee voice directly to organizational action.
Effective DEI dashboards move beyond compliance metrics to measure real inclusion—combining representation data with belonging sentiment, promotion equity, and employee voice. Below is a breakdown of each component, explaining what it measures and how Sopact Sense automates continuous DEI tracking.
Provide leadership with immediate proof of DEI progress. Three core metrics show representation, inclusion sentiment, and retention—the foundation of workplace equity.
Intelligent Column aggregates HRIS demographic data with pulse survey responses. Stats update automatically as new employees join and quarterly surveys close.
Connect metrics to why they changed. Each insight explains which interventions worked—sponsorship programs, ERGs, pay equity audits—and proves ROI on DEI investments.
Intelligent Grid correlates demographic shifts with program participation data. Plain English instructions: "Show promotion rate changes for employees with sponsors vs. without."
Balance quantitative metrics with lived experience. Shows what's working from employees' perspectives and where systemic barriers persist—critical for authentic DEI work.
Intelligent Cell extracts themes from open-ended feedback. AI categorizes comments by sentiment and topic (sponsorship, microaggressions, flexibility) in minutes.
Visualize where representation gaps exist across the organization. Proportional bars show actual percentages—making disparities immediately visible.
Intelligent Column calculates representation by level automatically. Links HRIS demographic data with org chart hierarchy—no manual Excel pivots.
Reveal pipeline leakage patterns. Color-coded metrics show where specific groups advance equitably (green) and where barriers emerge (yellow/red).
Intelligent Grid cross-tabulates demographic data by job level. Auto-applies color thresholds based on representation goals—flags concerning patterns instantly.
Turn insights into action. Each recommendation addresses a specific barrier surfaced in the data—pipeline leakage, bias training gaps, meeting culture, accessibility.
Intelligent Grid synthesizes patterns from qualitative feedback and quantitative gaps. Example: "If retention drops 15%+ at mid-level, recommend pipeline interventions."
Most organizations collect mountains of DEI data—demographic surveys, engagement scores, hiring metrics, retention rates—but struggle to turn those numbers into action. Teams spend weeks building dashboards that show what happened, not why it matters or what to do next. Meanwhile, leadership asks for proof of progress, employees want transparency, and compliance requirements keep growing. The result: DEI becomes a reporting exercise rather than a transformation strategy, and real equity gets lost in spreadsheets.
Dashboards show demographic breakdowns and percentages, but can't explain why gaps exist, what barriers employees face, or which interventions actually work. Leadership sees "representation improved 3%" but doesn't know if that's progress or tokenism.
HRIS holds demographics, engagement surveys capture sentiment, exit interviews reveal departure reasons, promotion data sits in spreadsheets. No single view connects hiring → experience → advancement → retention for different identity groups.
After presenting a quarterly DEI report, leaders ask "what should we do differently?" but the dashboard has no answers. There's no way to test whether mentorship programs improve retention or if unconscious bias training shifts hiring patterns.
Workforce demographics by role level, department, location, tenure
Identify where representation breaks down across the employee lifecycle
Analyze representation patterns: - Compare workforce demographics vs market availability - Show breakdown by seniority (entry → leadership) - Identify departments with largest gaps - Track change over time (YoY comparison) Surface insight: "Women represent 45% of entry-level but only 18% of VP+ roles"
Grid generates multi-dimensional view; Column aggregates by level; Dashboard reveals where pipeline breaks; Actionable targets emerge automatically
Engagement survey responses, demographic data, open-ended feedback
Understand which groups feel included vs isolated, and why
Calculate belonging scores by identity group: - Aggregate engagement questions (voice heard, authenticity, psychological safety) - Compare across race, gender, age, tenure - Extract themes from open-ended responses - Identify correlation with manager/team/location Return: "LGBTQ+ employees score 23% lower on 'authenticity at work' primarily due to..."
Column shows belonging gap; Cell extracts "why" from qualitative data; Leadership sees both metric + root cause; Interventions target actual barriers
Promotion rates, performance ratings, tenure, demographics
Detect bias in advancement opportunities controlling for performance
Compare promotion rates by identity: - Control for tenure, performance rating, department - Calculate promotion velocity (time to next level) - Identify managers with largest disparities - Statistical significance testing Generate: "Among high performers, white employees promoted 1.4x faster than Black employees"
Grid reveals patterns across cohorts; Row summarizes individual equity; Dashboard flags potential bias; HR investigates specific managers/departments
Exit interview transcripts, departure reasons, demographics
Understand why different identity groups leave at different rates
Extract departure themes by identity: - Categorize reasons (growth, culture, compensation, manager, work-life, bias/discrimination) - Compare theme frequency across demographics - Include direct quotes illustrating each theme - Identify preventable vs structural exits Return patterns: "Women cite 'lack of advancement' 3x more than men"
Cell codes each exit interview; Column aggregates themes by group; Dashboard shows why retention differs; Retention strategies target actual drivers
Compensation data, job levels, performance ratings, demographics
Identify unexplained pay gaps controlling for legitimate factors
Analyze pay equity by identity: - Compare compensation controlling for role, level, tenure, performance, location - Calculate median/mean gaps across demographics - Flag individuals with unexplained variances >10% - Estimate cost to close gaps Generate: "Median pay gap of 8% ($12K) for women in engineering roles; $2.4M to remediate"
Grid shows gaps across job families; Column calculates remediation costs; Dashboard prioritizes correction; Compensation team has action plan
Applicant demographics, interview pass rates, offer acceptance
Find where diverse candidates drop out of hiring process
Track hiring funnel equity: - Compare pass rates by stage (screen → phone → onsite → offer → accept) - Calculate drop-off disparities by identity - Identify interviewers with largest gaps - Compare source diversity (referral vs posting) Reveal: "Black candidates pass phone screen at same rate but 40% less likely to pass onsite"
Column shows stage-by-stage equity; Grid reveals where bias occurs; Dashboard pinpoints interview training needs; Sourcing strategy adjusts
Training participation, mentorship assignments, high-visibility projects
Ensure development opportunities distributed equitably
Compare development access by identity: - Training hours, conference attendance, certifications - Mentorship/sponsorship assignment rates - High-visibility project participation - Stretch role opportunities Calculate equity score: "Asian employees receive 30% fewer sponsorship opportunities despite similar performance"
Grid shows opportunity distribution; Row flags underinvested talent; Dashboard guides L&D resource allocation; Managers get equitable assignment guidance
Manager-level metrics: team composition, engagement, promotion, retention
Hold people leaders accountable for equity outcomes on their teams
Generate manager equity scorecard: - Team representation vs company benchmark - Engagement score disparities by identity - Promotion velocity differences - Retention rate gaps - Performance rating distribution equity Flag managers in bottom quartile: "Manager X promotes white reports 2x faster than others with same ratings"
Row creates individual manager scorecard; Grid ranks all leaders; Dashboard guides coaching priorities; Equity becomes performance metric
All DEI metrics updating continuously as HR actions occur
Track progress toward equity goals in real-time, not quarterly
Create living DEI dashboard: - Representation progress vs annual targets - Belonging score trends (monthly pulse) - Promotion equity tracking (updated with each cycle) - Pay gap status (refreshed quarterly) - Initiative effectiveness (A/B testing ERG programs) Share with leadership, board, employees (filtered views)
Grid powers continuous dashboard; Leadership sees current status anytime; Board gets transparency; Employees trust progress; DEI shifts from annual report to ongoing transformation
