New webinar on 3rd March 2026 | 9:00 am PT
In this webinar, discover how Sopact Sense revolutionizes data collection and analysis.
Training evaluation software with 10 must-haves for measuring skills applied, confidence sustained, and outcomes that last — delivered in weeks, not months.
Kirkpatrick’s four levels are not the problem. The data architecture beneath them is. Reaction lives in one tool. Learning lives in another. Behavior change lives in a third. Results live in a fourth. Nothing connects.
Sopact Sense issues a persistent learner ID at enrollment. Every instrument inherits it automatically, from baseline assessment through 12-month follow-up. The cascade closes because the ID never breaks.
The Kirkpatrick cascade, with one ID throughline
What it is
Training evaluation is the systematic process of measuring whether a training program produced what it set out to produce. A complete evaluation answers four questions: did participants react well to the program, did they actually learn, did they apply the skills back on the job, and did the organization see results (productivity, retention, revenue, error reduction). The first two questions are easy to answer in the room. The last two are where most programs stall.
Training evaluation splits into two halves: the in-the-room half (reaction and learning, measured the day of training) and the post-program half (behavior change and organizational results, measured 30 to 180 days out). Most programs measure the in-the-room half well. Satisfaction surveys and quiz scores are easy to collect on the day of training. The post-program half is where the cascade breaks. Connecting a 90-day follow-up response to the same learner's intake baseline requires one ID across every tool. Without it, Levels 3 and 4 are structurally unreachable.
The standard frameworks are the Kirkpatrick model (the canonical four-level structure used globally), Phillips ROI Methodology (which adds a fifth level converting outcomes to financial value), and the broader program evaluation discipline (formative methods that surface mid-program problems, plus summative methods that prove impact at the end). All three converge on one architectural requirement: every reported outcome traces back to an individual learner whose record stayed connected from intake through follow-up. Below: the seven methods, and how the architecture closes Levels 3 and 4 by default.
Half one
Reaction and learning, measured the day of training. Most programs handle this half well.
Half two
Behavior change and organizational results, measured 30 to 180 days out. Where the cascade breaks under disconnected tools.
Standard frameworks
The canonical four-level structure used globally: reaction, learning, behavior, results. Most training evaluation conversations begin and end here.
Extends Kirkpatrick with a fifth level converting training outcomes to financial value. The framework when funder accountability requires financial justification.
The broader evaluation discipline. Formative methods surface mid-program problems. Summative methods prove impact at the end. CIRO and Brinkerhoff Success Case Method fit here.
The Cascade Break
Most programs run all four Kirkpatrick levels. They just run them in four different tools, with four different ID formats, on four different timelines. The model is not broken. The data architecture beneath it is.
Level 1 satisfaction averages calculate on Monday. Level 2 pre-post deltas calculate three weeks later on a different participant list. Level 3 follow-up surveys go out to whoever the LMS last had an email for. Level 4 business outcomes live in HR or finance and never link back to who actually completed the training.
Without a persistent learner ID
Three days of analyst time per cycle reconciling four lists by hand.
The retrofit is matching by email. Email matching breaks when participants use work email at intake and personal email at follow-up. It breaks when names get misspelled. It breaks when participants change roles or programs. The 90-day response that would have proved behavior change lands in a row that does not link to the pre-program baseline of the same person.
The fix is not a smarter spreadsheet match. The fix is upstream: issue the ID once, at first contact, and let every instrument inherit it.
The Cascade Break is the structural failure where Kirkpatrick’s four levels get measured as four disconnected events. The model is not broken. The architecture beneath it is.
Sopact · Training Intelligence thesis
How it actually closes
One persistent learner ID issued at first contact. Every subsequent instrument inherits it. The record gets richer at each stage. By month six, the same row contains the full Kirkpatrick cascade for the same person, computed automatically.
Stage 1
Enrollment
Day 0
Stage 2
Baseline + post
Program week 0 to end
Stage 3
Reaction
End of session
Stage 4
Behavior
30 to 90 days post
Stage 5
Results
6 to 12 months
ID issued. Demographics + cohort tag stored.
Identical items, identical rubric. Per-person delta computed.
Open-ended scored against rubric at submit.
Personalized link tied to record. Manager parallel-form available.
Same row. Full cascade computable.
At month six, the funder report writes itself. The cohort end to shareable report cycle runs in hours, not weeks.
The substrate
Every layer works because every record carries the same persistent learner ID. Without that, longitudinal analysis is manual deduplication. With it, the analysis is a default output of collection itself.
01 · Cell
Intelligent Cell
Collection time · per response
Single-field analysis. Applied to one open-text answer or one file upload, with a rubric defined by the program owner. The score lands in a column inside the same record, not a separate dashboard.
In training
A 30-day follow-up answer to “describe a situation where you applied [skill]” gets scored against a behavioral rubric the moment it is submitted. The score and the reasoning land in the same row as that learner’s pre-program baseline.
02 · Row
Intelligent Row
Collection time · per learner
Multi-field analysis per record. Combines several Cells, structured fields, and uploaded files into a coherent learner view. The reviewer or program manager sees one consolidated profile, not five tabs.
In training
A learner’s baseline rubric, post-program score, reaction notes, and 30-day behavior evidence get rolled into a one-page learner brief. Manager observation, when collected, is folded in automatically.
03 · Column
Intelligent Column
Reporting time · cross-cohort
Cross-record patterns across all responses to one or more fields. Theme extraction, sentiment tracking, indicator computation across the full dataset. Replaces the manual qualitative coding workflow.
In training
Theme extraction across every cohort’s answer to “what is preventing you from applying this on the job?” surfaces the three structural barriers in two days, not three months.
04 · Grid
Intelligent Grid
Reporting time · full dataset
Full dataset analysis across every record and every field. Funder reports, cohort comparison, multi-program rollups. The two weeks before a board meeting compress into hours.
In training
One report aggregating all four Kirkpatrick levels across every active cohort, broken down by funder, geography, or demographic facet. Funder-ready output without an analyst week.
Where teams use it
Training Intelligence is the architecture beneath every page in this section. Pick the page closest to your work and go deeper.
The framework most training programs use, and the structural reason most stop at Level 2. The deep dive on what each level measures, how to instrument it, and how to keep the cascade from breaking.
Measuring knowledge and skill gain per learner. Pre-post pairing with identical items and identical rubrics, scored automatically.
SKILL TRANSFER · 30 TO 90 DAYSDid skills transfer to the job? Behavioral evidence at 30, 60, and 90 days, scored against the same rubric used at baseline.
QUESTION BANK · FOUR LEVELSPre-post matched questions across all four Kirkpatrick levels. Each question paired with the decision it feeds and the open-ended counterpart that explains it.
Cohort health, milestone completion, at-risk learners. The operational view that staff act on while the program is still running.
FUNDER · BOARDThe decision-ready output funders and boards actually use. Quantitative deltas, qualitative themes, and the story that connects them.
M&E · FRAMEWORK LEVELWhen a single program scales to a portfolio of programs. Indicators connected to instruments, instruments connected to records, records connected to outcomes.
What’s different
Most training stacks include both an LMS and a survey tool. Both produce useful data. Neither carries a learner ID across the gap, and neither analyzes open-ended responses against a rubric. The cascade closes when one record holds all four levels.
Capability
LMS platforms
Cornerstone, Docebo, Absorb
Survey tools
SurveyMonkey, Qualtrics
Sopact Sense
Training Intelligence
Persistent learner ID across instruments
ID exists inside the LMS. Breaks at every export to a survey tool, follow-up form, or HR system.
Embedded data and contact-list hacks. Breaks when participants use different addresses.
Issued at first contact. Inherited by every instrument from baseline through follow-up. Survives email changes and name corrections.
Open-ended response scoring against a rubric
Open-ended fields exist but are not analyzed. Manual review or external tooling required.
Qualtrics Text iQ surfaces aggregate themes. Output sits in a separate dashboard, not in the same record.
Rubric defined by the program owner. Score and reasoning land in a column inside the same record at the moment of submission.
Level 3 and Level 4 connected to Level 1 and Level 2
Course completion and quiz scores. Behavior change and business results live in other systems with no shared ID.
Reaction surveys are the home turf. Anything beyond requires manual stitching to other data.
All four Kirkpatrick levels in one persistent record. Pre-post deltas, behavior change, and business outcomes computable without reconciliation.
Cohort end to shareable funder report
Export, clean, merge with survey data, manually code open-ends, build report. Typically 4 to 6 weeks per cycle.
Same workflow. Different starting point. The cleanup tax is the same regardless of which tool collected what.
Funder-ready output drafts itself once the final follow-up arrives. Program staff edit narrative, do not assemble data.
Mid-program detection of at-risk learners
Login frequency, course completion. No qualitative signal from learner’s own words.
Mid-program pulses possible but require manual review of every response to spot risk signals.
Open-ended pulse responses scored at submit. Risk signals surface in real time. Theme extraction shows what the cohort is wrestling with this week.
Who runs it
Three customers. Three different training contexts. Same architecture: persistent learner ID at first contact, every instrument inherits it, the cascade closes by month six.
Lantern Network
Mentorship · internships · careers
373
Mentees served. 80 securing internships. 82 percent success rate.
The Lantern Network runs a three-pillar program: Inspire (Streaming Stories), Guide (mentorship matching and development), and Propel (paid internships at industry partners). Sopact Sense issues the persistent learner ID at mentee intake. Every subsequent instrument inherits it: mentor matching, mentorship session check-ins, internship application, internship outcome, alumni follow-up. The full participant journey lives in one record.
Connect with mentors. Land internships. Build your venture. Join 300-plus students who’ve transformed ambition into achievement.
The Lantern Network · program mission
The King Center
Seven training programs · replaced Qualtrics
10,000+
Stakeholder voices collected and analyzed across seven programs in twelve cities.
The Martin Luther King, Jr. Center for Nonviolent Social Change ran impactful training programs (Beloved Community Leadership Academy, Nonviolence365 Education and Training, Better Together for faith leaders) but qualitative feedback sat untouched because no one on the team had analyst capacity. Sopact replaced Qualtrics. Pre and post surveys now connect across cities. Open-ended feedback gets analyzed in real time, enabling dynamic discussion during the training itself.
Gathering open-ended feedback was always part of our routine, yet it remained untouched until now. Discovering automated insights was a game-changer, enabling real-time analysis and dynamic discussions during trainings.
Kelisha B. Graves, Ed.D. · Chief Research, Education, and Programs Officer
EnCorps
STEM tutoring · equity
20+ hours = 22-point gain
Tutoring hours optimal for measurable math score increase, validated across cohorts.
EnCorps prepares STEM professionals to tutor middle-school students from underserved partner schools. Before Sopact, student data lived in Salesforce, surveys lived in SurveyMonkey, and outcome tracking lived in spreadsheets. Sopact unified the sources, mapped indicators to strategic goals, and connected tutor-student relationships to diagnostic score change. The result: a defensible answer to the question every funder asks. Does it work, and how much of it works?
Quantifying impact in STEM education isn’t just a goal, it’s necessary. With the right data, we can shape the future of learning.
Kathleen Kostrzewa · Director, Strategy, Learning and Impact
FAQ
Eight questions program managers, evaluators, and L&D leads ask in their first conversation. Visible Q&A so search engines and AI assistants can index every answer.
The Kirkpatrick model is a four-level framework for evaluating training: Reaction (Level 1), Learning (Level 2), Behavior change on the job (Level 3), and organizational Results (Level 4). The model itself is sound. What breaks in practice is the data architecture beneath it. Reaction lives in a survey tool. Learning lives in an LMS. Behavior follow-up goes to whichever email the program last had. Results live in HR or finance. Without a persistent learner ID across all four, the cascade is four disconnected events instead of one connected story.
Training assessment measures an individual learner’s knowledge or skill at a specific point: a quiz, a rubric, a competency test. Training evaluation is broader. It evaluates whether the program produced the intended outcomes across the cohort: did learners apply the skills, did the work change, did the funder’s target outcome shift. Assessment is a building block of evaluation, not a substitute for it. The two questions need different instruments and different timing, but they need to be connected to the same learner record.
It can, if the rubric is the program owner’s, not the AI’s. Sopact’s Intelligent Cell scores an open-ended response against a rubric you define: the criteria, the weights, the scale. The score lands in a column inside the same record alongside the score reasoning, so a reviewer can audit any individual judgment. The output is traceable, auditable, and consistent across responses, which is the property manual coding cannot guarantee.
A persistent learner ID is issued at the first point of contact, typically enrollment. Every subsequent instrument (baseline assessment, post-program survey, 30 and 90 day follow-up, manager observation, alumni outcome) inherits that ID automatically. Per-person delta calculation is then a query, not a project. This survives email changes, name corrections, and multi-program participation. The link is built upstream at issue time, not downstream at match time.
Level 2 measures whether the learner acquired the knowledge or skill, typically through paired pre-post assessment using identical items and identical rubrics, with the delta computed per individual. Level 3 measures whether they applied it on the job, typically 30 to 90 days after the program through behavioral observation or self-report against the same competency anchors. Level 2 happens during the program. Level 3 happens after the program. Both require the same persistent learner ID to link the two measurements.
No. Sopact Sense is the evaluation and analysis layer. It complements an LMS rather than replacing it. The LMS handles content delivery, course completion, and quiz scoring. Sopact handles the persistent learner record, open-ended scoring, follow-up cycles, qualitative analysis, and funder reporting. Most customers run the two side by side. The integration point is the persistent ID: Sopact issues it, and downstream tooling references it.
The evaluation follows the program timeline, not a fixed duration. Level 1 runs at session end. Level 2 pre-post spans program start to program end. Level 3 measurement runs at 30, 60, or 90 days post-program. Level 4 business results usually require 6 to 12 months of post-program observation. The reporting cycle from cohort end to shareable funder report should take hours, not the 4 to 6 weeks typical of consultant-assembled reports. The architecture determines the speed, not the methodology.
For training programs that need longitudinal measurement and qualitative analysis, yes. The King Center moved off Qualtrics and now runs pre and post training surveys for seven programs across twelve cities through Sopact. The reasons are the persistent learner ID across instruments and the rubric-driven scoring of open-ended responses at submit time, both of which require custom embedded-data hacks in Qualtrics and break frequently in practice. For one-off surveys with no longitudinal need, either tool will do.
Close the cascade
A 60-minute working session. Bring a current cohort or a planned program. We map the four levels, the persistent ID, and the report you need at month six. By the end, you have a working evaluation architecture and a path to running it next cycle.
