Impact measurement software that joins surveys, case notes, financials, and stories on one participant ID. Built for evidence, not effort. Since 2014
Survey is yesterday. The 5% that funders see is the survey. The 95% that holds the truth lives in case notes, audio reflections, financials, and stories — and almost no platform joins them on one participant.
One stable participant ID across application, case note, survey, story, and ledger. Every outcome claim cites the source record. Built for the M&E lead, the impact manager, and the consultant who places the tool — not for the funder slide deck.
By Unmesh Sheth · Founder & CEO, Sopact · Updated May 2026
The shape of an impact measurement cycle
The reframe
Impact measurement, in its traditional form, is dead. It was always the funder's ask and the grantee's burden. The survey returned 5% of the context — the case worker's notes, the audio reflection, the financial document, the parent voice held the other 95%, and almost no platform brought them together under one record per person.
The dominant model since 2014: a vendor with consultants would help the nonprofit "build measurement capacity." Sold as a long-term partnership for outcomes maturity.
How it became a liability. The consultants left. The capacity rarely transferred. The report became the deliverable instead of the evidence behind it. Theory-of-change diagrams accumulated; the data never connected to the diagram. By the next funder cycle, the work was rebuilt from scratch.
Case management platforms — Bonterra Apricot, Efforts to Outcomes, SureImpact, Salesforce Nonprofit Cloud — sold "track every interaction." Click attended, click didn't, document the encounter.
How it became a liability. The system counted attendance and documentation, not movement on outcomes. The case notes that held the actual evidence sat in narrative fields, unsearchable, never reaching the dashboard. A board could ask "did they improve?" and the answer was always a slide and a story, never one connected record.
Sopact's one bet
One stable participant ID across application, case note, survey, story, audio reflection, and ledger entry. The qualitative material the dashboard usually misses surfaces alongside the survey signal. Every roll-up cites the source record. The connected record is the product — every view sits on top of it.
The deeper problem
Could you prompt your way to a one-cohort outcome summary with a foundation model? Yes. Run it twice and you get two different answers — this is how the models work, not a bug to fix. As the dataset grows, hallucination climbs: on Vectara's 2026 enterprise-document benchmark, every major reasoning model — GPT-5, Claude Sonnet 4.6, Grok-4, Gemini-3 Pro — fabricated information in more than 10% of summaries. The fix is not a better prompt. It is structural — and you cannot bolt it onto Apricot, ETO, SureImpact, Blackbaud Outcomes, or Salesforce NPSP. AI on top of old systems is the loudest pattern in the category right now — and the one that will quietly stop working in eighteen months.
The student enrolled at age 7 is the same record at age 17. The grantee at year 1 is the same record at year 5. The participant ID does not break across program redesigns, staff turnover, schema changes, or vendor migrations.
Before-and-after survey scores — the questions your program already uses for confidence, well-being, or skill — sit on the same record as case notes, audio reflections, exit interview transcripts, and program cost. The reasoning layer reads the qualitative material as evidence, not as decoration.
Every figure in a funder report, every theme in a board view, every outcome roll-up cites the specific case note, transcript, survey response, or ledger entry it came from. Auditors and program officers can verify in one click instead of one quarter.
This has been Sopact's day job since 2014 — before the generative AI category had a name, and well before the current wave of "AI on top of spreadsheets" tools showed up.
Definition
The category has a long history of misalignment. Funders ask for impact measurement because boards and regulators require it. Grantees treat it as a burden because the tooling was built to satisfy the ask, not to support learning. The result, over a decade, is a sector full of theory-of-change diagrams that no data connects to, and dashboards that report outputs (attended, completed, served) as if they were outcomes (improved, employed, retained).
The shift in the last two years is that the AI reasoning layer can finally read the qualitative material — the 95% of context that always lived in case notes, transcripts, and stories — and tie it to a participant identity. Survey is yesterday. The reasoning layer reads the case note and tells you which of the eleven participants in this week's pulse is drifting, and which case notes hold the explanation. That capability does not exist on top of a workflow-era platform; it has to be designed into the data architecture from the beginning.
What that looks like in practice: a youth services nonprofit reads outcome movement for 1,500 students across thirty caseworkers and three regions, joined to the case notes for any individual. A workforce program reads cohort-2024 wage outcomes at 18 months alongside the exit interview transcripts that explain the dropouts. A foundation reads grantee outcomes across forty-two grantee partners with citations to the original site-visit notes. A community health program reads before-and-after well-being scores across pre-mid-post with the case-worker narrative that explains the outliers. Same connected record per person, different programs.
Frame 03 · How Sopact fits the existing stack
Most teams already run a CRM, a case management system, an intake form tool, an accounting platform, and a reporting layer. Sopact Sense sits in the middle and holds one connected record per person — contact records flow in, outcome evidence flows out. The vendors on either side keep doing their job.
Contact records, intake forms, and case management entries flow in at enrollment.
One participant ID, longitudinal record growth, numbers and stories on the same row, every figure cited to its source.
Evidence-backed roll-ups, cost-per-outcome figures, and funder-ready reports leave Sense at the moment of use.
Frame 04 · The Tuesday question, not the year-end dashboard
The legacy stack was tuned for the year-end report. The impact manager's real job is the Tuesday question — the program officer asking what happened with this cohort, the board chair asking what the case notes say, the funder asking which outcome moved. Five of those questions, shown three ways.
of an impact manager's daily and regular work — cohort review, qualitative coding, funder reporting, evidence assembly, longitudinal pulse — runs on one connected record, not on five disconnected tools and a hired consultant.
Sixty minutes with Unmesh. Bring a real cohort, a real funder report, or a real Tuesday question. Walk away with a path that does not depend on a foundation model returning the same answer twice.
The anatomy of a roll-up
Every figure on a funder report descends from a single response written by a single participant. Sopact names the four layers so the question — and the citation trail — never gets lost.
One participant. One survey question, one case note, one transcript line, one ledger entry. The atomic unit. Coded against the instrument or the theme dictionary, and cited as the source of any later claim.
e.g. Participant 0418 · pre-program well-being question Q1 · score 3 · "I feel down most days this week."
One participant across all sources and stages — application form, baseline survey, six case notes, mid-program pulse, audio reflection, post survey, employment outcome. The longitudinal view of a single human, joined on the participant ID.
e.g. Participant 0418 · year-2 enrollment through year-13 graduation · 47 records across 6 sources.
One outcome across all participants in a defined cohort — pre→post movement on the primary instrument, theme distribution in the case notes, retention curve. The view the program officer asks for on a Tuesday.
e.g. 2024 youth-services cohort · n=47 · confidence pre→post +2.3 points · 12 named themes · 83% retention.
All cohorts, all programs, all outcomes — for the board, the funder, the regulator. Aligned to whatever reporting framework your funders ask for, cost-per-outcome computed from accounting. Every figure descends from a Cell and cites its source. Verifiable, defensible, ready for the audit.
e.g. 6 programs · 14 cohorts · 1,500 participants · 7 outcome measures · $48 cost-per-outcome · citation trail to source records.
From raw artifact to evidence
Survey responses, case-note PDFs, audio reflections, and quarterly pulse data arrive in different shapes from different sources. Sense makes each one queryable and citable on the participant ID.
Survey: 2024 youth-services intake Submitted: 2024-09-14 11:08 Response ID: srv_8821 Email: family contact on file Name: [participant] Q1 · "How confident do you feel about your ability to manage tricky moments at school this term?" Free text: "i dont know it depends i guess sometimes its ok sometimes not" Q2 · Likert 1–5 (confidence) Answer: 2 Q3 · "What is one thing that would help most this term?" Free text: "if my reading was better"
participant_id: P-0418
stage: baseline · 2024-09-14
program: youth-services-northland
caseworker: cw_07
confidence_baseline:
instrument: custom_5pt
raw_score: 2
text_evidence:
"i dont know it depends i guess
sometimes its ok sometimes not"
qualitative_signal: ambivalence
flagged_for_caseworker: true
stated_need:
category: literacy
raw: "if my reading was better"
bound_to_outcome: literacy_skill
citations:
- srv_8821 · Q1 · Q2 · Q3
ready_for_join: true
Case note — 2025-02-11 Caseworker: J. Davies Student: [name on file] Met before homeroom. Brought spare breakfast. Student was withdrawn, mentioned not sleeping. We talked about the maths test on Friday. Student said they panic looking at the page but did the practice with me. Read three pages out loud — improved from last term. Will check in Wed before period 2. Family contact later this week.
participant_id: P-0418
note_id: cn_4471 · 2025-02-11
caseworker: cw_07
stage: in-program week 22
themes_extracted:
- food_security_intervention
- sleep_disturbance · new_signal
- test_anxiety · recurring
- literacy_progress · positive
instrument_signal:
confidence: marginal_positive
source_evidence: "read three pages
out loud — improved from last term"
drift_signal:
sleep_disturbance → flag_caseworker
for follow-up before next pulse
citations:
- cn_4471 · paragraph 2 · 3
Audio reflection — annual 2025-09-02 · 8 minutes 14 seconds Participants: caseworker + student [transcript excerpt] "So last year you said the exams were the hardest. How did this term feel?" "Different. I still got nervous but I had the breathing thing. And I knew the maths the second time. The reading is still hard but not panic hard anymore." "And in class?" "I put my hand up twice this term. Once was wrong but that was okay."
participant_id: P-0418
reflection_id: ar_0214 · 2025-09-02
duration_seconds: 494
stage: year-2 longitudinal
transcript_themes:
- test_anxiety · reduced
- coping_skill_acquired · breathing
- literacy · still_difficult_but_calmer
- classroom_engagement · positive
evidence: "i put my hand up twice"
longitudinal_signal:
baseline_2024: confidence 2/5
reflection_2025: confidence 3.5/5
qualitative_movement: visible
primary_driver: coping_skill +
reduced_panic_response
ready_for_cohort_roll_up: true
citations: ar_0214 · 02:14–04:08
Quarterly pulse · Q3 2025 2024 youth-services cohort · n=47 Survey responses: 39 received Case-notes parsed: 41 (last 90 days) Audio reflections: 22 received Spreadsheet exports: - pulse_survey_q3.csv (39 rows) - case_notes_q3.zip (41 PDFs) - reflections_q3.zip (22 m4a) - attendance_q3.xlsx (47 rows) - quickbooks_program.csv Question from program officer: "Which participants are at risk?"
cohort: ys-northland-2024
n: 47 · reachable_q3: 41
primary_outcome: confidence + literacy
cohort_movement_q3:
confidence_avg_Δ: +0.8 points
literacy_avg_Δ: +0.4 points
retention: 87%
at_risk_flags: 6 participants
P-0418 — sleep_disturbance · new
P-0512 — survey says fine,
case notes say withdrawn
P-0623 — attendance dropped 40%
P-0741 — no audio reflection · 3 mo
P-0802 — instrument plateau
P-0915 — case notes name barrier
(transport) program missed
every_figure_cites_source: true
funder_export: ready
Frame 05 · The architecture underneath
The old promise was that impact measurement would be the layer where evidence met decision. Legacy made it storage — a place where surveys went to die and case notes never reached the dashboard. Three layers redo the work the legacy stack skipped.
Reads the case-note PDF, the audio reflection transcript, the survey response, and the ledger entry as evidence. Extracts themes, flags drift, surfaces outliers, and writes the citation trail back to the source record. The reasoning layer does not store the data — it reasons over it on every query. Sopact Sense is the system of record.
One stable participant ID across application, case note, survey, audio reflection, story, and ledger entry. Pre and post survey questions — the ones your program already uses — sit on the same row as the case notes and the cost data. The same participant who appears in a workforce program and a community health program is one record with two outcome tracks. The board view reconciles them without manual stitching.
QuickBooks, Xero, Sage Intacct, Bill.com — cost-per-outcome and program-cost figures pull from the accounting system at the moment of use. Attendance and activity records from Apricot, ETO, or Salesforce Nonprofit Cloud are referenced, not duplicated. The operational tools keep their job; the connected record keeps the identity.
Your theory of change, your logic model, the outcome framework your funder requires, the question set the regulator wants. Sense binds each one to the same connected record. The same participant data feeds the funder-specific report and the board view — without re-keying, without rebuilding.
cohort = ys-northland-2024 = 47 participant IDs from the September 2024 intake. The IDs are the same ones used in the case-management system, the survey tool, and the accounting project tag.Who this is for
Mid-tier nonprofits, foundations, and impact investors. The buyer who has outgrown survey-tool-plus-shared-drive but cannot afford a six-month data architecture engagement before producing a report. Five segments where Sopact is the right tool.
Frequently asked
Impact measurement software helps nonprofits and foundations track whether their programs move participants on the outcomes they promised. Traditional tools split the work — survey platforms collect responses, case management systems log activities, accounting tracks dollars, and the evidence lives in disconnected stores. Modern impact measurement brings surveys, case notes, financial documents, and stories together under one stable participant ID, so the story in a caseworker's observations sits alongside the survey signal funders see in a report.
Legacy impact measurement and case management tools were built as data-entry workflows wrapped in a dashboard. Sopact is built around one connected record per person — the participant identity persists across application, case note, survey, story, and ledger entries, and the reasoning layer can answer a Tuesday question by reading the source documents and citing them. The difference shows up when a funder asks for the qualitative evidence behind a number and you can return the source case notes in the same minute.
Yes. Your theory of change, your logic model, and whichever reporting framework your funder asks for all bind to the same connected record. Your custom outcome framework for a foundation grant, the survey questions your program already uses for confidence or well-being, and any portfolio-wide framework an impact investor needs — they all read from the same source data. No re-keying, no parallel spreadsheets.
Sopact connects, does not replace. Intake contact records flow in from HubSpot, Salesforce, or your case management system — Bonterra Apricot, Efforts to Outcomes, Salesforce Nonprofit Cloud. Outcome roll-ups, cost-per-outcome figures, and citation-backed evidence flow out to your accounting ledger (QuickBooks, Xero, Sage Intacct) and your reporting destinations (Looker Studio, Tableau, Power BI). Sopact Sense sits in the middle and holds the longitudinal participant ID.
That is the design point. A mid-sized nonprofit typically runs four to twelve programs, each with its own logic model, its own funder reporting cadence, and its own outcome measures. Sopact binds each program's outcomes to the same connected record per participant without forcing a single data model on every program. The same participant who appears in a workforce program and a mental health support program is one record with two outcome tracks, and the board view rolls both up without manual reconciliation.
The design point is start small, grow. Most mid-tier nonprofits and foundations have their first cohort report in under a month. The longitudinal thesis is testable on five participants. Begin with one program and one outcome, prove the citation trail works on real participant data, then bring additional programs onto the same connected record. Sopact does not require a six-month data architecture engagement before producing a report, which is the failure pattern at this scale with Salesforce Nonprofit Cloud, Blackbaud, and Bonterra deployments.
Mid-tier nonprofits (50 to 2,000 participants per cohort), foundations and CSR funders (5 to 500 grantee partners), workforce and education programs with longitudinal outcome obligations, school-based youth services with multi-region caseworkers, and impact investors who report against a portfolio framework. The design point is the team that has outgrown spreadsheets and survey tools but cannot afford a Salesforce architect on staff. For organizations under 50 participants total, a survey tool plus a shared drive is often still the right answer.
Yes, and this is where the connected record earns its keep. Historical participant lists in spreadsheets, old SurveyMonkey or Qualtrics responses, scanned intake forms, case-worker narrative notes, and even funder reports themselves are read in and joined to a participant identity. The qualitative material that historically sat in a shared drive and never reached a dashboard becomes searchable evidence with citations. Most nonprofits discover their existing data already supports the outcome claim they have been making — it was never connected.
You could prompt your way to a one-cohort summary. The trouble shows up in production. Ask Claude or GPT the same question twice on the same data and you get different answers — this is how the models work, not a bug to fix. As the dataset grows, hallucination climbs: on Vectara's 2026 long-document benchmark, every major reasoning model fabricated information in more than 10% of summaries; the marketed-as-most-capable reasoning models perform worse, not better, because they "think" their way past what the source actually says. The fix is not a better prompt. It is structural — the same participant ID at year 5 as at year 1, deterministic joins on case-note PDFs and survey responses and cost data, every roll-up citing the specific record it came from. The reasoning layer runs on top of that structure, not in place of it. This has been Sopact's day job since 2014.
The framework underneath
Six layers, four data dictionary types, the Cell → Row → Column → Grid roll-up, and how to design a longitudinal program around evidence, not effort. Co-developed with Melbourne Business School. Used by 30,000+ practitioners.
Related reading
A different starting point
Sixty minutes with Unmesh Sheth. No deck. We work a question you already need answered — a cohort review, a funder report, a Tuesday question from your board — against your real data shape. You leave with a path that does not require rebuilding the data each cycle.
