How do you combine results across programs or sites?

In short: Let each team run its own work, agree definitions for the few fields you will combine, and pool counts only where definitions, timing and population match. Show coverage, overlap and anything excluded beside the combined number.

A shared outcome label is a starting point, not proof that the numbers can be added. On the usual paths the problem arrives at quarter end. With surveys and spreadsheets, each site emails its file and someone spends a week reconciling columns that were never defined the same way. A CRM forces every site into one central model, and the admin who built it is the only one who can add a local question. A warehouse promises to pipe it all together and needs engineers you do not have. The result is the same: six systems, zero trust, and a board question that takes weeks.

How do folders give each site its own workspace?

In short: In Sopact Sense each site’s team gets a folder, builds its own surveys there, and asks the AI Assistant about its own data only. The organization owner sees aggregated results across every folder.

For the training team, Sites A, B and C each get a folder. Each coordinator runs intake, exit and the 30-day follow-up for their learners, and every learner gets a persistent ID at the first form. Site B, where most learners work shifts, adds a local question about shift patterns without asking anyone’s permission or waiting for an administrator. Site B’s Assistant sees Site B’s data; it cannot reach learners at Site A. The delivery manager, as owner, sees the combined results across all three.

The same arrangement fits affiliates, chapters, national teams and partner sites. It keeps local ownership where the work happens and gives the owner a picture without a quarterly merge. Decide who creates folders, who is added to each and who approves access, and record it in the change log your team keeps under the Lesson 4 roles.

Which fields should every folder share?

In short: Only the few you will combine, defined once and used the same way everywhere. Everything else can stay local.

Two sites can use identical words and still differ: one sends the follow-up at day 30, another at day 60; one counts a learner who missed the last session as a completer. Write the rule, the source field, the timing and who owns the definition. A shared place for definitions and a data dictionary the Assistant can use is coming soon in Sopact Sense; until then, the data steward keeps it as a team document and every folder follows it.

How do you decide whether a measure can be combined?

In short: Classify each proposed measure before adding anything. Only the first row gets pooled.

Keep “not comparable” separate from “not yet reported”. Both affect coverage, but they need different follow-up.

How do the numbers combine across three sites?

In short: Add numerators and denominators, not percentages, and report coverage beside the rate.

Spring cohort by site. Fictional; totals match the course example.

The pooled rate is (7 + 6 + 2) ÷ (11 + 9 + 5) = 15 ÷ 25 = 60% of respondents. Coverage is 25 ÷ 40 = 62.5%, and 15 outcomes are unknown. Averaging the three site rates instead, (63.6% + 66.7% + 40%) ÷ 3, gives 56.8%, because it weights Site C’s five replies the same as Site A’s eleven. If you want a site-level average, label it as one. For “how many learners used the skill”, use the counts.

The employer partner’s in-house course asks at exit whether learners plan to use the skill: 10 of its 12 completers said yes. That is a different question at a different time. It goes in its own panel beside the three sites, labeled, not added to the 15. Leaving it out silently would be as misleading as adding it.

A sentence the reviewer can reproduce: “Across Sites A, B and C, 15 of 25 completers who replied at 30 days used the skill at work; 25 of 40 completers replied. The partner’s in-house course measures intention at exit and is shown separately.”

Should you rank sites by their rates?

In short: Not from a table like this. Site C’s 2 of 5 is a question to investigate, not a verdict on its coordinator.

Site C also has the lowest coverage, 5 of 10. The five who did not reply may differ from the five who did. Check definitions and timing first, then context: did Site C start later, serve a different group, or send the follow-up late? Read its open answers against their records, using the codebook from Clean open-ended answers, with room for local codes. If some came in Spanish, Multilingual feedback shows how to review them.

With numbers this small, one more reply could move a site rate by 10 points or more. These are self-reports from the people who replied, and a difference between sites shows association at most. If the comparison will drive staffing or funding decisions, it needs a design built for that question. The people who know the sites make the call.

How do you avoid counting the same person twice?

In short: Decide whether you are counting people or participations, and use the persistent ID to tell them apart.

A learner who starts at Site A and finishes at Site B, or repeats the course, can appear in two folders. For “participations”, both count. For “people served”, they count once. Because each learner keeps one ID from the first form, the overlap is visible rather than guessed. Where you cannot reliably match people, such as partner-submitted totals, report participations and say so. Do not share personal details across organizations to get a cleaner headline.

What does one picture look like in practice?

In short: Local teams keep their own work, and the owner gets one view that is ready when the data arrives.

Open Play Foundation runs four sports facilities in Stellenbosch, South Africa, covering coaching, water infrastructure and food security. With its data governed at collection, ten program reports became one funder submission, and a water leak surfaced in real time because the data was connected. Its CEO put it as: “I’m digitizing our entire business through Sopact.” The King Center’s published story describes bringing survey evidence and qualitative feedback together across seven programs. Neither story validates the fictional numbers above.