What does governing the work mean for a small team?

In short: Every step of the data cycle has a named person and a date, every finding is checked before it travels, and what the AI may read is set once, in the workflow, by your program team rather than IT.

Picture the week the spring 30-day follow-up closes: 25 of 40 completers have replied. The delivery manager wants to know by Friday which completers used the skill at work and what stopped the others. Someone suggests pasting the export into ChatGPT. Nobody has said who checks the numbers before they go to partner employers, or whether learners’ names belong in that chat.

On the usual paths those questions get answered by whoever happens to hold the data. Governance here is the answer to a plain question: who decides, and what happens when they leave?

Lesson 1 walks through that comparison. This lesson is about running the last row week to week, in two halves: the people (roles, a calendar, a review, an action plan) and the controls (folders, field selection, a scoped assistant). Each fails without the other.

Who owns what on a small team?

In short: Three roles cover most workflows: owner, reviewer and data steward. On a team of two or three, one person may hold more than one, as long as they know which job they are doing at the time.

The steward is the role that usually goes unassigned. On the export-and-paste path it falls to whoever builds that quarter’s spreadsheet, so it changes hands without anyone noticing.

What does one cycle look like?

In short: Four steps, each with a date and an owner: collect, review, act, check. Put them on the calendar when you plan the program, not after the data arrives.

One cycle for the training team (illustrative).

Governed at collection, each 30-day answer lands on the learner’s existing record the day it is submitted. Maria’s record (ID 0417) already holds confidence 2/5 at intake and 4/5 at exit, 10 of 12 sessions and her mentor’s note about the mock interview she led; her 30-day answer joins the same thread. An Intelligence Cell reads each open-ended answer on arrival with the prompt you configured, so the review starts from linked, read answers rather than a fresh export.

How do folders give each team its own workspace?

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

For the training team, one folder holds the skills course: intake, exit, the 30-day follow-up and mentor notes, all on the same IDs. If the organization also runs a leadership program with another team, that team gets its own folder and builds its own forms, and neither team’s Assistant can reach the other’s data. The owner still sees results across both without anyone merging files.

The same pattern fits affiliates, chapters, national teams and program sites: each keeps control of its own surveys, and nobody waits on a central administrator to add a question. The combined picture is only meaningful for the few fields every folder defines the same way. If your work spans several sites, Many programs, one picture shows which fields to share and how to combine them. Decide who creates folders and who approves access, and write it on your handover page.

How do you decide what the AI may see?

In short: Two controls. Field selection sets which fields are sent to AI APIs, so names, emails and phone numbers can stay out of the model. And the AI Assistant stays locked until you pick which surveys it may use.

On the export-and-paste path the rule is whatever the person at the keyboard decides that afternoon; a spreadsheet with a name column goes into the chat whole. Governed at collection, the steward decides once, in the workflow. For the training team, name, email and phone are off. The barrier answer, mentor note and confidence score are on, because the question needs them. The ID stays on every answer, so a reviewer can still open Maria’s record; the model does not need her name to read what stopped someone.

Scope works the same way. Before the delivery manager asks anything, the Assistant is locked. She ticks intake, mentor notes, exit and the 30-day follow-up, and leaves out anything the question does not need, such as a payroll export. The answer can only draw on what was picked, and every line of it links to a record she can open.

Keep a line in your team’s change log for who approved the field selection and scope, and when, and revisit it whenever a form or field is added. Lesson 2’s deep dive on what the AI may see goes further into sensitive fields and small groups.

How do you run the review meeting?

In short: Keep it short and fixed: what came in, what it shows, what is missing and what you will do. Thirty minutes is usually enough for one workflow.

For anything the AI produced, the reviewer asks one question: does the record behind this line say what the summary says? If not, the finding waits.

How do you turn a finding into action?

In short: Pick one change your team can make, give it an owner and decide in advance how you will judge it.

Five learners saying they had no time to practise is a question to investigate, not a fact about all 40 completers. The team decides to test a short practice session after the next course, and writes the decision rule before any data comes in.

How do you keep the next cycle comparable?

In short: Change the program, not the measurement. Keep the question wording, definitions and timing the same unless there is a good reason, and record it when you change them.

Use the same follow-up question, the same definition of “used the skill” and the same window, day 25 to 35. Record who received the practice session. Keep the spring result as it is; correct errors openly with a note rather than replacing a number. Because the next cohort’s answers arrive in the same workflow, the comparison is ready when the wave closes.

Keep the change log as a team habit: what changed, why, who approved it and from which date, for questions and definitions as well as folder access, field selection and scope. Note the staff time the cycle took; if three people cannot sustain it, simplify before you expand. Change a question safely covers wording changes that would otherwise break the comparison.

Be honest about what the next cycle can show. A higher rate would be encouraging, but it would not prove the practice session caused it: groups and response rates differ, and “used the skill” is self-reported by those who chose to reply. An AI summary of the barrier answers is a reading, not a verdict; check it against the source. The decision belongs to your team. Lesson 5 shows how to define the measure so the comparison is fair.

What should a handover include?

In short: Someone else should be able to run the next cycle from your notes alone.

Keep one page with the roles, the calendar, where each form lives, the current definitions and the change log. Add the governance settings: the folder and who has access, the fields switched off for AI and the surveys the Assistant may use. List any IT dependency, such as organization-wide sign-in rules. Then ask a colleague to walk through the next cycle without asking you anything.

What does owning your own data look like?

In short: No IT ticket for every question, learning all year instead of a report at the end, and answers you can trace back to a record and trust.

Put the two halves together and the training team’s week looks different. The coordinator adds a question to the 30-day form herself. The operations lead sets what reaches the model. The delivery manager asks her question on Tuesday, not after a quarter-end export, and opens the records behind the answer before anything goes to partner employers.

In Sopact Sense today that rests on a persistent ID from the first form, folders, field selection, a scoped Assistant and answers that link to records. A shared place for definitions the Assistant can use is coming soon; until then, keep definitions in your working evidence plan and include them in the question.