Lesson 1 of 6 · Start here
What Is Connected Data Intelligence? Start Here Before You Add AI
Your team already collects enough. Applications, surveys, attendance, tickets, case notes, documents. What you cannot do is answer one question that runs across all of them and show the evidence behind the answer. This lesson explains why, what makes this different from every tool you already pay for, and how to work through the rest of the Academy.
In short: Connected data intelligence is the practice of keeping every record you collect attached to one continuing record per person or organization, with shared identifiers, agreed definitions and dated cycles, so a question that spans your tools can be answered and the evidence behind the answer can be inspected. It is less about new tools than about making the records you already have work together — and it is what makes AI safe to use on your own data.
You will leave with: one recurring decision written down, the records that should answer it, the gap that blocks it today, and the course path you are going to follow.
The problem, stated plainly
Here is a test you can run this week without changing anything.
In your next serious meeting, when someone challenges the number on the dashboard, watch what people open.
If the answer is a private spreadsheet, a program manager's notes, or a separate tracker that one person maintains, that is not a dashboard problem. It is a trust problem. The official system holds a version of reality that the people closest to the work do not fully believe.
Every system in your stack does its own job well. A CRM holds relationships and contacts. A case management system holds service delivery. A survey tool holds ratings and open comments. A spreadsheet holds the number people actually quote out loud. None of them is wrong. Each of them sees part of the same story, and none of them sees the person.
That was tolerable when reporting was annual and somebody had a week to reconcile it by hand. It stops being tolerable the moment you put an assistant on top, because an assistant will answer anyway. It will call a cohort disengaged without seeing the case notes, the people who never responded, the question that was reworded last cycle, or the program change that gives the result its meaning.
The danger is not an obviously broken answer. The danger is a confident, well-written, executive-ready answer that nobody can trace back to the evidence. AI does not create context the system never captured.
Three minutes on the same argument, and on how to work through the rest of this Academy.
Why can't a team with plenty of data answer one simple question?
In short: Each tool does its own job well, but nothing connects them. Answering a question that spans tools means exporting, matching and re-reading by hand, every cycle — so it gets done once a year, or not at all.
Follow a fictional three-person team that runs skills training for employers. Registration sits in a CRM, attendance in a spreadsheet, feedback in a survey tool. The delivery manager asks a reasonable question: "Which completers used the skill at work after 30 days, and what stopped the others?"
01 · REGISTRATION
Who enrolled? · CRM
02 · ATTENDANCE
Who completed? · Spreadsheet
03 · FOLLOW-UP
Who used the skill? · Survey tool
04 · REVIEW
What should change? · Nobody owns it
Four sources, four tools, no shared ID. The question spans all of them.
No single tool can answer it. A satisfaction score does not say who used the skill. Attendance does not equal completion until someone defines completion. A comment about "no time" explains one person's experience, not everyone's. And the one person who could stitch it together does it by hand, which means it happens once.
What actually breaks: four quiet failures
In short: Identity, reasons, comparability and definitions fail silently. Nothing errors out. The report still renders. It is just no longer defensible.
| What breaks | How it looks in practice | What it costs you |
|---|---|---|
| Identity | One person is a contact in the CRM, a client in the case system, a row in a survey export and a first name in a spreadsheet | Your team sees one person; an assistant sees four records. Nothing can be counted per person |
| Reasons | A score tells you confidence dropped; the reason sits in an open comment, a case note or an interview stored as loose text | You can report the number but cannot defend the explanation attached to it |
| Comparability | A new cycle brings a new form, new fields, a reworded question and sometimes new staff | Nobody can say whether the difference is a changed outcome or a changed question. The series restarts |
| Definitions | "Enrolled" means attended a session to one team, intake complete to another, status field active to a report | All three are calculated correctly and all three disagree, in public, in the meeting |
Bad data is rarely missing fields. It is missing context and unclear definitions. Before anything answers a question for you, your team needs a shared definition, a named owner, a stated population, and a way to see when that definition changed.
How is this different from a CRM, a case management system, a survey tool or a spreadsheet?
In short: Those systems store fragments and leave you to reconcile them later. Connected data intelligence agrees the identity, the definition and the cycle before collection, so records arrive already connected. That is the whole paradigm shift.
| System | What it is built to do | Where it runs out |
|---|---|---|
| CRM | Track relationships, contacts, pipeline and history with an organization | Holds a contact record, not a measured series. Open text has nowhere useful to live, and a second cycle is a second record rather than a second observation |
| Case management | Record what was delivered to one person, case by case | Strong on the individual case, weak across a cohort. Comparing this year's intake to last year's usually means an export |
| Survey platform | Capture one wave of responses cleanly | Each wave is its own export. People are identified by whatever email they typed. Comments come back as unstructured text nobody reads at volume |
| Spreadsheet | Let one capable person assemble the number quickly | The definition lives in that person's head. It is rebuilt by hand every cycle, has one owner, and leaves no trail anyone else can follow |
| Warehouse or BI dashboard | Combine data after collection and chart it | Needs technical staff, and inherits every identity and definition problem from upstream. It can show the number; it cannot show the sentence behind the number |
| Connected evidence layer | Keep every source attached to one continuing record, with identity, definitions, cycles and source text held together | It asks you to decide things up front — who the record is, what a word means, who owns it. That decision is the work this Academy teaches |
Read the table one more time and notice what it is not saying. It is not saying replace your CRM. Keep every tool that does a clear job well. The change is that the evidence stops living inside those tools and starts living in one continuing record they all point at. Intake, a follow-up response, a case note, an uploaded document and an outcome can all stay attached to the same person, in order, with the consent and visibility rules already applied.
The unit of value is not one survey response. It is one continuing record. That is what makes a question about change over time answerable at all.
Why does this have to be AI-native rather than AI added on top?
In short: Bolted-on AI reads whatever fragments happen to exist and fills the gaps with fluent language. AI-native means the structure, the coding rules and the source text are established as data arrives, so the assistant is reading a governed record instead of guessing at one.
Every system in your stack is shipping AI features. The pitch is the same everywhere: an assistant that can see a contact, a case, a survey and a history, and help your team act faster. The promise is real. It only holds when the record underneath is complete enough to trust.
Two things have to be true before an assistant is useful on your own data.
WHAT AI-NATIVE ACTUALLY MEANS
- Evidence is made usable as it arrives. A response is attached to the correct record, your coding and indicator rules are applied, the original passage stays visible behind any theme or score, and consent and anonymity rules are enforced before an assistant can read it.
- The structure can change without losing the series. Questions change. Programs change. A new funder asks for something nobody tracked. You describe the change in plain language, and the record stays continuous so this cycle is still comparable to the last one.
That second point is the flexible part, and it is the one most teams underestimate. In a traditional stack, changing what you collect is expensive — a form request, a schema change, a migration — so the series breaks or the question never gets asked. People still decide what to collect and what a finding means. The system's job is to make that judgment repeatable and inspectable.
The question then changes shape. Instead of asking someone to export a survey, hand-code comments, build a spreadsheet and join it back to program data, a manager can ask: why did confidence drop in the north cohort? — and get the number, the attribute that separates the group, the theme in the comments and the original wording behind it. The report is not rebuilt. It is a view of connected evidence.
That is the claim. Lesson 2 turns it into a test you can run on the tools you already have, by separating three things people all call AI: a standalone assistant you paste files into, an AI feature inside one tool, and a workflow where collection and analysis were designed together.
What self-governance means, and why it is the point
In short: Self-governance is the ability to change what you collect without losing what you have already collected — and to do it yourself, this week, without an IT queue, a vendor ticket or a consultant.
The speed at which your team learns is set by one thing: how quickly you can change a question and still compare the answer to last cycle. If a change takes a quarter and breaks the series, people stop asking. The measurement system quietly becomes a compliance exercise, and the real thinking moves back to the spreadsheet on someone's laptop.
Self-governance means four things sit with your own two- or three-person team rather than with a vendor or a technical department:
- Identity. You decide when records can be linked, when they cannot, and when consent says they must stay separate.
- Definitions. You decide what "completed" counts, who owns that definition, and everyone can see when it changed.
- Collection. You change a form, add a follow-up, or translate a question without filing a request.
- Reporting. You produce the report, and a reviewer can reproduce any number in it from the source.
This is also the honest answer to the governance question a board or a funder will eventually ask. Governance is not a policy document. It is whether a reviewer can open one number and walk back to the response it came from.
Who benefits from this, and for which work
In short: Any team that collects from the same people more than once, mixes numbers with open text, and rebuilds the same report by hand every cycle. The sector changes; the need to tell records and reporting periods apart does not.
Below is the work each Academy workflow course is built around. Find the row that matches the decision you are responsible for — that is the course you will pick after Foundations.
| If you manage | The recurring decision | What has to connect | Course |
|---|---|---|---|
| Members and networks | Who is disengaging before renewal, and what would bring them back | Join application, engagement history, pulse surveys, staff notes | Membership & networks |
| Casework | Which clients are stalling, and what barrier is common to them | Intake, assessments over time, case notes, referrals | Case intelligence |
| Training and programs | Did anything change after the training, and for whom | Registration, attendance, pre and post responses, 30-day follow-up | Training & programs |
| Applications and awards | Which applications to advance, consistently and explainably | Application, attached documents, rubric scores, reviewer comments | Applications, awards & grants |
| A portfolio | Which results across a varied portfolio are real and comparable | Reports from each organization, shared indicators, documents, multiple years | Investment & grant portfolios |
| Partners and suppliers | Which partners are delivering what they agreed, with evidence | Agreements, periodic submissions, verification documents, site notes | Partners & suppliers |
| Customer experience | Which accounts are at risk, and what is actually driving it | Account record, tickets, satisfaction scores, verbatim comments | Customer experience |
| Employee experience | What is driving attrition, in which teams, and whether it moved | Onboarding, pulse surveys, manager notes, exit interviews | Employee experience |
You will get the most from this Academy if your team is small enough that nobody's job title is "data engineer", you collect from the same people more than once, you care about the comments as much as the scores, and there is a report you rebuild by hand every cycle. You will find it basic if you already have a data team, a warehouse and clean transactional data, and your only open question is which chart to draw.
How should you work through the Academy?
In short: Foundations first, then one workflow course that matches the work you manage, then measurement and reporting when you need to defend results outside your team. In that order.
One warning about the order. Most teams want to start at step 3, because reporting is where the pain shows up — the deck is due, the numbers disagree, the deadline is Friday. Reporting is the last place the problem appears and the worst place to fix it. A report is only as defensible as the identity and definitions agreed before collection, which is step 1. Do step 1 properly and step 3 takes an afternoon.
Pick one workflow at step 2, not three. A single workflow carried all the way through teaches your team the method; three half-built ones teach nobody anything.
Start with a decision, not a new form
In short: A new form or tool cannot fix a question nobody has written down. Write the recurring decision as one sentence first.
DECISION SENTENCE · TEMPLATE AND EXAMPLE
If you cannot name the period, the owner or the action, you do not yet have a decision — you have a reporting habit. That is worth knowing before you build anything.
Give every source one job
In short: Give every source one clear job, and write down how it identifies people, when it records data and who owns it. The gaps become obvious on the page.
Source map for the training example (illustrative)
| Source | Its job | Identifier | Date recorded | Gap |
|---|---|---|---|---|
| CRM | Who enrolled, and for which employer | Contact ID | Enrollment date | Enrollment ID isn't shared with other tools |
| Attendance sheet | Who took part in each session | Learner name | Session dates | No agreed completion rule |
| Follow-up survey | Who used the skill after 30 days | Email address | Submission date | Can't be linked to an enrollment |
| Coach notes | Why someone struggled | None | Written in documents | Sits outside every dataset |
- Don't merge on a name. One learner can take two courses, and one employer can send several learners. Person, employer and enrollment are related records, not interchangeable rows.
- Keep anonymous responses anonymous. Don't link them just because it's technically possible.
- Pick one source of truth for identity. Usually the CRM holds contact details; other sources point to it.
Name the gap that stops you today
In short: Label the real gap before you change anything. A new survey platform will not fix an undefined completion rule.
| Gap | What it looks like | Where it's fixed |
|---|---|---|
| Identity | The same person appears under three emails | Lesson 3: persistent ID |
| Definition | "Completed" means different things to two staff | Lesson 3: framework and definitions |
| Timing | Nobody knows which survey wave a response came from | Lesson 3: time context |
| Evidence | Comments sit in a folder nobody reads | Lesson 3: surrounding evidence |
| Access | Only one person can open the attendance file | Lesson 4: owners and access |
| Upkeep | The report is rebuilt by hand every quarter | Lesson 4: the review cycle |
Before you put AI on top of anything
In short: Run one complete cycle through the checks in Lesson 2 before you buy or switch on anything. The first check that breaks tells you what to fix first — and it is almost never the thing you were about to buy.
Those checks are observable. Can your own team make a routine change. Does new material get analyzed as it arrives, including open text. Can the system tell a second enrollment from a follow-up. Can everyone see what a count means and which version applies. And the one most teams skip entirely: are you covering everyone, including the people who never answered, so a missing response reads as unknown rather than no.
So do not open with when can we add AI? Open with this.
Would we trust an AI recommendation based on the evidence we have today?
If the answer is no, the next step is not another feature. It is a connected, governed record your own team can run — which is what the next five lessons build.
Where does Sopact Sense fit?
In short: Most survey tools stop at collection. Sopact Sense gives every person and organization a persistent ID from the first response, so registration, attendance, follow-ups, documents and notes arrive already connected — and your own team manages it without an IT queue.
Sense is the connected evidence layer in the table above. It brings authorized records from the systems you already use, alongside surveys, open text, documents and notes, into one continuing structure. Each source keeps its role. The evidence stops losing its connection to the person, organization or event it describes. Keep the tools that already do a clear job well.
Add this to your plan
Open your working evidence plan ↗
Four things, in this order:
- Write your decision sentence using the template above.
- List three sources that could answer it, the identifier each one uses, and which source is the authority for identity.
- Name the one gap that stops you today: identity, definition, timing, evidence, access or upkeep.
- Take five approved, non-sensitive example records through the whole flow. Can a colleague say who each record describes, which period it belongs to, what a reported number counts and where the evidence came from — without asking you? Every point where they needed help is a requirement, not a verdict on a tool.
Then write down which workflow course you are taking at step 2.
Check your reasoning
For the training example: each month the delivery manager decides whether to change follow-up support, using completion records, 30-day follow-up answers and reviewed comments. The gap is identity — the survey uses email and the CRM uses a contact ID, so repeat enrollments can't be told apart. A missing response stays "unknown", not "no".
Frequently asked questions
What is connected data intelligence?
It is the practice of linking the records you already collect — forms, surveys, attendance, notes and documents — through shared identifiers, agreed definitions and dated cycles, so you can answer questions that span them and show the evidence behind the answer. It is less about new tools and more about making existing records work together.
How is connected data intelligence different from a CRM or a case management system?
A CRM holds a contact record and a case system holds service delivery, one person at a time. Both are built to store the current state of a relationship, not a measured series you can compare across cycles. Connected data intelligence keeps every response, note and document attached to one continuing record with dates and definitions, so change over time is answerable and the open text stays attached to the number.
What does AI-native data collection mean?
It means evidence is made usable as it arrives rather than cleaned up afterwards: each response lands on the correct record, your coding and indicator rules are applied, the original passage stays visible behind any theme or score, and consent rules are enforced before an assistant can read anything. AI added on top of existing tools reads whatever fragments exist and fills the gaps with fluent language it cannot source.
What is data self-governance?
Self-governance is the ability to change what you collect without losing what you have already collected, and to do it with your own team rather than through an IT queue, a vendor ticket or a consultant. It covers four things: who decides when records can be linked, what a definition counts, how a form changes, and whether a reviewer can reproduce a reported number from its source.
Do we need to replace our CRM or survey tool?
Usually not. Keep tools that do a clear job well. The first step is agreeing which identifier links records and which source is the authority for identity. Replacing a tool rarely fixes an undefined rule or a missing follow-up date.
What is the most common gap?
Identity. The same person is recorded under different names or emails across tools, so answers cannot be joined without manual matching. Definitions come a close second: two people count "completed" differently, and both are right inside their own system.
How is this different from a data warehouse or BI dashboard?
A warehouse or dashboard combines data after it has been collected, and usually needs technical staff. Connected data intelligence starts earlier: identifiers, definitions and dates are agreed before collection, so records connect as they arrive and a small team can maintain them. A dashboard can show the number; it cannot show the sentence behind the number.
Where should I start in the Academy?
Start with Foundations, all six lessons, and finish your working evidence plan. Then pick the one workflow course that matches the work you manage. Go to measurement and reporting last, when an audience outside your team needs results they can check. Starting at reporting is the most common mistake, because a report is only as defensible as the identity and definitions agreed before collection.
When you need more detail: Which shape is your data?
If you're unsure whether to track a person, an organization or an event, use this guide to choose the record. Come back with the identifier your question needs.
Read the method →