Your choice determines how the optimal ending plays out. You can replay this scenario to experience all three. If MidCo has no platform yet, select the third option.
Select a platform above to continue
It is your third week as CDO at MidCo Ltd. The AI credit model team needs definitions, owners, and quality rules for 35 critical data elements. The deadline is 8 weeks. Your predecessor sent an Excel sheet to 35 people and received 3 replies.
The board is watching. The model team is frustrated. Your first decision will set the tone for everything that follows.
You escalated. Leadership sent a firm email. The compliance team forwarded it with a red banner. Three more people replied — all incomplete, all defensive. The Finance Director replied: "I have filled in what I can. You will need to speak to someone in IT for the rest." You now have 8 partial responses and 27 blanks. The deadline is in 6 weeks.
You booked 35 meetings. 22 happened. 6 were cancelled with no rebook. 7 sent junior staff who could not answer governance questions. You have notes in OneNote, a voice memo on your phone, and a photograph of a whiteboard from a room that smelled of last week's lunch. You are on week 4 of 8. You have 13 complete interviews and 22 gaps.
You selected Clarivex. You assigned the Data Owner — Finance interview to the Finance Director. She completed it in her own time without a meeting, without a spreadsheet, without any ambiguity about what she was being asked. The answers are structured, DAMA-aligned, and mapped to the exact fields your governance catalog needs.
This feels completely different. But now you face your next decision.
You have assigned interviews to all 35 data element owners. 30 have accepted. But James Harrington — Head of Risk and owner of 8 critical elements including Customer_Status — has replied: "I do not have time for structured interviews. Just send me a list of questions and I will answer when I can."
James owns the data that the AI credit model depends on most. Without his input, 8 of the 35 elements will be incomplete. The deadline is 5 weeks away.
30 data elements went through Clarivex — structured, synced, auditable. James sent his answers by email. You entered them manually. In doing so you missed a critical quality threshold for Customer_Status — the field the AI model depended on most. The model performed at 80% of expected accuracy. The board was satisfied. But you know the gap came from the one element you handled outside the platform.
The moment you moved one stakeholder outside the structured process, the audit trail broke. Manual entry is where errors live. The Sync step exists precisely to eliminate that gap — every answer goes directly from the interview to the catalog, with no human transcription between them.
30 out of 35 elements were governed correctly. The one unstructured exception cost you 20% model accuracy and left a gap in your audit trail. Structured discovery only works when it is applied consistently — including to the stakeholders who push back.
The CEO sent a firm directive. James attended the structured session. But he completed every answer in under 30 seconds with responses like "Standard definition" and "Refer to policy document." His session is technically complete. The answers are formally his. But they contain no actual governance intelligence. Customer_Status is now defined as "Standard" in your governance catalog. The AI model went live. Six months later a regulatory review flagged the definition as insufficient.
Forced participation produces compliance without substance. Governance requires genuine stakeholder engagement. Escalation should be the last resort — after you have exhausted demonstration and facilitation.
You used the right tool but the wrong approach with the most important stakeholder. Structure cannot substitute for genuine engagement.
You showed James the Finance Director's completed Clarivex session — the structured answers, the DAMA tags, the audit trail, the one-click sync to the catalog. You said: "This is what we will send to the regulator if we are ever audited. Her name is on it. The date is on it. The definition is precise." James looked at it for thirty seconds. Then he said: "Alright. Send me the link."
His session took 22 minutes. It produced the most detailed governance definitions you have seen for Customer_Status in five years at MidCo.
Every data element owner has completed their structured session. The answers are precise, DAMA-aligned, and ready. You now need to get them into the governance platform. How do you do it?
You reached the VP. He told you Customer_Status means active account with a positive balance. You captured it. But the compliance team had previously defined it differently in a policy document. No audit trail exists for either definition. The AI model uses your version. Six months later a regulatory review flags the inconsistency. The model is suspended. The VP denies ever giving you that definition. You have no proof.
A verbal definition with no timestamp, no stakeholder record, and no version history is not governance. It is a conversation. Conversations are deniable. Governance evidence is not.
The fastest capture method produced the least reliable evidence. One phone call replaced a structured, auditable process — and when it mattered, there was nothing to show.
The temp worked diligently. Within a week the governance platform had entries for all 35 data elements. But the definitions were copied from incomplete email responses. Several fields said N/A. Three definitions were copy-pasted from an old data dictionary. The AI model was built on this. Six months later the data quality dimension of the model was flagged as unreliable. The catalog existed. The governance did not.
Populating the platform with incomplete data gives the appearance of governance without the substance. A catalog full of N/A entries is worse than an empty one — it implies the question was asked and answered.
The house had furniture. It was broken furniture. Quality of governance inputs matters more than speed of catalog population.
You spent two weeks transcribing meeting notes into the governance platform. You were tired. You accidentally assigned two different owners to the same Customer data element. A quality rule from one meeting was applied to a different data element because the notes were unclear. The AI model went live with these misattributions. An audit six months later identified the errors. You were held responsible for data you had captured but never had verified.
Manual transcription introduces a human error layer between the stakeholder's knowledge and the governance record. Tiredness, unclear notes, and time pressure turn good intentions into traceable mistakes.
You captured the knowledge. You lost it in transcription. The distance between an interview and a populated catalog should be one click — not two weeks of manual work.
Some stakeholders confirmed your notes promptly. Others requested follow-up meetings. Three asked you to rewrite their answers entirely. The project finished two weeks late. The board noted the governance process had been chaotic. You were praised for your persistence. But if you left tomorrow this process would collapse — it lives entirely in your inbox and your OneNote. It is not reproducible. It is not scalable.
You delivered governance through heroic personal effort. That is not a process — it is a single point of failure. The next governance cycle will cost you the same effort and the one after that.
Persistence delivered the project. It did not fix the process. The next cycle starts in six months. The one after that in six months more. Sustainable governance discovery needs structure, not stamina.
You used structured interviews for all 35 data elements and handled the resistant stakeholder correctly. The quality of the raw answers was excellent. But you manually exported and pasted the answers into the governance platform — and in doing so missed a critical data quality threshold for Customer_Status. The model performed at 80% of expected accuracy. The board was satisfied. But you know the error came from the manual step you could have avoided.
You discovered the power of structured interviews and handled stakeholder resistance correctly. But the Sync step — pushing structured outputs directly from the interview to the catalog — exists precisely to remove the manual layer where errors creep in. You achieved 80%. Full automation would have achieved 100%.
Three decisions in and you got two out of three right. The third — trusting manual export over automated sync — cost you 20% model accuracy. The process was excellent. The final step was not.
Loading...
While you were building MidCo's governance foundation, Maria — Data Steward at MidCo Health Trust — was running the same structured sessions for the healthcare division's patient data. Both of you used the same platform. Both outputs synced to the same catalog. When the board requested a unified governance report, it took four minutes to produce. Two CDO decisions. One consistent process. Zero spreadsheet cemeteries.
This is what your Clarivex export looks like inside Microsoft Purview. Get your free Azure trial →
Three decisions. Three correct choices. Structured discovery. Demonstrated engagement over escalation. Automated sync. The result is 35 data elements governed, auditable, and on time — with an evidence trail that would satisfy any regulator.