Written by The previous series discussed how DASUD could be adapted to the world of AI. Now, the question is about how and when the rubber meets the road. Road testing DASUD with AI, highlighted one key aspect – capturing all the AI use cases and the value that it brings. Some fundamentals never change – in Business Analysis you want to “shift left” as much as possible and get as close to the problem as possible. To me, the next challenge was capturing the wins of AI. When every man and his dog can slap “AI” on their offering, the real question becomes, “should we be solving this problem?”. We all have limited time and are expected to produce more. The one thing AI cannot do as yet, is help us understand the “why” we do what we do. Depending on when you read this post, the cost of AI and “tokenmaxxing” – using as many tokens as possible (bad idea), is coming to an end and you need to be really clear on what you’re using your AI for. And as such, it’s important to focus on the “why” and provide a framework that people can easily follow to maximise the return on investment from any AI development. 9/10 times, it’s actually saying “no”. So that one time you say “YES” – you know you’re on to a winner.
Inputs and Outputs
It was genuinely naivety on my behalf to think that people saw the same horizons as I did. Grateful for having completed my Certificate 4 in Training and Assessment in the VET sector, I now understood what it means to have a beginner mindset and to explain concepts as I go along. So when it came to AI, I realised it was more about the inputs and how well that problem was defined, that really helped cement what needed to happen next. Now I agree, that the world of AI is moving at an incredible pace and you can be forgiven for feeling that you are lagging behind. However, do not fear dear reader, this is hardly the case. The only ones that are worth listening to, are the ones that are in the “arena“. So here is my attempt at being in the arena for you.
Phase 1: Design the Intake Layer
Start by redesigning how ideas enter the system rather than how they’re judged – most approval bottlenecks are created upstream by messy, inconsistent intake, not by “too much governance”. A one-page intake form (problem, current process pain, data types involved, expected benefit) replaces ten-page business cases, which keeps grassroots submissions flowing instead of only the most persistent requesters getting heard. This is also where business learning starts: every submission becomes a data point on where the organisation actually feels AI pain, feeding a shared “problem inventory” leadership can see in real time.
Phase 2: Acquire Evidence Through Triage and Scoring
Move from opinion-based approval to a standardised scoring model covering business value, feasibility, risk, and strategic alignment, each weighted and calculated into a single composite score. Triage should be fast — five business days to filter duplicates and out-of-scope ideas — so people trust the process enough to keep submitting. This phase directly removes the classic bottleneck of “loudest voice wins,” while simultaneously building an evidence base (what data exists, what risk categories recur) that becomes organisational learning for future submissions.
Phase 3: Store And Centralise The Use Case Register
Every scored use case lands in a shared backlog/register with status, owner, and priority visible across the organisation — this is the “Store” discipline from DASUD applied to governance artefacts, not just data assets. Publishing an internal AI use case catalogue of approved, in-progress, and completed work does double duty: it prevents duplicate investment (a major hidden bottleneck) and inspires other teams by showing peer department wins. This register becomes the single source of truth executives use to track AI oversight and risk at a glance.
Phase 4: Use A Tiered Approval Workflow
This is where governance and speed stop competing: apply risk-tiered approval so low-risk, high-feasibility use cases fast-track through a lightweight sign-off, while high-risk or high-impact cases route to a monthly governance committee review. Embedding checkpoints into the lifecycle (requirements → design → test → deploy → monitor) rather than bolting governance on afterward is what lets teams move quickly while still being audit-defensible. Removing mandatory manager pre-approval for submission, but keeping structured governance at the decision gate, is a proven way to unblock volume without diluting control.
Phase 5: Delete, Review, And Close The Learning Loop
The final phase mirrors DASUD’s “Delete” discipline: retire or sunset underperforming use cases on a schedule, and feed measured outcomes back into the scoring criteria and intake templates. Organisations mature through three stages here — reactive (shadow AI, duplicate spend), controlled (centralised intake with consistent criteria), and strategic (a continuous innovation pipeline where the centre of excellence proactively mentors submitters and optimises the whole portfolio). Documenting this evolution publicly is itself powerful thought-leadership content, since it shows the governance model improving itself over time rather than staying static.
In an upcoming series, I will share my own thinking and how it can be expedited, given that there is an awesome YouTube video on my exact thinking process (unbeknownst to me and flattering that I am on the right path) which you can watch below. There are only 2 items, that I hadn’t considered and I will incorporate into my process going forward, so you can launch your AI strategy in the most efficient and effective way possible.
If you’d like assistance or advice with your Data Governance implementation, or any other topic (Privacy, Cybersecurity, Ethics, AI and Product Management) please feel free to drop me an email here and I will endeavour to get back to you as soon as possible. Alternatively, you can reach out to me on LinkedIn and I will get back to you within the same day!
Nigel D'Souza 