AI Data Companion
From “add AI” to designing an actual product.
“Bring AI into the data integration experience.”
AI was already mandated as part of the requirement, and I was asked to design the experience. But I didn't want to start by designing an AI feature — I wanted to understand what the user actually needed AI to help with.
We move data more than we realise.
Organizations constantly move data across systems for critical business initiatives:
Moving data sounds straightforward: Take data from System A and put it into System B.
- ✕ Inconsistently named across schemas
- ✕ Missing vital business context
- ✕ Scattered across disparate documents
- ✕ Described differently across systems
- ✕ Understood completely differently by different teams
The Moving-House Analogy
To anchor the design team and stakeholders around the real problem, I introduced a tangible physical metaphor:
Why it belongs here: Analyzed sample fields inside Box #104 — contains commercial credit facility identifiers, maturity dates, and counterparty entities matching Corporate Lending standards.
Meet Alia — The Data Analyst
Alia's Core Responsibility
Understand complex enterprise data, build trusted business knowledge, create accurate source-to-target mappings, and support multi-million dollar integration projects.
“AI magically generated a black-box mapping.”
“I understand this data well enough to trust the mapping I'm approving.”
Business meaning is documented clearly, agreed upon, and reusable across teams.
Source and target field relationships make complete technical and business sense.
Less manual grind spent discovering schemas and decoding obscure field names.
The analyst clearly understands the underlying reasoning behind every AI suggestion.
Understanding Comes Before Mapping
Before designing any AI UI, I mapped what the analyst actually had to do step-by-step:
The Product Wasn't Defined Yet
There was an AI requirement, but there wasn't yet a clear product model around it. Instead of jumping directly into screens, I worked backwards through collaborative discovery:
Interviewed multiple PMs across Data Ingestion, Catalog, and Integration to identify where AI could genuinely relieve cognitive load.
Created end-to-end storyboards and interactive prototypes to challenge initial assumptions about linear AI generation.
Structuring by Context: Groups & Products
Enterprise datasets are too vast to dump into a single AI prompt. We structured the product model around bounded domains:
Scope the AI to specific product domain.
Ingest schemas, sample records & specs.
AI creates initial shared business definitions.
Suggest source-to-target field linkages.
Analyst reviews, corrects & refines.
Persist approved knowledge for next time.
Designing for the Wait
AI processing is not instantaneous. If every generation step forced the user to wait, the product would devolve into a sequence of blocking spinner screens.
Analyst is trapped on loading screens, breaking cognitive momentum.
Analyst remains in continuous flow while AI works in the background.
AI finished drafting 18 glossary definitions and 24 mapping recommendations in background.
AI Assists. Humans Decide.
The core philosophy of the companion: AI reduces discovery friction; the analyst retains full accountability.
Proposes glossaries and source-to-target field matches based on data sample patterns.
Provides transparent reasoning: schema lineage, field statistics, and semantic similarity scores.
Adapts when the analyst overrides a recommendation and updates domain memory.
Inspects proposed business glossary and mapping candidates with context.
Corrects misclassifications, adjusts data types, and enriches business descriptions.
Formally signs off on integrations with full auditability and confidence.
Prototyping Under Ambiguity & My Contribution
Static screens were insufficient to evaluate asynchronous behaviors. I used Figma + Codex to build interactive prototypes that acted as thinking tools for leadership.
- • Shaping the core problem
- • Defining where AI fits
- • Agreeing on future journey
- • Product model structure
- • Current journey mapping
- • AI use-case exploration
- • Storyboarding & flows
- • Asynchronous interaction model
- • Figma + Codex high-fi prototypes
- • Human/AI interaction rules
- • Trust & explainability model
- • Future roadmap scoping
The Big Design Shift
AI needs bounded scope (Domain Groups & Products) rather than raw enterprise noise.
The business glossary is part of the reasoning process, not a side feature.
Analysts require transparent rationale before approving suggested mappings.
Design around the wait so users can stay productive across multiple products.
AI assists, suggests, and explains; the analyst remains the final decider.
Validated definitions become reusable enterprise assets rather than single-use outputs.
Shows how I untangle massive existing legacy complexity, silos, and conflicting operational workflows.
Shows how I lead 0-to-1 product definition, user modeling, and async interaction when nothing is defined yet.