Approach
Purpose first. Then the architecture.
Every engagement starts with five questions. The answers decide what we build, who is involved, and how we will know it worked.
The five questions
The discipline most data projects skip.
The questions come from Peter Drucker's work on effective organizations. Data programs rarely ask them, so they deliver platforms instead of results. How we apply them to data programs is our own, refined over 24 years in higher education: who sits at the table, how results are defined, and how the answers shape the architecture.
- 1
What is the mission?
What is the organization here to accomplish, and how will data help it show that it is doing so? For a public institution, the answer is public value: better outcomes for the people and communities it serves.
You receive
Mission and value alignment for the data program
- 2
Who is the customer?
Who depends on this data, directly and indirectly? Usually IT, research and planning, the operational departments, leadership and the board, and ultimately the people the organization serves.
You receive
Stakeholder map: primary and supporting customers
- 3
What does the customer value?
Accuracy, shared definitions, timely operational reports, a controlled transition, skills, and advanced analytics. We ask through interviews and a short survey, not assumptions.
You receive
User-values interviews and survey readout
- 4
What are the results?
What would be different outside the data team if this worked? We define results and indicators before we build, not after.
You receive
Results framework with indicators
- 5
What is the plan?
A phased roadmap that starts where value is highest, builds on the strengths you already have, and keeps reporting running throughout.
You receive
Phased roadmap with sequencing and costs
Everyone at the table
Make every group's strengths productive.
When IT owns the systems, research owns the definitions, and the operational departments own the business rules, no one can produce trusted information alone. Shared governance aligns those strengths. Research and planning join on day one, because they supply the meaning that turns data into information.
IT
Systems and security
Research and planning
Meaning and definitions
Operational departments
Business rules and context
Leadership and the board
Direction and decisions
One set of definitions
Shared, governed, documented
IT
Systems and security
Research and planning
Meaning and definitions
One set of definitions
Shared, governed, documented
Operational departments
Business rules and context
Leadership and the board
Direction and decisions
IT
Brings
Infrastructure, security, integration, and the systems of record.
Needs
A maintainable, secure architecture with less technical debt and less vendor lock-in.
Research and planning
Brings
Meaning: cohort and metric definitions, research design, and external reporting expertise.
Needs
Trusted, documented, reproducible data, and time to interpret it instead of reconciling it.
Operational departments
Brings
Context: the business rules of the departments that run day-to-day operations.
Needs
Accurate, timely operational reports and the skills to serve themselves.
Leadership and the board
Brings
Direction: priorities, decision rights, and the authority to govern.
Needs
A small set of trusted measures of mission and return on investment.
The plan: the Trusted Data Path
Diagnose, build the foundation, agree the metrics, then add AI.
Delivery is phased and starts where value is highest, often one department at a time, so reporting keeps running throughout. The platform is chosen to fit the skills and systems you already have.
The architecture
Three tiers, from campus systems to governed AI.
Canonical integration lands your sources and conforms them to shared business entities, keeping history. Analytical models organize the data around how you measure performance. Semantic intelligence defines each metric once, so people and AI read the same numbers.
Governance, security, quality, and lineage run through all three tiers, not as a step at the end.
Your systems
- Banner
- Colleague
- PeopleSoft
- Workday
- Slate
- Canvas
1
Canonical integration
Sources landed, cleansed, and conformed to shared business entities, with history preserved
2
Analytical models
Dimensional models of facts and dimensions, organized around how you measure performance
3
Semantic intelligence
Approved metric definitions in one semantic layer, read by people and AI alike
Dashboards and self-service
Leaders, managers, and analysts
Governed AI and agents
Assistants that answer from the same definitions
Governance · security · quality · lineage
How we deliver
Short cycles you can see, one subject area at a time.
- 1
Discover
Answer the five questions for this subject area and agree what a result looks like.
- 2
Model
Design the entities, measures, and definitions with the people who own them.
- 3
Build
Pipelines, models, and the semantic layer, in version control and under test.
- 4
Validate
Reconcile against trusted sources and the business owners’ own numbers.
- 5
Demonstrate
Show working results to the people who will use them, every few weeks.
- 6
Refine
Fold the feedback into the next cycle, then move to the next subject area.
After Refine, the cycle starts again with the next subject area.
Engineering practices
- Version control
- Every pipeline, model, and definition lives in Git and is reviewed before it ships.
- Tested transformations
- dbt models with automated tests for keys, relationships, and business rules.
- Metadata and lineage
- A catalog records owners, definitions, and where every number comes from.
- Governance in the workflow
- Named owners approve a definition before it reaches a dashboard or an AI tool.
- Knowledge transfer
- Your staff build alongside us and receive runbooks and training.
Cloud, on premises, or hybrid
There is no default platform. We weigh cost, skills, security, and lock-in with you, then recommend.
- Cloud
- Fastest to start and scale. A fit when policy allows data in your cloud tenant and your team wants managed services.
- On premises
- Full control for data that must stay on infrastructure you own, building on the database skills your team already has.
- Hybrid
- Sources and sensitive data stay on premises while analytics and AI run in your cloud tenant. Common during a phased migration.
Diagnosis
Two lenses, scored together.
Technology alone never fixes trust. The readiness assessment scores both lenses, because a strong platform without ownership decays, and good governance without a platform has nothing to govern.
Governance capacity →
Good rules, little to govern
Ownership exists, but the data platform cannot answer the questions yet. Build the foundation.
Ready to scale
Trusted data and the capacity to keep it trusted. Add metrics and governed AI.
Start with the basics
Neither the platform nor the ownership is there yet. Diagnose, then build both together.
Fragile platform
Good data without owners decays. Put definitions, stewards, and decision rights in place.
Data state →
How we define results
Results are outside the data team. Costs are inside.
A deployed platform is not a result. We agree indicators at five levels before we build, and we keep internal costs down: less duplication, less manual work, less vendor lock-in, and fewer security gaps.
Outside: results
- Strategic
- Operational
- Technical
- Governance
- Cultural
Inside: costs
Infrastructure, licences, staff time, duplicated reports, manual work, and vendor lock-in. Keep them low.
Strategic
- Leaders work from a small set of trusted, organization-wide measures
- Decision cycles get shorter
Operational
- Less time spent reconciling data
- Duplicate reports retired; inventories often shrink by a third or more
- External reporting that is repeatable and auditable
Technical
- Business logic encoded once, in a shared semantic model
- Less technical debt and vendor dependence
Governance
- Approved, documented definitions with named owners
- Fewer disputes over numbers in meetings
Cultural
- Leaders ask for insights, not extracts
- Analytics becomes proactive rather than reactive
Principles
“Information is data endowed with relevance and purpose.”
- 1
Purpose before platform
We start with your mission and the public value it creates, then choose the architecture. Technology follows purpose.
- 2
Build on strengths, not silos
IT, research and planning, and the operational departments each hold part of the answer. Shared governance makes those strengths productive together, and research joins on day one.
- 3
Results are outside; costs are inside
Success is measured by what changes for the people you serve and the leaders who decide, while keeping internal cost, duplication, lock-in, and risk down.
- 4
Information is data with relevance and purpose
Raw data becomes information only when definitions, business rules, and context are applied. That is why we invest in shared definitions and one semantic model.
- 5
Human sign-off on AI
Nothing ships, and no AI output reaches the people you serve, staff, or the board, without a named person approving it.
- 6
You own it at the end
Your staff build alongside us and receive runbooks and training. The platform, and the skills to run it, are yours.
Ways to work together
Project
A fixed scope with a fixed fee and a timebox. Best for assessments, builds, and pilots.
Embedded
Our architect and engineers work inside your team on your priorities for a set period.
Fractional
Senior architecture leadership a few days a month, on a retainer.
Start with a 30-minute readiness call
Tell us what you are trying to fix or build. You will talk with the architect who does the work.
Led by Victor Manchik, founder and principal data architect