Why does defining a data product take so much time?
The difficult part of data product delivery often begins before anyone writes code or builds a pipeline. A business team identifies an opportunity, discusses objectives, outcomes, available information, and possible solutions, then produces notes, presentations, spreadsheets, and actions.
Someone must turn that material into work a product team can evaluate and deliver. What is the business objective? Which use cases support it? Does the organization already have relevant products? Which candidates are needed? Who owns them? What is missing, and which governance requirements apply?
These questions take time because information is scattered, relationships are unclear, and stakeholders use different terminology. Traditional delivery relies on people collecting documents, interpreting discussions, preparing specifications, and coordinating reviews.
AI-assisted delivery changes how the preparation is done. Rather than starting with an empty document, agents help transform existing material into structured candidates, identify relationships and gaps, and prepare work for human review. The objective is not to let AI decide business needs; it is to reduce manual effort needed to prepare reliable information for those decisions.
How traditional data product delivery works
Consider a business unit trying to improve customer retention. A workshop brings together managers, analysts, engineers, and product owners to discuss the problem, existing customer information, possible indicators, and expected outcomes. Afterwards, someone must turn the discussion into a usable definition.
They clarify the objective, separate use cases from technical solutions, examine existing data assets and products, document missing capability, then consider ownership, dependencies, quality, access, and governance. Information often ends up across presentations, spreadsheets, product templates, Jira tickets, and technical documentation.
Experienced product managers and architects know how to translate needs into delivery requirements, but much of their time goes to collecting, restructuring, and reconciling information instead of evaluating the decisions that matter. When a use case, ownership record, or governance requirement changes, they must update several places and keep stakeholders aligned.
The bottleneck is not necessarily technical expertise. It is manual coordination between business demand and product delivery.
What AI-assisted delivery changes
A meeting transcript, workshop summary, requirements document, or set of use-case descriptions becomes input to a structured workflow. Agents interpret the material and identify possible objectives, use cases, candidate products, dependencies, and unresolved questions.
The output should not be a longer summary. It should be a structured representation of work the organization can evaluate. A retention workshop might yield one objective, three use cases, and several candidates. A workflow can separate those concepts, compare candidates with portfolio information, and highlight missing details.
The product manager reviews the result, corrects interpretations, confirms ownership, and decides which candidates deserve attention. AI accelerates decision-ready preparation; it does not prove that source material is correct, that a candidate has commercial value, or that development has been approved.
A banking example: from workshop notes to product candidates
A bank runs a customer-retention workshop with customer operations, analytics, data management, and technology. The group identifies a strategic objective: earlier identification of disengagement. It proposes use cases for finding declining engagement, targeting retention actions, and evaluating their outcomes.
The bank has customer profiles, transactions, and digital-engagement data, but the participants do not have a complete portfolio picture. At the end of the meeting, they have notes, a transcript, and possible next steps.
The manual approach
A product manager identifies the objective, separates use cases, consults the catalog, speaks with owners, and reviews documentation. They find that Customer 360 Profile exists but does not provide all required information. Customer Engagement Signals may exist elsewhere, but ownership and lifecycle status need clarification. They prepare Customer Retention Indicators as a candidate, document dependencies, identify governance requirements, create work items, and schedule reviews.
Eventually the team has enough structure to decide whether to move into delivery. Depending on complexity and information quality, organizing and validating what the organization already knows can take multiple working days across several stakeholders.
The AI-assisted approach
In a connected operating environment, the meeting material enters an AI-assisted workflow. Agents identify the retention objective and use cases, prepare candidate definitions, and examine the portfolio for related capabilities. Because the portfolio connects objectives, use cases, products, owners, dependencies, and lifecycle states, agents have more context than the workshop notes alone provide.
The workflow identifies Customer 360 Profile, finds Customer Engagement Signals, and flags the need to verify suitability, ownership, and availability. It prepares Customer Retention Indicators with proposed dependencies, quality considerations, governance questions, and visible gaps: no approved business owner, unclear intended consumers, and potential privacy and access review.
It records those questions rather than inventing answers. The result is a structured package of objectives, use cases, candidates, relationships, assumptions, outstanding questions, and proposed next steps. Stakeholders decide whether candidates advance, merge with existing products, or are rejected. Agents prepare the decision; they do not make it for the bank.
Why a generated document is not enough
Generative AI can turn notes into a convincing requirements document, but a document alone does not establish relationships to existing products, maintain lifecycle state, identify current ownership, or link later decisions back to a business objective. It does not prevent duplicate definitions or ensure stakeholders are using the same current information.
If each AI interaction produces an independent document, organizations can accelerate documentation while increasing fragmentation. A better approach turns extracted material into managed objects and relationships: an identifiable objective, connected use cases, candidate products tied to them, and references to existing products and dependencies.
Governance requirements and decision evidence remain attached to the relevant work. When a use case changes, teams can inspect its relationships to candidates, products, owners, and delivery commitments. The improvement is not faster document generation; it is moving from unstructured material to connected, manageable product work.
Human review must remain part of the process
An agent can mistake an example for a confirmed requirement, an idea for an approved objective, or an assumed dependency for an existing one. It can also propose a technically plausible product with insufficient business value. These errors become serious when generated output enters formal delivery without review.
Governed delivery distinguishes extracted facts, proposed interpretations, assumptions, and approved decisions. A product manager decides whether a candidate reflects the need. A proposed owner or authorised stakeholder confirms accountability. Governance functions determine which requirements apply and whether they are satisfied.
Human review is not a final checkbox on an autonomous process. It is part of the model that determines how candidates become approved products and proposals become authorised actions. The required review should reflect the decision: routine drafting can have less intervention, while investment, access, sensitive data, and production transitions need stronger accountability.
Governance should begin with the first product candidate
Traditional delivery often discovers late that ownership is unclear, sensitive information needs additional controls, or a proposed access method conflicts with policy. AI-assisted workflows can introduce governance earlier by examining a candidate's initial context.
Does it have an owner? Is the purpose clear? Are dependencies identified? Does use involve sensitive information? Are quality expectations documented? Are policy questions unresolved? Agents should not invent answers or declare compliance. They can make gaps visible while a product is still being evaluated.
This is Minimum Lovable Governance in practice: the smallest useful set of ownership, policies, controls, evidence, and review requirements follows the product through its lifecycle. At candidate stage, teams need enough information to understand responsibilities and constraints; during delivery and before operation, they need stronger evidence on implementation, quality, access, contracts, controls, and approvals.
Why a connected portfolio matters
Without portfolio context, an agent works primarily from input documents and can overlook an existing product that already serves the need. A connected portfolio gives it information about products, objectives, use cases, dependencies, ownership, governance requirements, and lifecycle states.
The agent may find that two proposed use cases depend on the same product, that a candidate overlaps work already underway, or that a solution depends on a product due for retirement. These observations help managers evaluate reuse, dependencies, and risk before commitments are made.
The portfolio knowledge graph provides relationships, machine-readable product specifications provide structured definitions, and governance and lifecycle information provides operational context. Results still depend on current, complete underlying information; outdated ownership or incomplete dependencies can produce incorrect recommendations.
From product candidate to managed delivery
Preparing a candidate is only the beginning. Once the business chooses to proceed, the product needs an owner, lifecycle state, technical responsibilities, quality expectations, governance requirements, confirmed dependencies, interfaces, contractual commitments, and organised delivery work.
Agents can help prepare specifications, identify incomplete requirements, and structure proposed delivery. A definition might provide the basis for an implementation backlog, but creating a backlog does not start implementation and generating a specification does not approve a product.
The operating model must preserve the difference between proposed work, approved work, active delivery, and completed outcomes. A connected portfolio gives continuity between initial business discussions and later delivery, monitoring, and review as priorities change.
Standardised agent workflows reduce inconsistency
Teams repeatedly interpret requirements, analyse use cases, examine capabilities, prepare definitions, identify governance gaps, and organise review. Without shared workflows, they use different prompts, templates, assumptions, and validation practices, making AI-assisted quality difficult to evaluate.
Standardised recipes define purpose, expected inputs, tools, stages, validation, outputs, operational permissions, and review points. A candidate-generation recipe can identify objectives, extract use cases, examine existing products, propose candidates, identify missing information, and prepare results for review. Another can focus on governance gaps or portfolio impact.
Recipes do not guarantee perfect output. They make recurring work more repeatable, inspectable, and manageable.
What Maysano changes
Maysano connects AI-assisted product delivery to a managed business and data product portfolio. It connects objectives, use cases, products, ownership, dependencies, governance, lifecycle, versions, and delivery context through a shared knowledge graph.
Portfolio Studio supports turning business input into structured product candidates and portfolio work. Agents can examine source material, identify concepts, prepare definitions, analyse relationships, and identify incomplete information. Results remain connected to the portfolio rather than becoming isolated documents.
Stakeholders can review candidates, refine definitions, evaluate relationships, and decide what proceeds. Standardised externally configured recipes, explainable operations, auditing, and operational controls combine AI-assisted preparation with governed review. Open standards under the Linux Foundation's LF AI & Data umbrella provide machine-readable definitions for products, contracts, and related operational information; Maysano adds connected portfolio and operating context around them.
What Maysano does not replace
Maysano does not replace business owners, product managers, architects, engineers, or governance reviewers. AI-generated candidates remain proposals until reviewed and accepted through the appropriate process. It does not replace warehouses, lakehouses, integration platforms, catalogs, metadata platforms, or delivery tools such as Jira.
Those systems continue their established responsibilities. Maysano connects business and product context to the work, supporting preparation, organisation, governance, and monitoring. It does not establish business truth through AI alone: objectives require confirmation, priorities require decisions, dependencies require evidence, and governance remains subject to enterprise authority.
How should an organization introduce AI-assisted delivery?
Begin with a recurring activity where teams already spend substantial time interpreting business material and preparing product work, such as the work after a business workshop. Select a real objective and use cases, gather relevant notes, requirements, and portfolio information, then use AI assistance to produce an initial structured set of candidates and relationships.
Compare the result with what an experienced product manager would normally produce. Check whether the workflow identifies objectives, separates use cases from products, recognises existing capability, exposes missing information, and preserves governance questions. Examine how much human correction is necessary.
Fast output that invents requirements is not an improvement. A useful implementation reduces repetitive preparation while preserving or improving decision quality. Once it works for a limited set of cases, extend it to specifications, governance analysis, dependency assessment, and backlog preparation, while keeping transitions between proposals, reviewed definitions, approved work, and actual delivery clear.
The future of data product delivery begins before implementation
Data product delivery will always need business judgment, technical expertise, governance, and accountable decisions. AI does not remove those responsibilities; it changes how much effort teams spend preparing information before exercising that judgment.
AI-assisted workflows can transform business discussions into structured candidates, connect them to existing portfolio information, identify gaps, and prepare work for review. The lasting value comes afterwards, when objectives, use cases, products, ownership, governance, lifecycle, decisions, and delivery remain part of one managed operating model.
The future is not simply faster documentation or more autonomous agents. It is a managed process that moves business intent into structured, governed product work with less manual coordination and stronger continuity between decisions.
Further reading
Open Data Product Specification