AI Products
Why a separate AI product layer is needed
Many companies adopt AI through scattered pilots: one department launches an assistant, another builds RAG, a third buys a developer tool, a fourth automates a process with LLM.
At first this looks like progress, but it quickly creates chaos:
- duplicate solutions appear;
- the same task class is implemented in different ways;
- pilots are not reused;
- it is unclear which AI tools already exist;
- the business asks for a new product for every case;
- IT, security, and architecture repeatedly review similar solutions;
- impact is hard to measure because the product base is not structured.
The operating model solves this through AI product portfolio management.
What is an AI product
An AI product is an internal solution, service, platform, or functional capability that uses AI and can be applied to one or several business tasks.
It should have an owner, target audience, task class, usage rules, support loop, quality criteria, security constraints, and links to initiatives and business impact.
An AI product is not always a large standalone system. It can be a platform, an applied service, a set of functions inside an existing system, or a reusable agent scenario.
Two AI product levels
Platform AI products
Platform products are base technology capabilities on top of which applied solutions and initiatives are built.
| Product | What it gives the company |
|---|---|
| Corporate LLM | Shared access to models, prompts, policies, logging, and guardrails |
| RAG platform | Search and answer generation over corporate knowledge |
| ML platform | Training, deployment, and monitoring of ML models |
| Code agent | Faster development, code analysis, testing, and documentation |
| Agent runtime | Environment for running, controlling, and monitoring AI agents |
Platform products answer:
Which base AI capabilities does the company have?
Applied AI products
Applied AI products are services built on top of platform capabilities and oriented toward specific business scenarios. They can use several platform products at once.
| Product | Built on | Task class |
|---|---|---|
| Knowledge base assistant | LLM + RAG | Document Q&A |
| Document generator | LLM + templates + workflow automation | Standard document preparation |
| Project manager assistant | LLM + RAG + MCP Atlassian + agent orchestrator | Statuses, minutes, risks, plans |
| Inquiry analyzer | LLM + ML classification + integrations | Inquiry analysis, routing, prioritization |
| Developer assistant | Code agent + repositories + docs | Development, review, tests, documentation |
| Compliance agent | LLM + RAG + control rules | Text, document, operation checks |
Applied products answer:
Which recurring business tasks can the company already solve with AI?
Relationship logic
An AI initiative should not automatically become a separate product.
Business need
↓
AI initiative
↓
Select suitable AI product
↓
Implement through a platform or applied product
↓
Reuse the result in other initiatives
A unique departmental task can be implemented as a configuration of an existing product. A recurring need across departments can become an applied AI product. A missing base capability needed by many applied products should drive development of a platform product.
Example
Bad approach: every department asks for its own RAG service: legal, HR, procurement, risk, IT. The company gets five similar solutions, five support loops, five architecture reviews, and five security approaches.
Right approach: the company develops one RAG platform and builds applied assistants on top of it for legal, HR, procurement, risk management, and IT support.
When an initiative becomes a product
Not every initiative should become a product.
It can remain a one-off project if the task is unique, the scenario does not repeat, no other departments need it, the effect comes from one-time automation, and support is not justified.
It can become an applied AI product if the scenario repeats across departments, the solution is reusable, there is a stable user group, a product owner is needed, regular improvements appear, and support/training/evolution are required.
It can lead to a platform product if a base technology capability is missing, the same component is needed by other initiatives, integration/data/security/monitoring needs repeat, or the company wants to accelerate future delivery.
Head of product portfolio role
The head of product portfolio ensures the AI product portfolio is managed, understandable, and useful for the business. The role answers: which products exist, which are used, which initiatives use them, which duplicate each other, which should evolve or close, which new platform capabilities are needed, and which applied products should scale.
AI product catalog
A minimum product card includes:
| Field | Description |
|---|---|
| Product name | Internal product name |
| Type | Platform or applied |
| Product class | LLM, RAG, ML, automation, code agent, agentic system, etc. |
| Owner | Who owns product development |
| Target audience | Who the product is for |
| Main scenarios | Which tasks it solves |
| Used platforms | Which base capabilities it uses |
| Status | Idea, pilot, production, evolution, retirement |
| Constraints | What the product must not do |
| Integrations | Which systems it connects to |
| Initiatives | Which initiatives use the product |
| Metrics | Usage, quality, impact, cost |
Product statuses
Product idea
→ Feasibility evaluation
→ Pilot
→ Production
→ Scaling
→ Evolution
→ Retirement
Platform products are evaluated by reliability, security, scalability, usage cost, availability to teams, and speed of onboarding new scenarios.
Applied products are evaluated by user value, adoption, result quality, effort reduction, business impact, and scenario repeatability.
Strategic portfolio decisions
Important portfolio decisions include:
- Investment choices — which AI products to grow and which to postpone.
- Task-to-product mapping — which use case class each product serves.
- Build / Buy / Partner — internal development, buying a solution, or partnering.
- Product maturity management — from experiment to scaled adoption.
The portfolio strategy should be reviewed at least quarterly based on new technologies, business priorities, initiative results, and maturity of existing products.
How AI products connect to initiatives
AI products and AI initiatives are different entities. An initiative describes a business need, value hypothesis, and implementation path. A product describes the reusable capability through which the need can be implemented.
One product can be used in several initiatives. One initiative can use several products.
Example initiative: reduce time to prepare analytical briefs. Used products: LLM, RAG, analyst assistant.
Key principle
The company must manage not only the flow of AI ideas, but also the portfolio of AI products through which these ideas are implemented.
Without this, AI adoption becomes a set of disconnected pilots. With it, the company gets less duplication, faster delivery, clearer architecture, managed security, reuse, and transparent accountability.