Authorized evidence
Documents, data, records and context made available for the engagement.
Broader analytical coverage.
Accountable judgment.
Alkemia’s cognitive agentic AI system: specialist agents, governed memory and proprietary domain knowledge supporting CIO, CFO and board decisions with traceable evidence.
Technology expands the analysis. People remain accountable for the judgment.
The Alkemia Decision Intelligence System is the cognitive agentic AI system Alkemia built and operates. The term “cognitive” is used in the sense of research on language agents: memory and knowledge held outside the model and a cycle in which the agent proposes, evaluates and selects actions. Alkemia adds one requirement to that cycle: every change an agent proposes to knowledge or method takes effect only after something outside that agent authorizes it (coded rules, a separate auditor agent or a person, depending on what is at stake), and the change is recorded. In every advisory engagement, specialist agents cover workstreams that would normally take several professionals. In a Cloud ERP engagement, those workstreams are licensing, contract negotiation with the vendor, sizing and infrastructure, architecture and conversion, business case and TCO, implementation-partner due diligence and market research. With the core of its delivery built on the system, Alkemia is an AI-native firm. The system organizes client-authorized evidence, mobilizes specialist analysis and preserves the chain from sources and assumptions to trade-offs and recommendation.
Alkemia works for whoever decides and signs.
Evidence remains connected to the decision it supports.
The system preserves context between what was observed, what must remain true and the decision at stake.
-
Authorized evidence
Documents, data, records and context made available for the engagement.
- Orchestration core
Alkemist
-
Specialist analysis
Domain agents examine the material against explicit criteria and defined boundaries.
-
Decision record
Evidence, assumptions, exposure and trade-offs remain connected.
-
Human judgment
Alkemia interprets the context and remains accountable for the recommendation.
-
Governance and review
The decision record identifies who holds authority, when the recommendation remains valid and when review is required.
Orchestration core
Alkemist mobilizes the right intelligence without erasing provenance.
Alkemist is the system’s orchestration core. It directs each question to the appropriate specialist agents, retrieves permitted context and keeps the analysis connected to its supporting sources.
- Domain-specialist agents
- Memory separated by context and access
- Alkemia method and proprietary domain knowledge applied without mixing client contexts
The system makes visible the basis, boundaries and authority behind each decision.
Each element remains connected to the part of the decision it supports.
Provenance
Where each piece of evidence came from and which part of the decision it supports.
Assumptions
What must remain true for the recommendation to stay valid.
Exposure and reversibility
Where commitments become binding, what exit would cost and which windows remain open.
Decision rights
Who recommends, approves and executes, and when the decision should be reviewed.
What supports “cognitive” and “AI-native”.
Mechanisms in operation in the system today.
- Choosing among alternatives
- In the intelligence cycle, the domain agent evaluates each market signal and chooses where it goes from a defined set of alternatives, such as discarding it, keeping it as a source, turning it into an intelligence note or using it to propose a change to the method. Coded rules refuse a choice that lacks the required proof.
- Whoever drafts does not approve
- Intelligence notes drafted by one agent go to a separate auditor, based on a different model family, which can only accept or reject them. Changing the method requires human approval.
- A record that is never deleted
- Automated changes the intelligence cycle makes to the knowledge base enter a permanent record, failures included.
- Explicit provenance
- Over 90% of the knowledge base carries a source reference (measured October 2026), and superseded knowledge is archived and preserved.
- Context isolated per client
- Each client’s material sits in its own index, beyond the reach of general search.
One system. Different decisions.
Each practice applies the same discipline of evidence, context and accountability to the executive moment at hand.
Strategy Advisory
Turns executive ambiguity into criteria, options, sequence and a defensible direction.
Enterprise Apps
Connects technical, commercial and contractual evidence to Cloud ERP and core-modernization decisions.
Enterprise AI
Defines where AI should operate, using which criteria and when to stop or reassess.
Where the system’s role ends.
Three limits define that role.
- Alkemia remains accountable for the recommendation.
- The client retains the final decision and its decision rights.
- The system operates within the evidence and access authorized for the engagement.
Frequently asked questions
What the system is and what it is not.
- Is the Alkemia Decision Intelligence System software?
- Yes. It is proprietary software, with specialist agents, governed memory and Alkemia’s proprietary domain knowledge. Today it supports every Alkemia advisory engagement, with each client’s material kept in an isolated index. It does not operate as autonomous decision software.
- What does it mean to call the system cognitive?
- Here the term is used in the sense of research on language agents (Sumers et al., Cognitive Architectures for Language Agents, TMLR, 2024): memory and knowledge held outside the model and a cycle in which the agent proposes actions, evaluates the alternatives and selects the one it executes. Alkemia adds an authorization and record requirement to that cycle. Any system that calls itself cognitive can be tested this way: two questions check the research sense (where memory and knowledge live, and how the agent chooses among alternatives), and a third checks the requirement Alkemia adds (who authorizes and records a change to the knowledge or the method). In Alkemia’s system, memory and knowledge live outside the model, in the knowledge base, in per-client indexes and in versioned records; in the intelligence cycle, the domain agent chooses where each market signal goes from a defined set of alternatives; and a change to the knowledge or the method takes effect only once coded rules, a separate auditor or a person, depending on what is at stake, has authorized it, and it is recorded.
- What does it mean that Alkemia is AI-native?
- An AI-native firm builds the core of what it delivers on AI agents, governed memory and proprietary knowledge. At Alkemia, that holds in every advisory engagement: specialist agents coordinated by Alkemist cover workstreams that would normally take several professionals (in a Cloud ERP engagement, from licensing to implementation-partner due diligence). Research, knowledge curation and method maintenance also run on the system.
- Does Alkemia have commercial relationships with vendors?
- None with SAP or with SAP integrators. Alkemia maintains commercial relationships with other technology companies, including paid referral agreements, and before evaluating a company it has a relationship with, it discloses that relationship to the client in writing. It receives no remuneration from vendors involved in the decisions it advises on, and its fees do not depend on the vendor chosen.
- Does Alkemist make decisions for the client?
- No. Alkemist orchestrates specialist analysis and context. Alkemia remains accountable for the recommendation, and decision authority stays with the client.
- What information does the system use?
- Only information made available and authorized for the engagement, combined with the relevant proprietary domain knowledge. The analysis preserves the connection between evidence, assumptions and recommendation.
- How does the system change advisory work?
- Every workstream the subject requires gets specialist coverage within the same engagement, and the recommendation arrives with its provenance, assumptions, exposure and decision rights made explicit.
Before committing capital, architecture or contract, make the decision traceable.
The initial conversation clarifies the decision, the exposure and the sign-off criteria.
Discuss a decision