Consulting · AI Systems
Own Your Data
The model is replaceable. The data is not.
A firm’s advantage in AI is not the model, which everyone can rent, but the data only it holds: its deals, its credit files, its research, its client history. The first job of an AI program is to make sure that advantage is used without being given away, which means knowing where data goes, what vendors may do with it, and how models are allowed to read it.
Where data goes
Every hosted API call carries the prompt, and often the firm’s documents, to a third party. Enterprise contracts commonly promise not to train on customer data, but retention, logging, sub-processors and jurisdiction vary and change. The default posture should be that sensitive data does not leave the firm’s environment, and that any exception is a documented decision with a contract behind it.
Retrieval, not absorption
Most business value comes from a model reading the firm’s documents at the moment of a question, not from a model that has memorized them. Retrieval architectures index internal data in a store the firm controls, fetch the relevant passages for each request and pass them to the model with the question. The model learns nothing permanent; the firm keeps the knowledge, the permissions and the audit trail.
Permissions and lineage
An AI system that can read everything is a data breach waiting for a prompt. Retrieval must respect the same access controls as the source systems, so a user sees only what they could already see. Every answer should be traceable to the documents it drew on, both for the user’s trust and for the record regulators and auditors will ask for.
Data as an asset
Cleaning, structuring and labeling internal data is unglamorous and is where most of the value is created. A firm that has done it can switch models freely, evaluate them honestly and, when it chooses, train on its own knowledge. A firm that has not is dependent on whichever vendor holds the messiest copy.
Practice
How Wesvel approaches it
- 01A data map is produced first: what exists, where, how sensitive, who may see it.
- 02Sensitive data is processed by models inside the firm’s environment or under contracts that prohibit training and mandate deletion.
- 03Retrieval systems enforce source permissions and record which documents informed each answer.
- 04Data preparation is scoped and funded as its own workstream, not assumed.
More in AI Systems
- Legacy LLMs vs Open-Weight ModelsWhat hosted frontier models and open-weight models each do well, what they cost in money and control, and how to use both without being captured by either.
- Training Your Own ModelsWhen to fine-tune or distill an open-weight model on internal data, how to do it without leaking it, and how to prove the result is better.
- Building Internal SystemsThe architecture around the model: orchestration, tools, retrieval, evaluation, monitoring, security and the infrastructure to run it.
- Governance and ComplianceModel risk management, human oversight, record-keeping, vendor diligence and disclosure for AI in regulated financial businesses.
- AI in Financial OperationsThe use cases that hold up in investment, lending and advisory operations, and the ones that do not, drawn from how Wesvel uses these systems itself.