ProductsEvals

Built around your institution

Know your AI works the way your firm works.

Evaluate one AI team against agreed criteria and your institution’s past work before it takes on live work.

Discuss Evals What you receive

Start with Evals as the first use of a full institution-specific foundation.

The question behind Evals

A plausible answer can still miss your standards.

An AI team can produce a plausible answer and still miss your sourcing standards, valuation conventions or methodology. Aperture evaluates its work against agreed criteria and your institution’s historical research.

The first deliverables

Begin with a clear evaluation.

  1. Agreed criteria

    Your reviewers’ standards, definitions and methodology, written down as evaluation criteria.

  2. A first evaluation

    One AI team assessed against those criteria and a set of your own past work, before it takes on live work.

  3. A findings report

    Where the AI team meets the standard, where it falls short and where human review should stay.

Evaluation continues for the agreed AI team and criteria as agents and models change. New criteria, additional AI teams and one-off private evaluations are scoped separately.

Illustrative example

Your reviewers set the standard.

A deal team wants an AI team to draft sections of an investment memorandum. Its senior reviewers set the sourcing standards, valuation conventions and house style. Draft sections are assessed against those criteria and the team’s past memoranda.

  1. The institution defines

    Sourcing standards, valuation conventions and house style.

  2. The work is assessed

    Draft sections against agreed criteria and past memoranda.

  3. Reviewers decide

    What is ready, what needs correction and what remains under human review.

The findings show what meets the agreed criteria, what needs correction and where reviewers should remain involved.

Illustrative findings excerpt

A finding your reviewers can act on.

Fictional case, figures and references. This illustrates the report format, not a client result or measured performance.

Agreed criterion
Use the agreed enterprise-value definition: equity value plus debt, less cash.
Observed work
The draft reports an enterprise value of $120 million.
Finding
Needs correction
Reviewer action
Reconcile the $10 million difference before the draft is used. The reviewer decides whether the corrected work meets the agreed criterion.
Inspect the worked comparison

Illustrative case inputs: equity value of $100 million, debt of $30 million and cash of $20 million.

$100m + $30m − $20m = $110m

Under this example’s agreed definition, the result is $110 million. The draft’s $120 million figure does not match. Actual definitions, permitted references and review criteria are agreed with the institution.

One institution-specific foundation

Start with the capability you need. Build on the same foundation as you grow.

Your licensed and internal data, connected and structured with permissions, checks and a retained record. New capabilities reuse those connections, checks and records, with additional work scoped as needed.

Each capability builds on the same institution-specific foundation.
EvalsCurrent capabilityRecord & ProvenanceCompliance & GovernanceSynthetic DataAI Teams
Your institution’s foundation
  • Licensed & internal data
  • Connections & permissions
  • Checks & retained record

Reviewed research can be retained in that foundation, so the next capability builds on what your institution already knows.

You control the foundation and own what is built on it.

Own agreed institution-specific deliverables under your contract. Aperture’s pre-built core is licensed on terms set in your contract. Third-party source and model licenses apply; hosting does not transfer model IP.

Under your contract, your data never trains AI models outside your institution.

A private briefing

Start with your institution’s work.

Discuss Evals