How Saphenia is priced and what sets the cost of a deployment

How much does Saphenia cost?

The cost of a Saphenia deployment depends on where your firm sits on the three levers. Two firms that both want an agent to run portfolio reviews can bring different volumes and kinds of data, and one pricing model covers both.

Lever 1: the size and complexity of your data

The data lever covers everything Saphenia connects to and configures: your firm's own record of facts, definitions, and history that every decision reads from. More connected data costs more, and complexity counts as much as volume.

Four things make data complex, and the decisions Saphenia takes on usually have all four:

• The data sits in files that are hard to read: contracts, filings, and feeds, each arriving in a different shape.
• The systems were built separately, and one answer needs several of them.
• The rules change while the work runs, with policies revised mid-task and regulation that differs by country.
• The history goes back years and still applies. Trading records and engineering specs are two examples.

A deployment that reads one clean reporting database sits low on this lever. One that reads scanned board packs, a loan system, a policy library with superseded versions, and a decade of records sits high on it.

Lever 2: the number of decisions

Each decision you hand to an agent adds to the price. It holds the decision logic, rules, and captured expert corrections for one workstream.

You start with one recurring decision, have it working in weeks, and add the next when you're ready. Once the sources are connected, Saphenia builds the decision logic in days. The decision keeps up as your data and rules change, so you pay for a workstream that stays current.

A second decision reads from the same record as the first. So if it needs only sources you've already connected, the data lever stays where it was and only the decisions lever moves.

Lever 3: bespoke agent or integration work

A source that needs a new connector adds to the quote, and so does an agent Saphenia builds for you. Saphenia runs with the model your firm chooses. A team that already runs an agent for this workflow keeps it, and the deployment connects that agent to the decision logic.

A worked example of the three levers

Take a made-up mid-sized lender whose Head of Credit wants an agent to handle loan product matching, picking which of the lender's products fits an applicant and why. Their lending criteria live in a policy library that the credit committee revises during the year. Applicant data sits in a loan origination system, while supporting documents arrive as PDFs and spreadsheets in a shared drive.

Here's where each lever sits for the first engagement:

• Data. Mid to high: moderate volume, high complexity. The supporting documents are unstructured, and the policy library keeps superseded versions beside current ones. One answer needs both the loan origination system and the shared drive, two systems built separately.
• Decisions. One, for loan product matching.
• Bespoke work. The lender has no agent in production for this decision, so the quote includes building one. The loan origination system already has a connector, which keeps the integration line small.

A year later, the same lender wants credit monitoring on their existing loan book. That second decision reads the loan system and policy library already connected, plus a new market data feed. The decisions lever moves by one, and the data lever moves by the new feed alone. If the feed needs a new connector, that adds a small integration line. Whether credit monitoring needs an agent of its own sets the bespoke line.

What affects the cost of deploying an AI agent platform at a bank?

At a bank, the same three levers set the price of a Saphenia deployment. Before it goes live, Procurement and InfoSec review the deployment, and at some banks Model Risk or Compliance does too. Those reviews set how long the purchase takes and sit outside Saphenia's price.

Saphenia runs in the environment you choose, so settle hosting early. The first call confirms how your choice enters the quote. Saphenia's ISO 27001 and SOC 2 certifications are in progress. The other facts InfoSec teams ask for are in security, hosting, and data ownership.

What to bring to the first call for an accurate quote

Saphenia scopes a quote from five pieces of information:

• The decision. Name the one recurring decision you want an agent to make first, and how often it runs.
• The sources. List every system, document store, and file type that decision reads from, with rough volumes and how far back the history goes.
• The rules. Say where the policies and criteria live and how often they change.
• The agent. Say whether you already run an agent for this work or want one built.
• The environment. Say where the deployment has to run and which reviews it must pass.

If you're weighing a purchase against building in-house, see the full cost of an internal build. For a first engagement on portfolio reviews, the guide to evaluating an AI portfolio review pilot covers what to test on your own documents.

How to frame the cost in a business case

The number to set against a Saphenia quote is what the decision costs you today. Someone on your team checks by hand every output an assistant drafts, because a wrong answer and a right one look the same. Multiply that checking time by how often the decision runs, and you have the cost that Saphenia cuts. Each answer arrives with its proof, and a person handles only the calls reserved for a human. The business-line owner who carries the decision can usually put a figure on that cost.

Ownership is the second line in the business case. Everything Saphenia builds from your data stays yours: the record of facts and corrections, the decision logic, the rules, and the proofs. All of it works with any model, and you keep it if you leave Saphenia. The money you spend on the first deployment builds that record, which every later decision reads from. The next model your firm adopts can use it from day one.