AI Transparency

Roiva is a platform for measuring and tracking the ROI of AI initiatives. We use third-party AI services to power in-app features. This page explains which services we use and how.

AI services we use

Anthropic Claude

We use the Anthropic Claude API for every AI feature in Roiva. Each one is listed below with what it sends. Every request that carries an account's data goes through one check, which refuses it while the account has Aggregate figures only turned on. So every feature in the first list is off for such an account, and Roiva sends nothing from the account to an AI model. An AI assistant of your own, connected over MCP, is yours rather than Roiva's, and is covered below.

Features that send your account's data

• The Roiva Assistant, the in-app chat. It sends your question, the conversation so far, and the account data it looks up to answer, such as initiative details, costs, cash value, freed capacity, metric readings and ROI figures, and the names or email addresses of an initiative's owner and approvers. It answers questions about your portfolio ("Which initiatives have the highest planned ROI?"), explains ROI calculations, helps draft initiative descriptions and hypotheses, and suggests improvements to value formulas. Off under Aggregate figures only.
• Setting up a new initiative started without a template: matching it to a library template, or proposing its expected figures, baselines and value formula. It sends the initiative's title, category, description and hypothesis, the assessment answers it was created from, the account's assumption figures, and the metric definitions that best match it, including any the account added. Off under Aggregate figures only.
• Estimating baselines a template's formulas need, once the account's own history has supplied what it can. It sends the template, your assessment answers, and each missing metric's definition and the formula that reads it. Off under Aggregate figures only: you record a baseline your answers don't give.
• Template recommendations. It sends the description someone types of your business and its problems, the account's industry, the names of the systems it has connected, and the titles of up to 25 of its initiatives. Off under Aggregate figures only.
• Reading documents you upload, such as briefs, decks and budget sheets, to propose an initiative's details, costs and metrics for you to review. It sends the text read out of the files and the metric definitions that best match it. Off under Aggregate figures only, which keeps no uploaded file. Importing a spreadsheet register of initiatives reads the sheet without AI; each initiative it creates is then set up as in the second item above, unless Aggregate figures only is on.
• Reading a business case for its planned figures, when someone uploads one on Edit Business Case or picks one there from the initiative's documents. It sends only the text read out of the file, and the figures it finds are offered beside the form's fields for a person to use or ignore. Off under Aggregate figures only.
• Reading a vendor's bill for its costs, when someone uploads a billing export, invoice, receipt or screenshot on an initiative's Costs tab. A spreadsheet's header and first rows are sent, and Roiva reads the rest of its rows itself; a PDF is sent as its text, or as the file where it holds none, and an image is sent as it is. The lines it finds are proposed for a person to tick and confirm, and nothing is booked until they do. Off under Aggregate figures only.
• Drafting Board Brief commentary, when someone asks for a draft of a quarter's management commentary. It sends the quarter's figures as the brief has them: the net-cash bridge and its totals, and the stage changes, go-lives, portfolio decisions with their recorded reasons, budget changes and new recurring cost lines, by initiative name. Off under Aggregate figures only.

Features that send no account's data

• The question box on Roiva's public documentation. It sends the question and the documentation pages that answer it, and is never handed an account, so Aggregate figures only doesn't apply. Don't type your figures there.
• Roiva's own upkeep. Checking integration setup guides against each vendor's public documentation, and checking the published sources behind Roiva's default assumptions for newer figures. Neither reads any customer's data.

Anthropic's privacy policy is available at anthropic.com/privacy .

Accuracy of figures

The assistant is instructed to quote cost, value, ROI, net and payback figures only as Roiva's reports return them, and not to calculate figures of its own. Its ROI figures come from the same calculation as Roiva's ROI pages. It is also instructed to link each figure it quotes to the Roiva page it came from. A language model can still misread or misquote a figure, so check it on the linked page before relying on it. The assistant panel says the same.

The assistant is instructed to restate a cost or a metric reading and wait for the user to confirm it before recording it. Costs it records are marked as recorded through the assistant on the cost list, in the initiative's activity and in the audit log. Value entries it records are saved as drafts, which count in no cash value, freed capacity or ROI until a person approves them.

How your data is used

We do not use your data to train AI models. Data you enter into Roiva — including initiative details, metrics, and connected integration data — is not used to train or fine-tune any AI model, by Roiva or by our AI service providers.

What each feature above sends Anthropic is limited to what it needs, and is processed solely to fulfill that request. It is subject to each provider's data processing terms linked above.

Google Workspace, Gemini and Google Cloud integrations

When you connect Google Workspace, Gemini in Google Workspace or Google Cloud, Roiva reads calendar events, Google's Gemini audit log or billing data to compute meeting time, monthly Gemini users and cost for your initiatives. For calendar events Roiva keeps each event's start, end and attendee counts, never its title, description or attendees. For Gemini it keeps a one-way digest of each person in place of their email, and nothing anyone asked Gemini or that Gemini wrote. This data is:

• Never used to train AI models, ours or anyone else's
• Sent to Anthropic only as the counts and totals derived from it, with the connection's name (for Google Workspace and Gemini, a Google email address or a Workspace domain; for Google Cloud, the name your team gave the connection), when you ask the in-app Assistant about them, and readable by an AI assistant your team connects over MCP (below)
• Otherwise kept within the sub-processors that host and run the service

Information we receive from Google, in the privacy policy, lists exactly what Roiva keeps from each Google service, and What Roiva keeps from each platform covers every integration.

Your own AI assistants (MCP)

An AI assistant your team already uses can read Roiva over the Model Context Protocol (MCP). Claude and ChatGPT connect by having a person sign in to Roiva and approve access to one account: an administrator, or any member if an Owner allows it. Microsoft Copilot Studio, Microsoft 365 Copilot, Amazon Quick and Gemini Enterprise sign people in the same way, through a client an administrator registers for the account. Claude Code signs in the same way as Claude, or can use a token. Cursor, Gemini CLI, Google Antigravity and other MCP clients use an access token an administrator issues. It is on every plan, the free account included. The assistant can read your account's initiatives, costs, cash value, freed capacity, and ROI. It cannot change anything, and every lookup it makes is recorded in your audit log.

Where your account allows it (Owners and Admins by default), the assistant can also draft a monthly figure Roiva asks a person to type in, such as a ticket count from an export, with the source it came from. The draft waits in Roiva until a person accepts or declines it. A figure accepted from a draft is marked Drafted by AI, and the value it calculates then waits for an approver like any other entry.

It can also draft the charges on a vendor bill Roiva doesn't connect to, such as an invoice you share with it, as an initiative's costs. Roiva checks each line as it would a bill you upload and flags any that may already be counted, which start unticked. Nothing is booked until a person reviews the lines and books them; those costs are marked Drafted by AI and count toward ROI at once, as costs a person types do.

And it can draft an initiative's business case: its planned figures, each with a basis line saying where it came from and the source it cites. Nothing changes until a person uses the figures they want on the business case form and saves them, and a basis line it wrote is marked Drafted by AI where Roiva shows the business case's bases, until someone rewrites it. It can't draft one for an initiative that is in review.

The answers Roiva returns go to that assistant and its AI provider, which your organization chose and has its own agreement with. That provider is not one of Roiva's sub-processors, and your agreement with it decides whether it may retain or train on what it receives.

Questions

If you have questions about how Roiva uses AI services, contact us at hello@roiva.ai .