FlowSharp - analytics, ROI and governed AI

FlowSharp measures the process while the process works. This creates better decisions about automation, AI, cost and continuous improvement.

1. Measuring the process while it works

Every item moving through a Flow produces operational knowledge:

• time in phases;
• completed or failed tasks;
• manual steps;
• automations executed;
• bottlenecks;
• estimated cost;
• potential savings;
• AI output;
• generated events.

This measurement lets teams discuss the process using data, not impressions.

2. Time, bottlenecks and throughput

Analytics views help understand:

• how many items enter and exit;
• where items remain idle;
• which phases take longer;
• which tasks fail more often;
• which processes are improving;
• which processes are getting worse;
• which phases are saturated (WIP limit reached), with admitted items, queued items and saturation percentage.

The point is not only performance control. It is understanding where to act.

3. Cost, automation and ROI

FlowSharp moves the conversation from "can we automate?" to "is this the right place to automate?".

The ROI simulator helps reason about:

• human cost of a task;
• task frequency;
• time that can be saved;
• automation or AI cost;
• estimated savings;
• return over time.

4. Automation goals

A team can set a goal, for example increasing task automation from 25% to 45% by a given date.

The same goal appears in Analytics and in the Automation Plan, with a bar showing current level, target and the gap to close. The two areas are linked so teams can move from reading the metric to the initiatives intended to change it.

This turns automation into a governed program, not a set of isolated initiatives.

5. AI inside the process

FlowSharp integrates AI in several places:

• assisted Flow generation;
• Copilot in the designer;
• AI tasks inside Flows;
• configurable AI assistants;
• integrated Flow Agent;
• external agents via MCP;
• multiple AI providers;
• Data Shield and data minimization.

6. Flow Agent and assistants

The Flow Agent helps users:

• understand which Flows exist;
• choose which Flow to start;
• fill initial data when needed;
• read information about items and processes;
• work within the user's permissions.

The Flow Agent inherits the Data Shield: if the agent's AI provider has the shield enabled, sensitive fields of the items consulted in chat are replaced with pseudonyms before every call to the model — across the whole conversation, including tool results. The user keeps seeing the real data in the answers; a shield icon in the chat header signals that protection is active. If the model uses a shielded value to search items or propose a start, the real value is restored when the operation is executed. The same protection applies to the flow_agent tool exposed to external assistants via MCP.

AI assistants let teams standardize prompts, behavior, output and conversation style for specific use cases.

7. Data Shield, Data Gate and human control

When AI enters business processes, the point is not only calling a model. You must decide which data it can see, when it can work and who validates the result.

FlowSharp supports:

• sensitive fields;
• reversible pseudonymization (the model sees tokens, the operation uses real values);
• data minimization to the data contract;
• protection against missing data;
• action audit;
• manual tasks before or after AI;
• separation between operational providers and anonymization providers.

The Data Shield is enabled at AI provider level (Management → AI) and covers both AI tasks and statuses inside flows and the Flow Agent chat.

The Data Gate avoids AI on empty or incomplete data. Human work provides control, validation and responsibility.

8. Security, audit and governance

FlowSharp treats governance and security as part of the product.

Important elements include:

• authentication and roles;
• permissions for Flows and folders;
• audit log;
• session revocation when roles change;
• AI data protection;
• controlled uploads;
• API keys shown only once;
• key expiration and scope;
• encrypted secret variables, with reveal restricted to Admins and recorded in the audit log;
• idempotent item and event creation (no duplicates on retries or double clicks);
• tenant separation;
• read-only mode for users who only need to consult;
• tracked item history and transitions.

This makes FlowSharp suitable for processes where you need to know not only what happened, but also who did it, when and why.

9. Quality, outcomes and deviations

Analytics measure more than speed and cost. They also measure the quality of the journey:

• the win rate shows how many positively closed cases account for all positive and negative closures;
• the average score gathers the signal left by the phases a case has crossed;
• deviated cases reveal how much work is leaving the expected path;
• on the Process Map, KO percentage, score, waiting and deviations show where the process is losing value.

These indicators are not for judging people. They help identify places where the process too often forces rework, returns or exception handling.

10. Documenting and sharing the process

From a Flow or a Process Map you can create a print view ready to become a document PDF. The document brings together the diagram, phases, tasks and metrics for the selected period; for a map, it can include its linked Flows as well.

It is useful for a review, a handover, a customer meeting or a governance snapshot. The print view keeps the language of the process: it is not a technical export, but a document that stakeholders and operational teams can read.

11. Flow KPIs, history and forecasts

Each Flow can define item metrics and aggregate KPIs over its declared data and process metrics. Item values update as the case changes; flow KPIs can become dashboard cards with a number, gauge, target or sparkline.

History is snapshotted every night. After definitions change, Recalculate history queues a rebuild of values without blocking use of the Flow. The trend chart compares multiple KPIs with:

• windows from 30 to 365 days;
• dual axes for different units;
• base-100 mode for comparing heterogeneous trends;
• an internally calculated forecast with a confidence band.

When there is not enough history, FlowSharp declares the forecast insufficient instead of inventing a result.

A KPI can have thresholds and a value- or label-based goal, with a date and a baseline frozen at creation time. Nightly evaluation distinguishes achieved, missed, on track and at risk. Marked thresholds and goals produce events only when the condition changes, avoiding repetition; events can feed notifications, automations or a new item.

Dashboards and Analytics can be filtered by folder, including subfolders. Technical or test Flows can be excluded from statistics so they do not distort operational indicators.

12. Automation Plan

The Automation Plan turns an opportunity into a governed initiative. A task can originate from:

• candidates in Process Map Diagnostics;
• a What-if scenario, as one initiative or one per node;
• the map's AI Analysis;
• a Flow's ROI simulator;
• manual entry, for an idea born outside the metrics.

Each initiative records title, origin, operating owner, planned date, external reference, baseline cost, investment and expected savings. Measured and estimated contributions always remain separate.

On first activation, FlowSharp creates the Automation Backlog system Flow. Initiatives become its items and move through a governed path from proposal to planning, work, go-live and verification. Key statuses and the system family are protected, while the team can add intermediate steps.

The global page shows cross-map initiatives and rollups for investment, expected savings and realized savings. When an automation is live, verification compares the frozen cost with the current measured cost; if no measurable source exists, the result can be declared manually and remains marked as such. Candidates already in progress are recognized to prevent duplicates.

13. Baselines and Saving Challenges

A cost baseline snapshots a process before improvement. From Process Map settings, an Admin selects the observation period and reviews:

• measured cost per case for Flows;
• volume and operating cost;
• estimated manual work;
• data quality and completeness.

A value can be corrected only with a tracked reason. Freezing makes the baseline immutable, assigns a SHA-256 fingerprint to its cost components and creates a new version when another snapshot is needed.

A premium Saving Challenge links the baseline to duration, success-fee percentage, recurring costs and a minimum fee. In draft it shows a preview of periods; activation locks the terms and later changes become effective-dated amendments. Each period compares baseline cost, real volume, measured cost, AI cost and recurring costs to produce net savings and accrued fee. A consolidated period no longer changes.

The list and detail keep measured savings and estimated capacity separate, and show cumulative values and perimeter health. The simulator tests terms before the contract is created. Excel export becomes available after the first consolidated period and includes contract, monthly detail and per-Flow detail, with protection against formulas in free-text cells.