Uniphore Help Center Portal

Process Intelligence offers two methods for data import:

The best solution for your environment might involve one or both methods. For example, if you have a repository of historical transcripts in JSON format or audio/video files, you can manually upload those artifacts. If you install Chronos to record customer sessions, that data is automatically ingested into Process Intelligence for analysis via a connector you configure.

Depending on your level of access, some controls, pages, or information may not be available to you. Contact your administrator if you have questions about role-based access.

To begin, go to the home page > Agents > Data Layer > Process Intelligence.

The Cognitive Analysis page for your project opens. This is the default landing page for your project, and the best place to get started. If you have access to multiple projects, you can select a project via the Project dropdown available on each Process Intelligence page.

Depending on your level of access, some controls, pages, or information may not be available to you. Contact your administrator if you have questions about role-based access.

The process list displays all discovered processes in your project.

Each row in the process list represents a single discovered process and includes the following columns:

Via Process Explorer, you can drill down into a specific process to understand exactly how it is executed and which variants or alternate paths exist. While your organization may have one documented "happy path" for a process, analysis of actual recordings typically reveals several variants: different paths agents actually take to complete the same process. These variants may reflect different outcomes, escalation paths, workarounds, manual interventions, and more.

To open a process, either click its name in the process list or click View Process in the Actions column.

From here, you can view and compare paths, review metrics, see root cause analysis and recommended actions for each path, and export a report about the process.

Process View Overview

When you open a process, you are viewing it at an aggregate level, so the view reflects all recorded occurrences of that process across your dataset, not a single instance. Once opened, the process view provides:

• Paths used in the selected process.

• The Map view, which shows the process flow for the selected path or paths. The Map view includes additional insights about root causes and potential next steps.

• The Analytics view, which provides deeper information on the identified process bottlenecks, root causes, and possible recommendations to improve the process.

• A link to create an SOP for the process via the SOP Studio.

You can use this view to see which processes are running in your organization and where bottlenecks happen.

Map View

The Map view displays the process flow for the selected path, and shows each step in sequence, including branches, decision points, and outcomes.

Toggle the level of detail shown in the flow to focus on the information most relevant to your analysis.

You can drag (move) or resize paths:

1. Click + to zoom in.

2. Click - to zoom out.

3. Click the Fit icon to fit the path in the view.

To view all of the paths side-by-side, change from Single View to View Paths.

This provides an at-a-glance view of all of the paths. Use Compare Paths to view the path in the Variant Analysis page and get a more detailed comparison between two or more paths.

Insights

The side panel provides general insights into the selected path, including Root Cause Analysis and suggested actions.

Click any step in the process flow to drill down and see more information. You can explore the source material (video, transcript, or audio) that Process Intelligence used to build the map. Use these sources to validate accuracy and understand the context behind specific steps or decisions. The detailed view also highlights the identified bottlenecks, the applications involved, and the activities performed for that step in the path.

To learn more about a specific occurrence, click View Occurrence Map.

The Occurrence Map for the specified occurrence opens.

The Occurrence Map shows all data associated with that interaction: the metadata Process Intelligence extracts for the occurrence, transcripts, videos, and system event logs.

This view shows the system's interpretation of a single run alongside its raw evidence. It is the screen for validation, debugging, and root-cause investigation of one specific occurrence.

The Occurrence Map has three main sections: Evidence, Occurrence Flow, and a side panel with AI Analysis. All three are time-synchronized. Selecting a step on the map jumps the video to the right moment, scrolls the transcript to the right turn, and highlights the matching log event.

Variant Analysis answers the same underlying *efficient vs. inefficient* question through different filter combinations. Common patterns:

To get a more detailed comparison between two or more paths, compare multiple paths via the Variant Analysis page or by clicking Compare Paths on the Process Explorer page. Use this to identify or examine the differences between two paths.

To compare paths:

An Automation Report is generated for each new process as it is created.

The report card displays the process identifier, the number of instances and steps analyzed, and a step automation breakdown bar showing how many steps are fully automatable, partially automatable, or require manual handling, along with an overall readiness percentage.

To open an automation report, click Analyze Process. The report opens in the Automation Analysis page, displaying the process name, instance count, step count, and a summary of automatable steps, with the Automation Score in the upper right.

Automation Analysis

The Automation Analysis page provides a detailed assessment of the selected process's automation potential. It shows how many steps are automatable and to what degree, surfaces governance and compliance considerations, and gives an overall readiness score to help you prioritize automation efforts.

Automation Score

The Automation Score is a 0–100 score that reflects the overall automation readiness of the process. The score is accompanied by a readiness label: Low, Medium, or High.

Step Breakdown

The step breakdown displays six metrics that together describe the automation profile of the process:

• Total Steps - The total number of steps in the analyzed process.

• Highly Automatable - Steps that can be fully automated without human intervention.

• Partially Automatable - Steps that can be partially automated but require some human involvement.

• Manual Only - Steps that require full human intervention and are not automatable.

• Governance Review - Steps that require governance review due to factors such as PII, financial data, or high blast radius.

• Human Approval - Steps that require an approval gate before proceeding.

Key Takeaway

Key Takeaway provides an AI-generated summary of the overall automation opportunity for the process, including the systems involved, their suitability for automation, and any recommended approaches or considerations. Click See more to expand the full summary.

Overall Automation Readiness

The Overall Automation Readiness bar shows the percentage of steps that are fully or partially automatable out of the total steps in the process. Two additional metrics are displayed below the bar:

• Automatable Time Share - The percentage of total process time that could be automated.

• Reliable (API-only) Share - The percentage of process time that can be automated using API-only methods, representing the most stable and reliable automation approach.

Governance Summary

The Governance Summary provides a compliance, blast radius, and data sensitivity profile across all steps in the process. Any applicable compliance frameworks are displayed as tags in the upper right of the section.

The following governance metrics are displayed, each showing the number of steps affected:

• PII - Steps that involve Personally Identifiable Information.

• PHI - Steps that involve Protected Health Information.

• Financial - Steps that involve financial data.

• Approval Required - Steps that require human approval before proceeding.

• Audit Log - Steps that must be logged for audit purposes.

• Irreversible - Steps whose actions cannot be undone.

Blast Radius shows the worst-case impact scope if a step fails, broken down by category and percentage of steps affected.

Data Sensitivity shows the highest-sensitivity data class each step touches, broken down by category and percentage of steps affected.

By Application and By Step

The By Application and By Step tabs provide further detail:

• By Application - Per-tool feasibility, confidence, and usage share across the process, sorted by activity volume.

• By Step - A per-step breakdown showing each step's type, name, applications involved, and feasibility rating. Expand any row to see application methods, decision logic, and human-thinking notes.

Automation Artifacts

The Automation Artifacts page provides the structured outputs generated from the automation analysis. To access it, click Automation Artifacts under the BUILD tab.

Workflow Steps

The Workflow Steps section lists all steps in the process, each labeled as a START, STEP, or END. Each step displays a workflow ID, a description, and any input and output entities associated with that step. Click a step to expand it and see full detail.

Tools

The Tools section lists the automation tools generated for the process. Each tool displays a name, type tags such as Agent or Mutating, a description, behavioral instructions, and a parameter summary showing the number of inputs and outputs. Click a tool to expand it and see full detail.

Skills

The Skills section lists the automation skills generated for the process. Each skill displays a name, a skill definition ID, and a description. Click a skill to expand it and see full detail.

Downloading Artifacts

To download the full artifact package, click Download Artifacts.

Report Lifecycle

Automation reports are generated once and remain static. A report will remain accurate for the life of the process because new data files create new processes rather than updating existing ones.

The Process Optimization tab displays projected savings across your discovered processes based on estimated handle time improvements. The following summary metrics are displayed at the top of the page and you can use them to get a quick sense of the overall optimization opportunity across your processes before drilling into the detail below.

Below the summary metrics: a Configuration panel, a Process Breakdown table, and an AHT Comparison Chart. Click How are savings calculated? at the bottom of the page to see the formula used to calculate monthly savings for each process.

The Automation tab models the return on investment of automating your discovered processes based on configurable cost and implementation parameters. The following summary metrics are displayed at the top of the page and you can use them to quickly assess the overall automation opportunity and investment outlook across your processes before reviewing the rest of the sections.

Below the summary metrics: a Configuration panel with scenario saving and comparison tools, a 3-Year Cumulative Savings Projection chart, a Sensitivity Analysis, an Automation Readiness Matrix, a Savings Contribution by Process chart, and a Process Comparison table.

Working with Scenarios

The Scenarios feature lets you save up to three named configurations of the sliders for comparison. To save a scenario, enter a name in the Scenario name field and click Save. Saved scenarios are listed below the Configuration panel and can be deleted individually using the delete icon.

Once you have saved two or more scenarios, click Compare Scenarios to open the Scenario Comparison modal. The modal displays each scenario's configuration inputs and resulting metrics side by side, with the best-performing value highlighted for each metric.

Use scenarios to model different cost and timeline assumptions side by side. For example, a conservative baseline, an optimistic fast-ramp option, and a high-cost cautious approach, so you can present stakeholders with a range of outcomes rather than a single projection.

To create a new SOP:

The new process is added to the SOP Library. It is complete as is, or you can edit it and make further changes.

To edit an SOP, click the process name. The SOP Editor opens, and shows a list of filters and the process flow map:

You have two options for editing an SOP: 

To make changes directly via the SOP Editor:

To rename an SOP:

To delete an SOP:

You can export an SOP in PDF or Word format.

The Include Images and Include Flow Chart options apply only to Text View. Select Text View before configuring these options. Include Flow Chart means to include the Process Map as a graphic/flow chart in the SOP.

Chronos (https://uniphore.github.io/chronos/index.html) is a recording tool used to capture, redact, and upload interactions from which business processes are identified, called screen recordings. Chronos is available as a Chrome extension and a desktop app with a lightweight installer and a fully offline installer with bundled dependencies.

You need the following information (credentials, paths, etc.) when creating connectors.

To create a connector, click Add Connector. The Add Connector panel opens.

Click the type of connector you want to create.

To edit a connector:

Redaction settings control how Process Intelligence treats Personally Identifiable Information (PII) in the data artifacts as they are uploaded. You can specify whether to enable redaction and how information will be redacted: Replaced with a Black Box, Blurred, or Pixelated.

To test a connector:

To delete a connector:

Click a file in the listing to open it in the viewer. Depending on the file type, you can view, read, or listen to it.

You can manually upload additional data artifacts, such as audio, video (standalone and multi-part explainer videos), and transcript (JSON) files, by clicking Upload Files.

Before uploading data files to Process Intelligence, you should edit them to remove any PII. To have Process Intelligence redact PII on manually uploaded data artifacts, contact Uniphore.

bundle_id,file_name,sequence,agent_id
P001,video_part1.mp4,1,agent_1
P001,video_part2.mp4,2,agent_1
P002,video_v2.mp4,1,agent_2

Whenever new source files (video sessions captured by Chronos and manually uploaded data files) are available, you need to trigger discovery to identify new processes.

To trigger discovery on any newly available data files and Chronos sessions:

Depending on your level of access, you may be able to delete existing data files.

To delete a data file:

Here is a list of permissions by role. Administrator users have all of the User permissions as well.

Limited to Admin users.

To create a user:

Limited to Admin users.

When editing a user, you can change the user's password, name, or role. Click a user to edit it. Then click Save to save the changes.

To reduce the number of re-takes:

Before you begin clicking, take 15–30 seconds to say what this process is. Something like "This is the monthly journal-entry close — I'm in Oracle EBS, and the goal is to post and approve the close for each ledger".

You don't need to recite a checklist. The algorithm picks up context from whatever you say naturally: process name, the systems involved, the expected outcome, anything not obvious from the screen. Just frame the recording so the rest of it makes sense.

Talking through what you're doing, even loosely, meaningfully improves the output. You do not need to provide a play-by-play of every click; the algorithm can infer most mechanical actions from the screen on its own. What it cannot infer is information in your head: why a step matters, the business reason for a choice, or context that isn't visible on screen.

A few habits that go a long way (when they fit naturally):

It is okay to stay quiet through a stretch of routine UI work; the algorithm will still capture the steps. The narration adds context that the screen alone cannot provide.

It is important to note that business processes are graphs, not straight lines. The algorithm can only model the branches it has some signal about. The path you actually click is captured either way; alternate paths only show up if you describe them.

When you reach a decision point, the most useful thing you can do is briefly state what you are evaluating and the alternatives. For example: "I'm checking whether the variance is over $500. If it is, I file a manual adjustment; if not, I move on to the next ledger." Then continue and perform whichever path applies.

Pausing briefly on the screen you are evaluating helps too. It gives the algorithm a clean moment to anchor the decision step on. But you do not need a strict procedure; a single sentence covering the criteria and the alternatives is usually enough.

Ideally, you can demonstrate a non-trivial alternate path live (e.g., an exception, escalation, or retry). If not, a sentence describing what would have happened on that path allows the algorithm to represent it as a synthetic instance.

Business rules are the durable knowledge that defines how the process is supposed to work: thresholds, validations, compliance constraints, and SLAs. They enforce SOP correctness, surface governance requirements, and define what automation logic is allowed to do.

The algorithm can pick up rules that are visible on screen (error messages, threshold prompts, validation banners), but rules that live only in your head, such as "variances over $500 need manager approval", or "we never refund a pending transaction", only get captured if you mention them. So when a rule influences something you're doing, a quick callout helps.

Common categories worth flagging when relevant:

The list above is not exhaustive. Call out whatever rules genuinely apply.

Many processes repeat steps, such as closing ledgers one at a time, working through a queue of tickets, and retrying after a validation failure.

Some steps are checks rather than actions: verifying row counts match, confirming debits and credits net to zero, reconciling totals, and comparing IDs across systems. These look similar to ordinary review screens; however, they play a different role: they are the points where the process formally gates on data integrity.

A quick verbal callout, such as: "I'm checking that debits and credits net to zero before posting", helps the algorithm represent these as their own verification steps rather than folding them into the surrounding action. That distinction matters downstream for audit, training, and any automation that needs to preserve the gate.

The algorithm picks "representative timestamps" for each step and extracts the corresponding frames as screenshots that visually document each step. These are used in generated SOPs, in process visualizations, and as ground-truth references when reviewing automation candidates. Frames are picked from your video, so the video must contain good frames.

Do:

Avoid:

Preferred length: 15–55 minutes per process. Videos under 55 minutes are processed in a single pass, resulting in the most coherent output.

Format: Standard screen-recording formats are accepted (MP4 / MOV / WebM). MP4 with H.264 video and AAC audio is the safest choice.

Frame rate: 24–30 fps is sufficient. Higher rates only inflate file size.

If the full process is too long to record comfortably in one sitting, or you need to stop and resume (across breaks, days, or shifts), record it as multiple part-files. Our solution joins the parts in the correct order and processes them into a single video, provided you specify the correct order in a manifest file.

When to use multiple part-files;

How to record the parts:

bundle_id,file_name,sequence,agent_id
P001,video_part1.mp4,1,agent_1
P001,video_part2.mp4,2,agent_1
P002,video_v2.mp4,1,agent_2

Narrate the outcome of the run: "The journal has been posted and approved. The close is complete for this ledger."

Briefly summarize what came next or what would happen on a different outcome (escalation path, follow-up, etc.).

Stop recording after the final screen has finished rendering. Do not cut off mid-action.

Pre-Flight Checklist

Post-Recording Checklist

Process Intelligence does not rely on manually documented workflows. Instead, it derives processes automatically by analyzing recorded interactions: screen recordings, audio, transcripts, and system logs. The result is a data-driven model of how work actually happens, rather than how it is assumed or intended to happen. This distinction, observed reality versus documented assumption, is central to the value Process Intelligence provides.

Process Intelligence is built on a two-level data model. At the bottom level are occurrences: individual recorded instances of a task being performed. At the top level are processes: aggregated models built from many occurrences. Understanding this hierarchy is essential: when you view a process in Process Explorer, you are seeing a composite picture derived from all recorded occurrences, not a single instance. When you want to examine a specific interaction, go to Process Occurrences.

Process Intelligence identifies all the distinct ways a process is actually executed and represents each as a path. Paths may reflect different outcomes, escalation routes, workarounds, or agent-specific behaviors. The All Paths view combines all variants into a single map to show the full range of possible execution sequences. Recognizing and analyzing path variation is foundational to understanding process health and identifying improvement opportunities.

Rather than treating inefficiency as a qualitative observation, Process Intelligence quantifies it using the Friction Index: a 0–10 score calculated from signals such as system switching, repeated loops, and silence. This makes operational drag comparable across processes, agents, and time periods. A process with a Friction Index of 8 can be directly compared to one with a Friction Index of 3, enabling prioritization of improvement efforts based on data rather than intuition.

Process Intelligence is designed around a workflow that moves from raw evidence to optimization:

Each stage builds on the previous one. The workflow is designed to be repeatable. As new recordings are continuously uploaded, the process models stay up to date, and the cycle begins again.

Process Intelligence supports analysis at two different levels of granularity, and choosing the right level depends on the question being asked. Aggregate analysis, available through Process Explorer and Variant Analysis, answers questions about patterns across many interactions: What is the most common path? Where do bottlenecks typically occur? Instance analysis, available through Process Occurrences and the Occurrence Map, answers questions about a specific interaction: What exactly happened during this call? What anomalies were detected? Both levels are necessary for a complete understanding of process performance.

A key analytical capability in Process Intelligence is the ability to compare paths directly. Use Variant Analysis to place two or more paths side by side, apply filters (such as agent or AHT range), and view metrics for each. Combined with the AI-generated Root Cause Analysis insights, this enables you to move beyond identifying that a problem exists to understanding why it exists, and what to do about it.

In Process Intelligence, SOPs are not static documents that you write once and file away. They are generated directly from observed process data and can be regenerated or updated as processes evolve. Use the SOP Editor to configure scope (e.g., top 80% of paths vs. all edge cases), apply filters, and edit the resulting process map before exporting. This means SOPs reflect actual operational behavior rather than idealized descriptions, and can be kept current as that behavior changes.

Because Process Intelligence ingests recordings of real agent interactions, it necessarily handles sensitive data. PII redaction is built into the ingestion pipeline: personal information is removed from files as they are uploaded, before they are stored or analyzed. Redaction is configurable at the connector level, allowing organizations to control the method used (Black Box, Blurred, or Pixelated). This ensures that the analytical value of interaction data can be captured without retaining personally identifiable details.

Process Intelligence maintains a direct link between its analytical outputs and the underlying source material. In Process Explorer, you can click any step in a process flow to access the original video, transcript, or audio used by Process Discovery to build that step. This transparency allows you to validate the accuracy of process models, understand the context behind specific steps or decisions, and build confidence in the data driving optimization recommendations.

Connectors define how Chronos screen recordings and other explainer videos, transcripts, and audio files are ingested into Process Intelligence, specifying where files originate (Google Cloud Storage or SFTP), how they are accessed, and how PII within them is redacted on upload. Once uploaded, files are stored within Process Intelligence and are visible on the Data View page. Process Intelligence supports multiple customers through a multi-tenant architecture, meaning each organization's data is stored and accessible only within their own environment.