From financial planning to intelligent decisions.
One environment that takes a business from its planning model and data to analysis, decisions, actions and governed distribution - with every authoritative figure computed by a deterministic engine and traceable to its source.
Pre-production build · early access by request
Planning still breaks between the spreadsheet, the dashboard and the decision.
Cogniartha is built around one question: can a single environment take a business from model and data to decision, action and distribution, with every figure traceable?
Configuration before planning
Planning platforms often need months of specialist configuration before anyone can actually plan.
Dashboards you can only look at
Numbers are shown in one place, while data entry, action and follow-up happen somewhere else.
AI numbers nobody can trace
An assistant can write a figure fluently, yet cannot say which formula, version or record produced it.
Everything between the model and the decision.
Each capability below is labelled by status. Demonstrated means it is shown working in the recorded demo of the current build.
Enterprise planning and modelling Demonstrated
- Models created by describing the business, or by defining record types, attributes and measures directly - reviewed and approved before going live
- Measures with units and lever roles; drivers whose formulas are parsed, never executed as code
- Weighted, effective-dated organisation hierarchy
- Controlled data entry and in-place write-back, validated and audited, with stale edits refused
- Budgets as their own data kind; variance computed by the engine
Specified Dedicated budgeting, planning and revenue-projection buttons are specified but not in this build.
Planning in detail →AI and ML intelligence Demonstrated
- Planning goals in plain words, turned into a structured proposal that is validated and confirmed by the user before anything runs
- Machine-learning forecasts, always stored and shown as predictions
- Anomaly findings, and sourced market-intelligence signals that can feed a scenario
- Screens generated from a description, then edited freely like any hand-built screen
- The language layer never produces an authoritative figure
Specified A deterministic forecast is specified but not in this build.
How AI is kept separate →
Deterministic financial engine Demonstrated
- Exact decimal arithmetic - no floating point, no clock and no randomness
- Driver-based calculation, roll-ups across the weighted hierarchy, variances and margins
- Every result linked to its driver version and calculation run, with a replay check
- Any figure can be opened to show its provenance, formula, inputs and contributing units
- Goal-seeking scenarios search on engine figures; lever changes can be evaluated with no AI at all
Dashboards and screen authoring Demonstrated
- Composable screens from a toolbox of 46 typed objects - charts, grids, KPIs, gauges, maps, entry objects, filters, rich text and images
- Semantic binding to the measures, units and dimensions of the live model, limited to what the author may see
- Containers, tabs, filters scoped to an object, a container or the whole screen
- Preview, immutable versions, duplication and structure-only templates
- Editable grids, grouping, drill-down and drill-through
Operational actions Demonstrated
- Buttons with eight action classes: ML forecast, driver-engine recalculation, scenario, write-back, export, email, navigation and sequences
- Every click runs with the clicking person's permissions and is audited with its path and outcome
- Authors choose the AI path or the deterministic path for scenarios
- Long-running work shows live progress in words, never numbers
Distribution and governance Demonstrated
- Reusable, versioned email templates with dynamic values filled in per recipient
- Attachments in PDF, PNG, PowerPoint, Excel, CSV and Word
- Scheduled sends and event triggers on thresholds, variances, shortfalls and changes
- One separately composed document per recipient - never one report filtered afterwards
- Publishing rules, approvals, oversight and an audit log with before and after state
In the recorded build no mail relay is configured, so sent copies arrive in each recipient's in-platform notifications.
Distribution in detail →
One cycle, one environment.
- 1
Model & data
Approved model; records validated, written back and pre-aggregated.
- 2
Analysis
Screens composed for each viewer, with provenance behind every figure.
- 3
Planning
Forecasts labelled as predictions; scenarios computed by the engine.
- 4
Action
Buttons recalculate, forecast, write back and export as the clicker.
- 5
Distribution
Scheduled or triggered sends, one scoped copy per recipient.
Real screens from the running build.
These are frames from the recorded demonstration on illustrative demo data - not mock-ups.
Watch the complete product, end to end.
A narrated walkthrough of the running build, from tenant setup and modelling to screens, actions, security and governed distribution. Capabilities that are specified but not yet built are labelled as such in the film.
Chapter 1 of 29 · Opening
- 01Opening0:37
- 02The Problem0:55
- 03Platform Overview1:05
- 04Tenant, Users and Security1:14
- 05Model Building1:13
- 06Data and Financial Model1:09
- 07Data Entry and Write-back1:04
- 08Forecasting and Planning0:55
- 09AI and ML Intelligence1:01
- 10Deterministic Financial Engine1:07
- 11Traditional Dashboard Experience0:47
- 12Screen Authoring1:18
- 13Visual Object Toolbox1:07
- 14Semantic Data Binding0:54
- 15Interactive Planning Screens0:53
- 16AI-generated Screens1:21
- 17Action Buttons1:12
- 18AI versus Deterministic Execution0:42
- 19Live Execution and Progress0:59
- 20Governance0:58
- 21Security and Row-Level Security1:07
- 22Email and Distribution1:02
- 23Scheduled and Event-triggered Distribution1:05
- 24Per-recipient Security1:25
- 25Auditability and Traceability0:51
- 26End-to-End Business Scenario1:15
- 27Technical Architecture1:24
- 28Enterprise Value0:37
- 29Closing0:17
Language understands. ML predicts. Only the driver engine computes.
Intent, layout, narration
Turns requests into proposals that are validated against an allow-list and a schema. No code path writes a model response into a stored number.
Forecasts and anomalies
Every forecast, trend and anomaly finding is stored and shown as a prediction, never mixed into actuals.
Authoritative figures
Exact decimal arithmetic, each result tied to its driver version and calculation run.
Evaluated for the actual person
Access has two independent axes - organisation units and measures. Viewers, clickers and recipients each get figures composed for their own scope; out-of-scope data and missing data look the same.
Auditable by design
Writes, publishing, button executions, sends and trigger firings are recorded with actor, target and before-and-after state. In validation A PostgreSQL row-level-security backstop is built and being validated.
Cogniartha does not claim any security certification or compliance attestation at this stage.
Help shape the platform.
For finance and FP&A leaders, enterprise planning teams, consultants and design partners who want to evaluate an engine-first approach to planning and decision intelligence.
Requests are reviewed individually. Access depends on fit and availability, and the platform is still pre-production.
AI & ML Consultants
Cogniartha is a product of AI & ML Consultants, presented by Manjunatha M.
About Cogniartha →