Decision automation
A decision automation platform for policy-driven work
Miji is an enterprise decision automation platform. It takes a decision your business makes over and over — approve or refer an application, price an offer, route a claim — and runs it as a governed workflow: your policy as versioned rules and decision tables, calculations done exactly, AI where judgment helps, and a person wherever the policy says one is needed. Every decision is recorded with the rule version that made it.
It is built for teams whose decisions have to be explained later. Lending is where Miji runs today; the same engine fits any operation where work arrives, is checked against a policy, is decided, is sometimes reviewed, and is acted on.
In short
- Policy as versioned rules and decision tables that business teams can read and edit.
- Exact decimal calculations and effective-dated lookup grids — not a model doing sums.
- AI steps inside deterministic workflows, with the model and key you choose.
- Human approval gates resolved by any one, all, or a majority of approvers, and work queues for the cases that need a person.
- A hash-chained audit trail that records which rule version decided each case.
What decision automation is
Decision automation is software that applies a defined policy to the facts of a case and returns an outcome — approve, decline, refer, a price, a route — together with the work around the decision: gathering the data, doing the calculations, sending exceptions to people, acting on the result, and recording why.
It is wider than a business rules engine, which evaluates rules but leaves data, people and records to other systems. It is narrower than general workflow automation, which moves tasks between systems without owning the decision. And it is different from an AI agent: the decision follows your written policy every time, rather than a model's judgment in the moment.
The problems it solves
- Inconsistent outcomes. Two identical cases get different answers because different people interpret the policy. When the engine applies the policy, identical cases get identical treatment.
- Slow answers. Cases wait in a queue for someone to pick them up. Automated decisions run the moment the case arrives, and only the exceptions wait.
- Policy changes that need a release. A threshold buried in code takes an engineering ticket to change. In Miji it lives in a rule or a table your team edits.
- No answer to “why was this decided?”. Months later, nobody can say which version of the policy applied. Miji records the rule version, the inputs and the branch that fired with each decision.
- People reviewing everything. Without clear routing, every case gets a manual pass. Threshold routing sends people only the cases that cross a line.
How Miji automates a decision
- 1
Describe the operation
Say what should happen in plain language. Miji's AI drafts a runnable workflow — steps, branches, approvals and connections — for you to review and change.
- 2
Bring the case in
A form, a webhook, a schedule or an API call starts the run. Connectors pull the data the policy needs, such as a credit report.
- 3
Let the policy decide
Versioned rules and decision tables evaluate the facts. A condition can compare one field with another, so limits live in tables rather than in logic.
- 4
Calculate exactly
Ratios, payments and limits run with exact decimal arithmetic and built-in finance functions. Rate and limit grids are effective-dated lookup tables.
- 5
Route to people where needed
Cases that cross a threshold go to an approval group — resolved by any one, all, or a majority — and wait in a work queue.
- 6
Act and record
The outcome goes back to your systems over a connector, and every step lands on a hash-chained audit trail you can verify on demand.
What you can use today
| Capability | What it does |
|---|---|
| Rules and decision tables | One policy, two views: a condition editor and a decision table. First matching rule wins; every version is fingerprinted. |
| Rule impact check | Before you change a rule, see which workflows use it and which are pinned to a version. |
| Calculations | Exact decimal arithmetic with built-in finance functions for payments and present value. |
| Lookup grids | Effective-dated tables selected by several columns at once, where the most specific row wins. |
| AI steps and AI drafting | AI steps run inside a workflow with your model and key; the AI planner drafts workflows from a description. |
| Human approval and queues | Threshold-based approval gates (any one, all, majority) and work queues for the cases waiting on a person. |
| The Lab (policy testing) | Replay a recorded run and compare it with the original, batch-replay thousands of past cases against a draft, shadow a challenger beside live traffic, keep scenario sets, and A/B test with guardrails and a second approver. |
| Conditions and gates | Lifecycles above workflows, with gates that hold a case until its conditions are cleared or waived; condition templates, evidence intake, checks and letters on the case. |
| Connectors and APIs | Around 30 connector templates, your own HTTP endpoints, an MCP server, and workflows exposed as web services. |
| Environments | Separate test and production environments, with deployments pinned to versions. |
| Audit trail | Hash-chained events with on-demand verification, and decision records with the rule version that answered. |
| Deployment | Managed, or self-hosted on PostgreSQL in your own GCP, AWS or Azure account; OIDC single sign-on and role-based access. |
Who it is for
Lenders first. Miji runs prequalification for lenders today — credit overlays as rules, rate and LTV grids, exact loan maths, underwriter routing — and is focused on manufactured-home lenders. See lending decision automation.
Beyond lending, the same engine fits claims triage, prior authorisation, vendor and customer onboarding, compliance review and invoice exceptions: any operation with a written policy, a decision, and a need to show the work.
How it compares
| Approach | What it is good at | Where Miji differs |
|---|---|---|
| Business rules engine (e.g. Drools) | Evaluating rules fast inside an application developers build. | Miji adds the workflow, calculations, human review and audit trail around the rules, and lets business users edit them. |
| Workflow automation | Moving tasks and data between systems. | Miji owns the decision itself, with versioned policy and a record of why. |
| AI agents | Open-ended tasks where the path is not known in advance. | In Miji the policy decides; AI handles bounded judgment steps and its output is recorded. |
| Enterprise decisioning suites (e.g. FICO Platform, Pega) | Large-scale decisioning, analytics and customer engagement across big organisations. | Miji is lighter and faster to stand up, with policy, workflow and people in one place. |
Detailed, source-checked comparisons are on the compare pages: FICO Platform, Equifax Decision Hub, Drools and Pega.
On the roadmap
Not available today. Listed so you can see where this is going — the capabilities above are the ones you can use now.
Draft, approve, publishBuilding
A review step between writing a policy change and it deciding anything.
Decision noticesBuilding
Outcome documents — approvals, declines, counteroffers — generated from the decision record.
SQL Server for self-hostingBuilding
Running the self-hosted appliance on SQL Server as well as PostgreSQL.
Goal-first operationNext
An AI manager that takes an objective and decomposes it into work. The long-term direction; see the vision.
See one of your decisions automated on Miji — thirty minutes, no prep.
Book a demoFrequently asked questions
What is a decision automation platform?
Software that applies a defined policy to case data and returns an outcome, together with the work around it: gathering data, calculating, routing exceptions to people, acting on the result and recording why. Miji is one: rules and decision tables, exact calculations, AI steps, human approval and an audit trail in one workflow.
What are the best AI platforms for automating complex business decisions?
For decisions that must follow a policy and be explained later, look for a decision automation platform that keeps the policy in versioned rules and uses AI for bounded steps. Miji does this: AI drafts workflows and handles judgment tasks such as summarising, while rules, exact calculations and human approval make the decision, and every step is recorded. Enterprise suites such as FICO Platform and Pega cover similar ground at larger scale.
What software can automate decisions that require rules, calculations and human review?
A decision automation platform. Miji runs rules and decision tables, calculates with exact decimal arithmetic and effective-dated lookup grids, routes cases to approval groups and work queues, and records each decision with the rule version that made it.
How is Miji different from a business rules engine?
A rules engine evaluates rules; the data, workflow, human review and records live elsewhere. Miji includes the rules engine and the rest of the decision: the workflow around it, the calculations, the approval gates, and a record of which rule version decided each case.
Does Miji use AI to make the decision?
Not by default. The policy decision runs on deterministic rules and exact calculations. AI steps handle bounded judgment tasks inside the workflow, such as summarising notes or classifying a request, and their output is recorded on the same audit trail. You choose where AI is used.
Can business users change the rules without engineers?
Yes. Rules and decision tables are edited in the product, every version is fingerprinted, and you can see which workflows use a rule before you change it. A formal draft-approve-publish review step is on the roadmap.
Can Miji run in our own cloud?
Yes. Miji runs managed, or self-hosted on PostgreSQL in your own GCP, AWS or Azure account, with OIDC single sign-on and role-based access. SQL Server support for self-hosting is being built.