Glossary

The decisioning and automation glossary.

Plain definitions of the terms we use across decisioning, process automation, integration, and applied AI. Written for the people deciding whether to build, replace, or scale these systems, not for a spec sheet.

Rules engine: Software that evaluates business rules (if/then logic, decision tables, rate cards) against inputs and returns a decision. The point of a rules engine is to hold that logic outside application code, so the business can read, test, and change it without a software release.
BRMS (Business Rules Management System): A platform for authoring, versioning, testing, and governing business rules at scale, with an audit trail of who changed what and when. A rules engine executes rules; a BRMS manages their whole lifecycle.
Decisioning: The discipline of making a defined operational decision (eligibility, price, approval, routing) consistently, at speed, and with an explanation. The decision is the product; the platform is just how it ships.
Decision table / decision graph: Two common ways to express decision logic. A decision table maps combinations of inputs to outcomes in rows and columns; a decision graph chains decisions and sub-decisions into a traceable flow. Both make logic legible to non-engineers.
Decisioning platform: A system that combines a rules engine, data access, workflow, and governance so that decisions can be built, deployed, and audited as a managed capability rather than scattered through code and spreadsheets.
Rating engine: A decisioning system specialised for calculating a price or premium for a defined unit (an insurance policy, a loan, a consignment) in real time. See our note on real-time rating at scale.
Rating vs pricing vs decisioning: Rating calculates a number for a unit. Pricing is the business decision about what the number should be, given margin and strategy. Decisioning is the broader category (eligibility, approval, qualification) of which rating is one specialised case. Projects that fail at rating usually failed to separate the three.
Validation vs verification vs approval: Validation checks a submission against a defined ruleset and explains the result. Verification proves an identity or fact. Approval grants permission. Large-scale systems get into trouble when these are conflated. See validation at scheme scale.
Process automation: Coordinating the steps, hand-offs, and systems that move work from request to outcome. Distinct from decision automation: process is how work flows; decisions are the choices made along the way. Most real engagements need both.
Straight-through processing (STP): When a transaction completes end to end with no human touch because every decision and step could be made automatically. The realistic goal is rarely 100% STP; it is a high STP rate with a well-designed path for the exceptions.
Exception handling: What happens when a case does not fit the happy path: how it is explained, who reviews it, against what service level, with what audit trail. In high-volume validation and decisioning, the exception path is the product, not an afterthought.
Real-time vs batch: Real-time returns an answer synchronously while a person or system waits (latency budgets in the tens to low hundreds of milliseconds). Batch processes accumulated work on a schedule. Real-time is a cost, not a virtue: use it where something is genuinely waiting.
Latency and P99: How long a decision takes to return. P99 is the latency the slowest 1% of requests still beat, a more honest target than an average because it reflects the experience under load.
Audit trail: An immutable record of each decision: the inputs, the rule version in force at the time, the outcome, and the reasoning. In regulated work the system must answer not just "what was decided?" but "what would have been decided six months ago?"
iPaaS (integration platform as a service): A platform for connecting applications and moving data between them with managed connectors and workflows, rather than bespoke point-to-point integration code. Workato is an example we deploy.
Workflow orchestration: Coordinating tasks, services, and integrations into a reliable end-to-end flow, with retries, error handling, and visibility. n8n is an open-source orchestration tool we deploy and support.
RPA (robotic process automation): Automating repetitive, structured, screen- or system-level tasks by mimicking the steps a person would take. Strong for stable, rules-bound work; brittle when the underlying systems or formats change. Often the wrong tool where an API or a decisioning approach would be cleaner.
Agentic AI: AI that can plan and carry out multi-step tasks toward a goal, gathering context and using tools along the way, rather than answering a single prompt. Useful for complex, mixed-channel, context-dependent work. See agentic AI for mid-market.
Agentic vs autonomous: Agentic systems take initiative across steps but operate within defined bounds and oversight. Autonomous implies acting without human control. In regulated operations the useful pattern is agentic-within-rules: the AI does the heavy lifting; business rules govern the outcome.
Deterministic vs probabilistic: A deterministic system gives the same output for the same input every time (business rules). A probabilistic system gives a likely answer that can vary (most machine learning and LLMs). Regulated decisions usually need determinism; context-gathering and synthesis suit probabilistic AI. Good designs combine them.
Grounding: Anchoring an AI model's answer to specific, citable source material rather than its general training, so the response is accurate and traceable. Clean structured content and schema make a site easier for answer engines to ground against.
Low-code / no-code: Tools that let people build or change logic, workflows, and interfaces through configuration rather than hand-written code. The value in decisioning is that the people who own the rules can change them safely, without queuing behind an engineering release.
Decision Pyramid: Our framework for separating the layers of a decision (the data it needs, the logic that governs it, and the experience that surrounds it) so teams fix the right layer instead of rebuilding the whole stack. See the Decision Pyramid.

Got a workload in mind?

Tell us where work gets stuck.

If a definition here matches a problem you are carrying, we can tell you what fixing it actually looks like.

Talk to us