AERO, the fastest solver in DecisionRules

Speed that holds at scale.

AERO is the new solver in the DecisionRules business rules engine. It keeps your rules in memory after the first call and runs nested rules in the same process, so complex decisions return in milliseconds.

4ms

median for a real customer's insurance quote: 600 conditions across 13 rules, in one API call.

Proven in production

Teams running their decisions on DecisionRules

More than 1B decisions a day run through DecisionRules, for airlines, telecoms, banks, insurers and retailers.

1 · Speed

Complex decisions, in milliseconds.

A small test rule makes for a faster headline number. We measured a real customer's production decision instead: 13 rules and 4 subflows in one API call. Here is the case, step by step.

Customer case study

One of the world's largest insurance groups · name withheld at the customer's request

01 · The decision

A motor insurance quote, built in one real-time API call while a customer or agent configures the car and the cover on screen.

02 · The result

4 ms

median for the whole decision, in one call 1

Measured at 500 requests per second, same setup as the benchmark below.

03 · What it runs

Decision Tables and Lookup Tables
12
Scripting Rule
1
subflows, connected in one Decision Flow
4
rows
230
conditions in total
600

04 · Why it is fast

After the first solve, AERO keeps each rule in memory, so later solves skip fetching it from the database again. Nested rules run inside the same process, with no internal HTTP request between them, so your rule set can keep growing with your business.

2 · Infrastructure

Big decision capacity.
Smaller infrastructure.

Your cloud bill grows with every pod you run. One AERO process puts every available CPU core to work, so a single pod carries more traffic before you need the next one. When you do, add pods and scale out.

Throughput

100,000+

decisions per second on 3 × AWS c7g.xlarge

Proof of concept

See AERO's speed on your own rules.

Our solution consultants set up a proof of concept with your rules and data, so you measure the latency yourself.

Bracey Parr DecisionRules Expert

Bracey Parr

DecisionRules Expert

Book a demo

Benchmark

The full benchmark, test setup included.

One latency figure says little without the conditions behind it. We publish the full distribution and the test environment, and we rerun the same test on your rules in a proof of concept.

Percentiles

p50 and p95 for every workload. The slow requests stay visible.

Measured over HTTP

k6 calls the REST API the way your service will, at a fixed 500 requests per second.

Validated responses

A request counts as successful only when it returns 200 and its content passes validation.

Reproducible on your rules

Our solution consultants share the test setup and rerun it on your own rules in a proof of concept.

Load test results

500 requests per second, constant

Every latency in the table was measured under this load. That is more traffic than a typical customer deployment handles.

Load
500 requests per second, constant
App nodes
3 × AWS c7g.xlarge, 4 vCPU, 8 GiB
Load generator
k6 on 1 × m7i.2xlarge
Cache
In-memory, enabled
Redis
cache.t4g.micro
Network
Same AWS region
Errors
0%
Date
3 September 2026
Book a demo with your rule
Latency and throughput results measured at 500 requests per second
Workload Metric Result
Decision Trees median 2.59 ms
Decision Tables median 2.94 ms
Decision Flow, small median 3.35 ms
Decision Flow, medium and large median 4.98 ms
Scripting Rules, data transformation median 7.84 ms
All 13 test cases median 3.83 ms
All 13 test cases p95 5.29 ms
Slowest case outside Scripting Rules, sequential Decision Flow p95 9.87 ms
Max throughput, one pod requests per second 35,000+

1. Measured in the environment above with warm cache. Network latency between your service and DecisionRules is not included, and HTTP or database calls inside a Decision Flow take as long as those systems need. Benchmark results, not a contractual commitment.

Tech specs

AERO tech specs

Full reference in the DecisionRules documentation .

API
POST /rule/v2/solve/:ruleId/:version , JSON in and out
Bulk solve
Array of objects in one request. Parallel worker pool, results in input order.
Caching
Rules held in memory per space, loaded on first call
Observability
Solver version and correlation ID in response headers. OpenTelemetry (OTLP) export of traces, logs and metrics to Grafana, Jaeger, Datadog or any OTLP-compatible collector, with shared traceId and spanId across requests, audit records and logs.
Audit
Audit logs for root and nested rules
Deployment
Self-hosted Docker images. Kubernetes with Helm charts.
Security
SOC 2 Type II, ISO 27001 certification

FAQ

Questions about AERO

Tell me more about AERO