Same engine
Deterministic, probabilistic, and spatial matching behave identically in both paths.
Performance & scale
Real-time requests and file uploads run the same matching engine with the same rules. Pick the one that fits your workflow, not your accuracy needs.
Same answers
Both paths maintain complete matching parity, returning identical person IDs and decisions.
Different shapes
Real time handles 1 to 100 records per call. Files handle thousands to tens of millions of records.
Real-time matching
Section titled “Real-time matching”Each real-time call has a mostly fixed overhead: securely authenticating the request, initializing the match, and assembling the returned attributes. Because that baseline overhead is similar whether processing 1 record or 100, sending more records per call makes each record significantly faster.
| Records per call | Average time per call | Effective time per record |
|---|---|---|
| 1 | ~6.2 s | ~6.2 s |
| 10 | ~6.8 s | ~0.68 s |
| 25 | ~7.9 s | ~0.32 s |
| 50 | ~8.9 s | ~0.18 s |
| 100 | ~10.4 s | ~0.10 s |
Individual-level matching with base attributes. Individual calls vary depending on network conditions.
What you ask for affects speed
Section titled “What you ask for affects speed”Call duration also depends on which match levels you request and how many enrichment attributes you include. Lighter requests return faster.
| Request configuration | Records | Average latency | Effective time per record |
|---|---|---|---|
| Digital level only, base attributes | 10 | ~6.4 s | ~0.64 s |
| Individual level, base attributes | 10 | ~6.8 s | ~0.68 s |
| All levels (no spatial: I, H, D, A) | 10 | ~11.6 s | ~1.16 s |
| All 5 levels including Spatial (I, H, D, S, A) | 10 | ~21.5 s | ~2.15 s |
| Individual level + full enrichment catalog (500+ attributes) | 10 | ~22.3 s | ~2.23 s |
| All 5 levels + full enrichment catalog (500+ attributes) | 100 | ~39.8 s | ~0.40 s |
Spatial matching performs additional geographic proximity routing, and retrieving large attribute sets requires broader feature store assembly. Requesting only the levels and attributes you need keeps response times fast.
File matching
Section titled “File matching”Files are processed in bulk data warehouse operations, so processing time scales efficiently with file volume. These benchmarks include all five match levels and full attribute enrichment.
| Input records | Matching time | Speed per record |
|---|---|---|
| 1,000 | ~3.5 min | ~210 ms |
| 10,000 | ~4.1 min | ~25 ms |
| 100,000 | ~5.1 min | ~3.1 ms |
| 1,000,000 | ~7.7 min | ~0.46 ms |
| 10,000,000 | ~16.5 min | ~0.10 ms |
| 50,000,000 | ~65.5 min | ~0.08 ms |
Upload and download times depend on network throughput and file size. Large lists can be uploaded as up to 50 shards in parallel. Files, shards, and formats →
Tips for faster matching
Section titled “Tips for faster matching”- Batch real-time calls. Send batches of up to 100 records per call instead of one record at a time.
- Ask only for what you need. Request the match levels and enrichment attributes you will actually use.
- Send person IDs you already have. A recognized
iag_person_idskips rematching and refreshes data directly. Person IDs → - Use files for big lists. Lists beyond a few thousand records are faster and simpler processed through file matching.