Agentic AI for campaigns.

Autonomous voter outreach — fully managed, start to finish.

Pilot · 2026
+877%

Conversion vs. legacy channels. Same district. Same budget.

Best result observed in 2026 pilots. Individual results vary.

The channels you pay for are reaching fewer people every cycle.

Bulk-blast SMS gets filtered, email open rates are falling, and Meta CPMs keep climbing. And every message is still drafted by hand, one at a time.

Sources: Campaigns & Elections (10DLC enforcement, Feb 2026); VoterVoice 2025 Advocacy Benchmark; AdAmigo Meta CPM benchmarks, 2026.

Built by intelligence veterans.

Persuasion and targeting are all we’ve ever done. We built Polis-1 to do it for your campaign.

Other platforms hand you software. Parla does the work.

You set the strategy. Polis‑1 finds the voters, writes every message in their language, and sends at the moment they’ll act.

Powered by Polis‑1 · patent‑pending

Under the hood.

Polis‑1 models every voter in the district: propensity, persuadability, language, and the moment they’ll act. It turns that into the sends you approve.

polis‑1.internal · model-ops
Live
prod · us‑west‑2 · single‑district scope · encrypted at rest synced 6 min ago · run 7f3a9c
core · v4.2.1 442,180voters modeled 1,240signals / voter 94.2%confidence 6hrefresh Calibrated

Persuadable identified

184,200
41.6% of file

High-propensity

96,400
score > 0.7

Predicted lift

+6.4pt
vs holdout

Signal match rate

92%
file → signals

Qini

0.31Δ+0.05

Precision

0.81@0.5

PR-AUC

0.941+0.03

Inference

40msp95

Recall

0.88@k

Holdout

p<.01n=8k

Projected vote share · to election day

Deployment

Environmentprod
Regionus-west-2
DatasetCA-04 442k
Checksuma3f9c1

Propensity surface · Polis-1

PersuadableLocked

Voter clusters · k=6

Optimal send window

6a1p9p

Inference stack

Identityrecord linkage
Features1,240 signals
Propensityboosted trees
Persuasionuplift causal
Send-timecontextual bandit
Languagetransformer

Language mix

English48%
Spanish27%
Mandarin14%
Vietnamese8%
Cantonese3%

District coverage

Low High

Cumulative lift by decile

Polis-1 telemetry

Build7f3a9c
Calibration0.98
Feature drift0.4%
Segments live312
Languages11 native

Live inference

8s
Scored 1,240 voters
gbdt
26s
Uplift recomputed · cohort 4
causal
1m
Send-time re-solved · subj a3f9…2213
bandit
2m
Language routed · ES
xlm

Representative Polis‑1 console. Modeled per district.

Compliant by design.

What a campaign can legally send is a maze of federal, state, and local rules. Regulators fine the ones who slip. Every message Parla sends already clears it.

The right disclaimers

The required “paid for by” on every message, to the letter in every jurisdiction.

AI disclosure

Added automatically wherever the law requires it.

State and local rules

The ordinances layered on top of federal law, applied for your district.

Consent, opt‑outs, and quiet hours

Only reachable voters, every STOP honored instantly, never sent at an unlawful hour.

The results, in their words.

“Our mail vendor quoted $175K for the cycle. Parla covered the district for $40K.

Candidate · County Supervisor · California · 2026

“We got flagged as spam on the first SMS run. Parla didn’t.”

Candidate · State Senate · California · 2026

“Our digital firm spent $90K on Meta for fewer impressions. Parla moved more voters.”

Candidate · US House · California · 2026

Previous SMS vendor: English only. Parla: Spanish, Mandarin, Vietnamese — same week.

Candidate · US House · California · 2026

Our mail vendor quoted $175K for the cycle. Parla covered the district for $40K.

Candidate · County Supervisor
California · 2026

We got flagged as spam on the first SMS run. Parla didn’t.

Candidate · State Senate
California · 2026

Our digital firm spent $90K on Meta for fewer impressions. Parla moved more voters.

Candidate · US House
California · 2026

Previous SMS vendor: English only. Parla: Spanish, Mandarin, Vietnamese — same week.

Candidate · US House
California · 2026

Win your race.