Home / How it works
One engine, ten stages, and a named person at every checkpoint.
This page walks a program from brief to pipeline: what the agents do at each stage, who approves it, what you see, and what you can hold us to in the program agreement.
Every stage has an owner, an output and a checkpoint.
Read across for a stage, down for a role. Agents do the work in the first lane, our people sign in the second, and you stay in control in the third.
Yellow diamonds mark a Saravus person signing. White diamonds mark a decision that stays with you. Pausing a program and honoring an opt-out are the only actions an agent takes on its own, because both reduce risk.
Ten checks run on every lead before a person releases it.
- Capture integrityForm, timestamp and source page verified
- Consent proofRoute and wording valid for the recipient's country
- Email validityDeliverable, with role and catch-all addresses handled per your rules
- Identity and roleA real person, a current title and the right buying-group seat
- Account matchBelongs to your target account list
- Program rulesGeography, company size, industry and title filters
- DuplicatesWithin the program, against history and across programs
- SuppressionsCustomers, open opportunities, competitors and opt-outs
- Fraud and bot signalsVelocity, device and pattern screening
- Qualifying answersYour questions, answered and recorded
Every program states where its contacts come from and who controls the data.
General information, not legal advice. Each country rulebook is reviewed by counsel before we deliver into that country, and any country without one is treated as opt-in.
Service levels you can write straight into the agreement.
| Commitment | What it covers | How it's set |
|---|---|---|
| Feasibility and price | Reach by country and role, consent route and a firm price | Two business days, standard |
| Pacing | Daily delivery against plan, with a tolerance band and a shared report | Agreed per program |
| Acceptance window | Days you have to return a lead with a reason | Agreed per program |
| Make-goods | Replacement inside the window, or a credit if we can't replace | Agreed per program |
| Response time | A person answers every message about the program | Agreed per program |
| Suppression | Customers and open opportunities excluded from every program | Standard |
| Data return and deletion | Program data returned or deleted at close, confirmed in writing | Agreed per program |
| Billing | Rates per accepted lead, covered role or program, in euros or US dollars | First program billed on accepted leads only |
We deliver into the marketing stack you already run.
Field maps, campaign member status and lead source values are set once and reused. Direct connectors are switched on per program; until then we deliver in your template or through the platform's supplier upload.
Fourteen agents, each with one job and a hard limit.
| Agent | Job | What it can do on its own |
|---|---|---|
| Orchestrator | Plans each program and routes the work | Plan only |
| Brief reader | Turns briefs and TALs into a spec ledger with citations | Draft only |
| Feasibility planner | Counts reach by country, role and size | Draft only |
| Quote explainer | Explains the price, which is calculated from the rate card | Draft only |
| Audience builder | Matches accounts, maps roles, sets consent per contact | Staging only |
| Content writer | Writes the master in the program's brand kit | Draft only |
| Localizer | Writes each language from the approved master | Draft only |
| Sequence designer | Builds outreach sequences | Paused sequences only |
| Pacing controller | Watches delivery health | Pause, never resume |
| Reply handler | Classifies replies and drafts answers, disclosing that it is AI | Honor opt-outs; replies need approval |
| Lead checker | Runs the ten checks and writes the audit trail | Reject, never release |
| Signal watcher | Spots intent surges and missing roles on engaged accounts | Propose only |
| Returns analyst | Maps return reasons and tightens checks | Draft only |
| Reporter | Writes the narrative around figures from our database | Draft only |
Every agent run is logged with its inputs, output, model, prompt version and the person who approved what followed.