ADI AGENTS · ORIGINATION
Pre-qualification in conversation, not in a form
A web and voice assistant that captures the lead, pre-qualifies inside your credit policy and carries it to the offer.
The operational jump
From form to offer
Both models get the same traffic. What changes is how many reach the end.
Form and scripted chatbot
ADI origination agent
Results in production
Impact that is measured
Aggregated platform results, measured in production. Not attributed to any specific account.
How it runs
From the first sentence to the offer
Web or voice
The applicant arrives through the site widget, WhatsApp or the phone, inbound or outbound.
Free speech, no fixed order
Information is extracted from speech rather than typed into a field. The applicant answers at their own pace.
Score and policy
HTTP Tools query your systems during the conversation: score, credit policy and eligible products.
Deterministic eligibility
The model runs the dialogue; the decision happens outside it, in code, against your rule.
Product and terms
Amount, term and rate presented with correct pronunciation of values and percentages.
Lead in your system
Data extracted from the transcript and delivered to the origination system by signed webhook.
Capabilities
Converting without rushing
Out-of-order capture
If the applicant volunteers three facts in one sentence, all three are captured and never asked again.
Pronunciation dictionary
Amounts, rates, terms and documents read as a trained agent would read them. CPF, CURP and RFC included.
Cross-channel resumption
Starts on the site, continues on WhatsApp, ends on voice. The context comes along.
In-conversation quoting
Term and amount recalculated in speech, within the eligible products.
Strict topical focus
The agent does not offer opinions outside origination. Out of scope, it routes to an operator.
Escalation with the full thread
The operator arrives with the complete conversation and the data already collected.
Feasibility
Connects to what you already have
Credit engine, bureau and CRM connected through a documented interface. No core changes.
Onboarding in 1 to 3 weeks. With an existing SIP trunk, the voice channel goes live in 2 to 5 days.
Inside the rules
Originating without leaving policy
Guardrails in code
Deterministic filters before and after the model. Prompt injection and PII exfiltration blocked at the edge.
Tokenised PII
Personal data is tokenised before any call to a model provider.
Traceability
Every interaction reconstructable: what the agent said, on which version, under which controls.
Closed catalogue
The agent executes only actions registered as enums validated outside the model.
The full cycle
One engine, five agents
One compliance approval covers all five stages. You start with the one that hurts most and expand with no new project.
Onboarding
Sign-up and activation with deterministic identity validation.
View →Transactions
Voice-authenticated in-app operations, in seconds.
View →Collections
Negotiates inside the defined policy and logs the commitment in the CRM.
View →Support
Resolves on first contact, with context across all three channels.
View →FAQ
Frequently asked questions
Does the agent decide credit?
No. The agent runs the conversation and collects the data; the eligibility decision runs outside the model, against your policy, and returns a deterministic result. What the agent communicates is the result of that lookup.
How does the lead reach our system?
Through a webhook signed with HMAC-SHA256 at the close of the conversation. The fields are defined beforehand in Data Collection, with a type and an instruction per field, and extracted from the full transcript regardless of the order the information appeared in.
Does it work on our site without heavy development?
The web widget is embedded in your site and authenticated. Agent configuration, including prompt, knowledge base and tools, is done in the no-code editor.
What if the applicant switches channel halfway?
The context comes along. The conversation can start on the site, continue on WhatsApp and end on voice without the applicant repeating what they already said.
How does the Risk team audit this?
Every interaction is recorded and reconstructable, with the agent version and the controls active at the time. Audit logs contain no personal data — PII is tokenised before any call to a model provider.