Objections, Not Questions
An applicant who asks "why do you need my SSN" gets an answer from both. The one who says "I'll finish it later" needs persuasion, and that is what the model is trained on.
Sales persuasion, from the first signal to the next sale.

Sellence vs Gradient Labs
Gradient Labs runs customer operations for banks and lenders. Sellence is a sales model that texts each customer one to one, in your name, until the application is finished.
01
A customer operations agent answers the applicant's question and passes the case on. A persuasion model answers the objection and finishes the sale.
02
| Dimension | Sellence | Gradient Labs |
|---|---|---|
| What it is | Persuasion model for financial services | AI agents for customer operations in finance (per site) |
| Goal of each conversation | Finish the application and fund the account | Resolve the query or complete the case (per site) |
| Who persuades the applicant | The model, in your name. Questions needing a licence go to your team | The agent answers questions and links back to the step left (per site) |
| How long the thread runs | Weeks, one thread per applicant that remembers every reply | Sent when the applicant drops off, or on a schedule (per site) |
| Channels | iMessage, WhatsApp and text, with voice and web chat. No email | Chat, email, text and voice (per site) |
| Objections and persuasion | Trained on 3.5M+ real sales conversations | Not publicly stated |
| Scope beyond the sale | None | Disputes, KYC, collections, support, back office (per site) |
| How the result is measured | 2-month test against customers it did not text | Not publicly stated |
| Entry price | Paid pilot, quoted per company, then a 12-month term | Not publicly stated |
Sellence loses on breadth of operations, email and named bank customers. It wins on who persuades, objection handling and measurement.
03
An applicant who asks "why do you need my SSN" gets an answer from both. The one who says "I'll finish it later" needs persuasion, and that is what the model is trained on.
One applicant, one thread. The model remembers the fee question from 3 weeks ago and picks up there, in local time, until the account is funded.
Half the eligible applicants get the model, half keep your process. The step you count as finished is compared between the 2 groups after 2 months.
04
Nations Lending runs it on its loan applicants. Generali runs its conversations on it.
05
You are a bank or lender with a support queue, disputes, KYC checks and collections to run, and you want one vendor across all of it. Gradient Labs publishes a lending pilot and rollout at LHV Bank (FinTech Futures, read 4 October 2026) and covers email, which Sellence does not. Sellence does one job: the application that was never finished.
06
Share of abandoned applications that get funded, half with the model, half without, over 2 months.
Share of conversations finished without a loan officer or agent touching the thread.
What each agent says to "I'll come back to it": a reminder, or an answer to the objection.
Compliance review of the transcripts: consent on file, STOP texts obeyed, fee claims in approved wording.
07
Gradient Labs runs customer operations for financial services: support, disputes, KYC, lending cases. Sellence is a persuasion model that handles the objection and follows up for weeks until the application is funded.
No. Identity, eligibility and credit decisions stay with your KYC provider, your core banking system and your sponsor bank. Sellence runs off the updates they send and writes the outcome back.
Gradient Labs does not publish pricing on the pages we read. Sellence runs a paid 2-month pilot on one journey, measured against customers it did not text, quoted per company, then a 12-month term.
Weeks. Our engineers connect it to your systems, your compliance team approves the wording, and your team either approves each reply or lets it send within rules you set.
When you need one agent across support, disputes, KYC and collections, with email in the mix. When the number to move is applications finished and funded, you need the model.
Sources: gradient-labs.ai and /guides/best-ai-agents-for-lending, trysalient.com, cascading.ai, taktile.com, fintechfutures.com on LHV Bank (read 4 October 2026), gradient-labs.ai/use-cases/incomplete-application-follow-up (read 5 October 2026), sellence.com/llms.txt. Written by Sellence, every competitor fact from its own pages.
The Agent Handles The Application. The Model Finishes The Sale.