Short answer
Aide meets all six criteria: SOC 2 Type II, an AI level per topic so a person signs off regulated replies, an audit log, tests on past conversations, account lookups through an API, and a published price. Fin and Lorikeet also meet six; Decagon, Gradient Labs and Sierra meet five.- Options compared
- Aide, Fin, Lorikeet, Decagon, Gradient Labs, Sierra, Ada.
- Criteria
- SOC 2 Type II, human sign-off, audit record, tested on history, account lookup, published price.
- Order
- Most criteria met, then the lowest cost for a 1,000-conversation month.
- Compliance
- All seven state SOC 2 Type II on their own sites.
- Checked
- November 16, 2026, on each vendor's pricing, security, industry and help pages.
- Disclosure
- Aide makes this page and is one of the seven.
The six criteria
Each is applied to what the vendor's own pages said on November 16, 2026; anything they didn't confirm is marked "Not confirmed" and doesn't count. They come from a fintech inbox: card declines, transaction disputes, account restrictions, interest questions and requests that must go to a person.
- SOC 2 Type II: stated on the vendor's own site.
- Human sign-off: a reply or action waits for a person's approval, or the AI drafts for a person to send.
- Audit record: what the AI did in each conversation is recorded for compliance to review.
- Tested on your history: it can be tested on your own past conversations before launch.
- Reads the account: it reads the customer's record from your own system before it answers.
- Published price and self-serve start: no sales call needed.
Tools are ordered by criteria met, then by the lowest cost for the month below; ties with no published price are alphabetical.
Side by side
| Tool | SOC 2 Type II | Human sign-off | Audit record | Tested on your history | Reads the account | Price and start |
|---|---|---|---|---|---|---|
| Aide | Yes, security page | Yes, each topic is AI Agent, Drafts or None | Yes, per conversation, plus a two-year audit log | Yes, checks on the last 50 conversations, replays | Yes, by API, on email, helpdesk tickets and WhatsApp | Yes, free plan; $149 |
| Fin | Yes, financial services page | Yes, holds a draft (beta) or pauses a Procedure | Yes, inputs, AI decisions, handoffs, triggers | Yes, Simulations from a real conversation | Yes, data connectors or MCP | Yes, 14-day trial; $297 |
| Lorikeet | Yes, pricing page | Yes, human review by scenario, approval gates | Yes, every interaction, model choice and action | Yes, simulations from imported tickets | Yes, live account context | Yes, 30-day trial; $2,397 |
| Decagon | Yes, its security page | Yes, Agent Assist drafts for an agent | Yes, Watchtower monitors every conversation | Yes, uploaded transcripts become tests | Yes, core banking systems, read and write | No, demo |
| Gradient Labs | Yes, home page | Yes, on dispute outcomes; no copilot | Yes, QA agent reviews every conversation, audit trail | Yes, batch tests on historical conversations | Yes, API or CSV upload | No, demo request |
| Sierra | Yes, September 2026 post | Yes, Live Assist drafts for reps | Yes, a reasoning trace per decision | Yes, annotated conversations become tests | Yes, systems of record | No, no price on its site |
| Ada | Yes, financial services FAQ | Not confirmed; hands off to a person | Yes, logged per conversation | Yes, test cases from a real conversation | Yes, core banking and custom systems by API | No, pricing page books a demo |
The seven, in order
1. Aide: best for per-topic control, a self-improving agent and a free start
Each topic is set to AI Agent, Drafts or None, and new topics start at Drafts, so a dispute or an account restriction is drafted for a person to send or handed to your team. Aide applies the topics, levels and procedures your team sets, and the rules stay yours. A custom connector looks up the customer who wrote in and reads the fields you pick, such as a payment date or an amount due. The Action Trace shows each conversation's topics, sources and handoff notes, and the audit log exports as CSV. Every night Aide compares its drafts with what your team sent, tests each change and lists it with an Undo. It runs inside Zendesk, Gorgias and Front, or on its own. More in Aide for financial services and security.
2. Fin: best for outcome pricing on the helpdesk you already run
Fin runs on Intercom's helpdesk or on Zendesk, Salesforce and others at $0.99 per outcome, 50 outcomes a month minimum on your own helpdesk. Procedures run card disputes, fraud claims and KYC checks and pull account data through data connectors or MCP. Human in the Loop, in beta with the Intercom helpdesk, has Fin ask a teammate to approve before it replies. More in Aide vs Fin.
3. Lorikeet: best for regulated teams that want forward-deployed support
Lorikeet is built for financial services, healthcare and insurance. It files disputes, processes payments and verifies identity in your systems, across chat, email, voice and SMS. Start and Scale plans include forward-deployed support. The Start plan is $2,100 a month paid annually plus $0.99 per resolved chat, email or SMS ticket; escalations are not charged.
4. Decagon: best for banks and lenders that want core banking integrations
Decagon's financial services page covers consumer banking, lending and wealth, and says it reads and writes core banking, loan and card processing systems. Trace View shows the agent's decision path, and its security page lists PCI DSS and HIPAA beside SOC 2 Type II. More in Aide vs Decagon.
5. Gradient Labs: best for disputes, collections and KYC run end to end
Gradient Labs is built for financial services, with specialist agents for lending, disputes and KYC on email, text and voice. Its disputes agent gathers evidence, decides the outcome and sends it to a person for sign-off before filing the chargeback. Pricing is by outcome with no platform fee, quoted through a demo, and its FAQ says it offers no copilot for human agents.
6. Sierra: best for taking card and ACH payments inside the conversation
Sierra's payment infrastructure holds PCI DSS Level 1 certification for card and ACH payments on voice and chat, and in September 2026 it announced AIUC-1 certification. Live Assist drafts replies for care reps. More in Aide vs Sierra.
7. Ada: best for banks and credit unions that run procedures as Playbooks
Ada turns procedures such as PIN resets and card replacements into Playbooks, connects core banking, payments and fraud systems through APIs, and lists HIPAA, PCI DSS and AIUC-1. Its docs record a reason for every handoff; an approval or draft step is not listed on the pages we checked. More in Aide vs Ada.
What 1,000 conversations a month costs
Assume, for the math only, that the AI works all 1,000 conversations, resolves 300 and writes 2 replies in each.
- Aide: 2,000 AI replies fit Pro, which includes 4,000: $149.
- Fin, on your own helpdesk: 300 outcomes at $0.99: $297.
- Lorikeet: the Start plan at $2,100 plus 300 resolutions at $0.99: $2,397.
- Decagon, Gradient Labs, Sierra and Ada: not published.
Per-resolution bills grow as the AI resolves more: at 600 resolved, Fin is $594 and Lorikeet $2,694. Aide bills AI replies, never the replies your team writes. Every plan is on Aide's pricing page.
How to choose
- Control topic by topic, nightly learning and a free start: Aide.
- Outcome pricing on the helpdesk you already run: Fin.
- Forward-deployed support for a regulated rollout: Lorikeet.
- Disputes, collections and KYC handled end to end: Gradient Labs.
- Card and ACH payments inside the conversation: Sierra.
- AI on phone calls: Fin Voice, Lorikeet, Decagon, Sierra, Ada or Gradient Labs.
- Start without a sales call: Aide (free plan), Fin or Lorikeet (trials).
Sources
Checked November 16, 2026 against each vendor's own pages: Fin's pricing, financial services page, Human in the Loop, approvals in Procedures and simulations for Evals; Lorikeet's pricing, financial services page, simulations and FAQ; Decagon's financial services page, security page, Simulations post and Agent Assist post; Gradient Labs' home page, pricing and evaluation guide; Sierra's financial services page, trust page, Live Assist, AIUC-1 post and agent development post; Ada's financial services page, Simulations docs, handoff reasons and pricing page. Also checked and left out: Glia, Posh AI and Kasisto, platforms for banks and credit unions, and NiCE Cognigy and Forethought, none of which shows a price on its home page; eesel AI, whose site says its SOC 2 Type II audit is in progress; and Zendesk AI agents, which run inside Zendesk and are compared in Aide vs Zendesk AI. Aide's facts come from its shipped product and its security page. Aide makes this page and is one of the seven; to correct a fact, write to support@aide.app.