Zocks

Every downstream function depends on one conversion working: spoken client conversation turned into structured records. The models are the product, not a layer above one. Founded in 2022, the company has never had a pre inference version of itself. Without the inference nothing is left, because there is no recording to fall back on and no manually maintained database underneath.

The agentic layer extends the same dependency across an adviser's whole book, answering questions about the client base and proposing actions. The company describes a multi agent architecture with separate agents assigned to particular workflows.

The oversight design is published, and it is sensible. The agentic layer proposes rather than acts. It surfaces an opportunity across the book and suggests a next action, and the adviser completes that step with a deliberate click. That puts a human decision between the model's judgment and anything that reaches a client. The company states plainly that the agents operate with the user in full control.

Enterprise deployments carry fine grained access controls and a named supervision function, a specific regulatory concept in this market rather than a general word. The drafting outputs are less clearly gated than the agentic ones. Client emails and completed intake forms are generated automatically, and nothing published describes a review step before they leave, or whether generated text is marked as such.

The accuracy claim is superlative and unmeasured. The company states it sets the industry standard with unmatched accuracy and speaker attribution in complex multi participant conversations. It publishes no figure behind that, no word or attribution error rate and no benchmark against any named alternative. There is no evaluation method or drift statement either. The claim carries weight, because attribution decides which participant in a family meeting is recorded as having said a thing. The note becomes the firm's record of a conversation that was never recorded and therefore cannot be replayed to check.

Several enterprise customers are named openly. They are large, checkable institutions rather than descriptions of a firm type, covering broker dealer networks, adviser aggregators and an insurance group. One further wealth manager selection was separately announced. A stated base of more than five thousand financial firms gives scale.

The quantified claim, that advisers save more than ten hours a week on administrative work, is repeated consistently across funding announcements and product material. It is entirely self reported. The named institutions and the number never join: no case study attaches a measured result to any identified firm.

Zocks' privacy and security page says data captured by Zocks is never used for training or improving large language models. Conversations are never recorded. Each firm gets its own retention policy, users can set retention at the object level, and transcripts and summaries can be deleted on demand or on a schedule by the hour, day or month. Transcripts can be archived to Smarsh, Global Relay and Proofpoint for supervision, and customer data is removed when a customer leaves. The privacy policy says AI features use LLM technology from trusted third party providers without naming them, and nothing states what those providers retain.

On privacy the vendor has done something structural rather than said something reassuring. The platform captures and attributes multi participant client conversations without making recordings. That architectural decision removes an entire category of sensitive material from existence rather than protecting it after the fact. It is stated consistently across product pages and every announcement, not once in marketing.

Retention is described as the customer's decision, covering how much is kept and for how long, which places control with the regulated firm that carries the obligation. The detail behind it is missing. Nothing published says what structured data is retained once the conversation is gone, and there is no public retention schedule, deletion commitment or data processing statement.

A SOC 2 Type II attestation is named on Zocks' security page, which publishes its control set in four areas: infrastructure, organizational, internal procedures, and data and privacy. Controls listed include unique authentication to production systems and databases, restricted access to encryption keys, firewalls and the production network, and MFA and encryption on remote access.

Access is revoked on termination within set times, and penetration testing runs annually with a remediation plan. Formal retention and disposal procedures govern data, and customer data is removed when a customer leaves. A subprocessor section is part of the same page. The auditor and report period are not named.

A technology supplier of this kind needs no license and has none. The downstream question is live and only partly addressed. Client meeting records and the notes derived from them are books and records for a regulated advisory firm, subject to retention obligations and to examination. The company does name a supervision capability for enterprise firms, which acknowledges the obligation exists. What is absent is the specification.

Nothing says whether records produced by the platform satisfy retention format requirements, or how a firm evidences the completeness of a note produced from a conversation that was never recorded. Nor is it said what an examiner would be shown.

Transcription and speaker attribution can perform differently across accents, dialects and speech patterns, and a poorer result means a less accurate record for the client concerned. The opportunity engine raises a second question. It scans a book of business and surfaces which clients merit a next action, such as held away assets, planning opportunities and major life events, so it directs adviser attention toward some clients and away from others.

Any systematic tendency in that ranking shapes who gets served. No evaluation across speaker populations is published, and nothing explains how opportunities are ranked or whether that ranking has been reviewed.

Zocks' terms of service, last updated 8 October 2025, provide the service as is and as available, with no warranty beyond a mutual promise of authority and legal compliance. Liability on both sides excludes indirect and consequential damages and is capped at the fees Zocks received in the six months before a claim.

Zocks defends customers against claims that the service infringes intellectual property, and the customer defends Zocks against claims arising from its own breaches or unlawful use. California law governs. No accuracy commitment, correction obligation or service level appears in the terms. Because Zocks keeps no recording, a note that misstates what a client said cannot be compared with the source, and the note becomes the firm's record of the conversation. If a dispute later turns on what a client instructed, the firm has only the machine generated note to go on.

Zocks describes its stack as proprietary and names no language model provider, model family, version or inference host on its product pages, funding announcements or press material. Its security page lists AWS as a subprocessor for hosting. The privacy policy says AI features use LLM technology from trusted third party providers without naming them, and Zocks states that no client data is stored on LLMs. The MCP server lets a firm connect Claude, ChatGPT and Gemini to its Zocks data. Those are the firm's own assistants reaching in, and naming them says nothing about whose model listens to a client conversation and writes the note.

Zocks reads from and writes back to more than 40 systems through out of the box connectors, a suite of REST APIs and an MCP server, and its summary for AI assistants names them. CRMs are Wealthbox, Salesforce, Redtail, Practifi, XLR8, Zoho, HubSpot, SmartOffice, Microsoft Dynamics 365, AdvisorEngine and the SS&C Black Diamond Elements, Engage and Edge products.

Conversation data populates planning, onboarding and proposal systems, and Zocks says it fills more than 370 fields in eMoney or RightCapital from a single conversation. Holistiplan, Wealth.com, Orion, Black Diamond reporting and PreciseFP are connected too. Meetings come in from Zoom, Teams, Google Meet, RingCentral, GoTo Connect, Zoom Phone, Teams Phone, Intulse, Webex and landline or mobile calls. Archiving runs to Smarsh, Proofpoint and Global Relay, Zapier reaches other tools, and the MCP server connects Claude, ChatGPT and Gemini. Field mapping for each connection is not published.

The platform is delivered as hosted software and nothing about topology is published. There are no published hosting regions, storage location or tenancy statement. Nor is there a private deployment option for enterprise firms or any contractual residency commitment. The protocol server raises a sharper question, and it is unanswered. When a firm reaches its own client conversation data from a general purpose assistant, that data crosses into an environment the vendor does not control. Nothing states where the resulting processing happens or whose terms then govern it.

Prices are public, per user and per month. Essentials costs $67 billed annually or $80 monthly and covers meeting preparation, notes without recordings, CRM, planning, portfolio and tax planning integrations, pre meeting and follow up emails and unlimited meetings. Professional costs $117 annually or $140 monthly and adds Zocks MCP, form filling, reporting and analytics, client profiles, an advisor coaching center, team sharing, admin assistant seats and one onboarding session.

Ultimate costs $184 annually or $220 monthly and adds native scheduling, automatic replies to client emails, document extraction, Zapier workflows and three onboarding sessions. Annual billing gives two months free, add ons are listed at $25 a month, and a free trial is offered.

The stated range runs from solo practitioners through mid sized practices to enterprise deployments, with more than five thousand financial firms claimed. Named enterprise customers span independent broker dealer networks, registered investment adviser aggregators and a mutual insurance and annuity group, a real spread of institution types within advice.

Enterprise capability is described in terms that match how these firms are actually organized: organizational hierarchies, fine grained access control and supervision at firm level rather than a single tier product. Two limits matter, geography and adjacency. The footprint appears to be the United States alone. Investors describe ambitions across wealth, banking and insurance, but the published customer base sits almost entirely in advice.