Jiminny

The product is model output end to end and the removal test fails cleanly. Transcription across more than sixty languages, automatic notes and summaries, scoring against coaching frameworks, a question answering interface across calls and deals, and automatic population of customer record fields from conversation content are all model dependent, and stripping them leaves a call recorder with a playlist feature. That is not a diminished version of the same product.

The company has operated this way since 2016, so this is a long standing speech product rather than a recent repositioning. Held below the top band because the transcription layer that everything rests on is licensed from a third party engine rather than built, so the vendor's own modelling contribution sits in the scoring and coaching layer above it, and because the coaching workflow, playlists, snippets and manager tooling around the intelligence are conventional software that carries real weight in what buyers describe choosing it for. Ask which capabilities are the vendor's own models rather than the licensed engine, and what remains if the speech provider is unavailable.

The platform writes into the system of record on its own and no boundary is published. Automatic population of customer record fields, tasks and events from conversation content means the product reads what was said on a call and creates or changes records that pipeline reporting and forecasting then depend on, without a person entering them.

Nothing published describes which fields it may write, whether it overwrites values a person entered, whether any approval step precedes a write, or what an administrator can constrain by field or role. Separately the silent call attendance capability is an autonomy question of a different kind, since it is a human action rather than a machine one but is equally unbounded in what the public record describes: nothing states who may use it, whether its use is logged, or whether the representative being coached is told. Ask which fields automatic record filling can write and whether it overwrites human entries, and whether silent attendance is logged and visible to the representative.

The speech engine is named, which almost nothing in this category does. Independent coverage identifies a specific third party speech to text provider behind transcription, supporting more than sixty languages, and the provider publishes its own account of the relationship, so the disclosure is corroborated from both sides rather than resting on a vendor claim.

That matters more here than a generic model statement would, because transcription is the foundation every other feature reads from, and knowing which engine produces it lets a buyer evaluate that engine's published accuracy independently and ask where audio travels. Two gaps hold it below the top band.

No provider, family or version is named for the layer above transcription, meaning the scoring, the question answering interface and the record filling, which are the features that generate decisions rather than text. And no accuracy figure is published by the vendor for transcription in any language, for call scoring against human reviewers, or for the field values written back into the customer record. Ask which models power scoring and the question answering interface, and for word error rate by language.

Consistent independent placement and no measured outcome from either side. Reviewers converge on a clear qualitative verdict, describing a strong manager experience, effective coaching workflows and good value for the segment, and one directory places the product ninth in its category ranking. Contemporaneous press records roughly sixteen and a half million dollars raised, which evidences the company rather than the product.

What is absent is measurement of the thing being sold: no published study shows ramp time reduction, win rate improvement or quota attainment across a stated customer population over a stated period, and no accuracy figure is offered for the call scoring that managers act on when coaching individuals.

The same review corpus records a specific operational complaint, that data presentation lacks flexibility for tailoring reports to particular measures, which is a usability finding the vendor does not acknowledge. Ask for ramp or win rate improvement measured across a stated cohort and period, and for scoring agreement against human reviewers.

The platform sends internally rather than to prospects, which narrows this axis considerably. Notifications and nudges reach representatives and managers through messaging and email inside the customer's own organisation, and playlists and snippets circulate for training. There is no sequencer, no dialer campaign and no outbound delivery to external contacts, so consent capture and suppression have little to attach to and the axis applies in a limited form.

The one external touch is the call itself, where the obligations are recording and monitoring law rather than messaging law, and those are addressed on a different axis. What is unaddressed here is whether any of the internal notification traffic carries conversation content about external parties outside the platform's access controls, for instance a summary of a customer call pushed into a general messaging channel. Ask what conversation content the messaging notifications carry and which channels they can be directed to.

Jiminny acts as processor for recordings, transcripts and CRM data under a data processing addendum its terms incorporate, and its privacy notice relies on the EU, UK and Swiss Data Privacy Frameworks for transfers, with a named data protection officer and a UK registration with the Information Commissioner's Office.

The trust center lists 19 subprocessors with purpose and location, among them AWS hosting in the United States, Germany and Ireland, AssemblyAI and Gladia for transcription, and OpenAI, Google Cloud and Fireworks AI for Ask Jiminny. The privacy notice leaves the legal basis for recording prospects, and notice to them, with the customer as controller. Customer data is deleted on request, three years after upload by default, and 30 days after termination.

No external corpus is licensed. Everything Jiminny holds originates in the customer's own calls, email and meetings, so there is no purchased database, cooperative or third party feed to trace. Audio and text leave Jiminny's systems for named subprocessors, among them AssemblyAI, Gladia, OpenAI and Fireworks AI, and the July 2026 terms contractually restrict those providers from using customer data to train or improve their services.

The terms do not say whether recordings, transcripts or scores improve the scoring and coaching models Jiminny builds itself, and the restriction does not cover AI assistants a customer connects through its MCP server. An external participant on a sales call is recorded, transcribed by a third party and scored, and what Jiminny's own models learn from that material is unstated.

Integration runs through sanctioned routes on both estates the product touches. Native applications are published for two major customer record platforms through their own marketplaces rather than built as private connectors, meeting and call recording operates through the conferencing platforms own participant mechanisms, and messaging notifications run through the standard application surface of the messaging platform.

Mobile applications are distributed through the two main application stores, which is a further sanctioned distribution path. Nothing resembling credential storage, stream interception or unsanctioned automation appears in anything located.

Held below the top band because no stated conformance position against any specific platform's terms was found, which is what separates the bands here, and because the silent call attendance capability operates inside conferencing platforms whose own policies on undisclosed participants are not addressed anywhere in what the vendor publishes. Ask whether any named platform conformance commitments exist contractually, and what conferencing platform policies govern the silent attendance feature.

Jiminny's July 2026 terms answer the training question for the outside models it uses. Its AI features run on large language models from third party providers and on Jiminny's own machine learning technology, and Jiminny contractually restricts those providers from using customer data to train or improve their services.

The terms say nothing either way on whether recordings, transcripts or rep scores train Jiminny's own models, and the restriction does not reach AI assistants a customer connects through the MCP server. Retention is stated: data is deleted 30 days after termination or three years after upload, and the subprocessor list names the providers audio and text reach, among them AssemblyAI, Gladia, OpenAI and Fireworks AI. The terms bar customers from using the AI features for emotion recognition, biometric classification, or recruiting, monitoring or terminating employees.

A marketed capability places an undisclosed third party inside a live customer conversation. Independent review coverage describes managers joining calls invisibly to send private coaching messages to the representative while the conversation continues, with the other party unaware anyone else is present, and treats that concealment as the platform's principal differentiator rather than as an incidental feature.

Nothing published describes what the external participant is told, whether any announcement covers silent attendance as distinct from recording, or how the capability behaves in jurisdictions requiring every party to a conversation to consent. That is a different exposure from recording. A recorder can be disclosed once at the start and the person can decide whether to continue; a silent listener joining part way through cannot be consented to by someone who does not know they arrived.

This grade rests on the capability itself rather than on the vendor's own framing, which was not verified on this pass, and the architectural reading is what places it: the product is built so that a person on the call cannot know who is listening. Recording, transcription and scoring of that same conversation compound it. Ask what the external party is told about silent attendance, whether the capability can be disabled per jurisdiction or per account, and what consent the vendor considers necessary before it is used.

Jiminny integrates natively with HubSpot and Salesforce, logging notes, fields and follow up actions into the CRM and answering questions about calls and deals inside HubSpot, and it reaches other CRMs through Integration.app. A customer API has its own public developer reference, and an MCP server gives connected AI assistants read only access within each user's own permissions through OAuth. Depth at the edges, such as custom objects and sync conflicts, is not described on the main site. Capture spans phone, video and email, with Jiminny Voice as its own telephony option on Twilio, and users sign in through their Microsoft, Google, Salesforce, Slack or HubSpot accounts.

Jiminny hosts on AWS, and its subprocessor list places AWS in the United States, Germany and Ireland, while its site describes EU data centers and secure EU and US storage. The privacy notice says personal data is stored and processed in the United States, Ireland, the United Kingdom and other locations, with customer data held only where the DPA and commercial agreement allow.

Transcription runs through AssemblyAI in the United States and Ireland and Gladia in Germany, and meeting bots through Recall.ai in the United States and Germany. How a customer's region is assigned, the tenancy model and recovery objectives are not on Jiminny's pages. A disaster recovery plan is in place and uptime is stated at 99.9%.

Jiminny runs a trust center that lists SOC 2 Type II, GDPR, CCPA, HIPAA and the EU, UK and Swiss Data Privacy Frameworks, and holds its SOC 2 Type II and Type 1 reports, a penetration test summary, a system architecture and data flow diagram, an AI legal and security guidance paper and its DPA behind an access request.

It also states annual third party audits and penetration tests, a disaster recovery plan, cyber insurance, device management and SSO for staff access, with a public status page and a security address for vulnerability reports. Audit dates are not shown on the public page. In the product, customers get SSO, two factor authentication and granular permissions, and data is encrypted in transit and at rest on AWS.

Jiminny's site carries no pricing page: buyers are sent to book a demo or start a 14 day trial with no card. The Gong comparison page promises one flat fee with no usage meter and names no figure, unit or plan. The July 2026 terms set out contract mechanics rather than prices: a minimum number of users per license type set in the order form, a separately charged Jiminny Voice license with standard rates above 1,350 minutes in 30 days and for international calls, an extra fee to import more than three months of calls or a year of email at onboarding, and a 5% renewal increase when the order form sets no renewal price. A buyer cannot budget a purchase from Jiminny's own pages.

Jiminny's July 2026 terms set out how a customer leaves. Contracts renew for the same period unless either side gives 60 days notice, the renewed license count is whatever the account holds 60 days before the end, and renewal pricing rises 5% when the order form sets no other figure.

After termination, customer data including recorded calls stays available for download or export for 30 days and is then deleted, and data is otherwise deleted three years after upload unless agreed in writing. Whether scores, coaching history and Ask Jiminny answers leave in a usable format alongside the recordings is not addressed. During the term, recordings can be exported or shared by link, and the customer API gives programmatic access.

The platform operates no campaign sending infrastructure. Email and messaging traffic consists of notifications and coaching nudges directed at the customer's own staff rather than at prospects, so there is no pooled sending estate, no shared reputation and none of the warmup, rotation, bounce or complaint machinery this axis normally examines.

Call recording rides on the conferencing platforms rather than on vendor operated telephony, so call completion is not the vendor's to manage either. What is knowable is thin because there is little to know. Nothing published states whether notification mail originates from vendor domains or the customer's own, or who configures authentication, which is a small question given the internal audience but the only one this axis can ask. Ask which domain sends platform notifications and who configures authentication for it.

The served segment is clear, consistently described from outside, and supported by a commercial structure built for it. Independent reviewers place the platform with small and mid market business to business teams with meeting and call heavy motions, one naming teams under twenty representatives specifically, and identify the absence of a mandatory platform fee as the thing that makes conversation intelligence reachable for buyers previously priced out of the category.

That is a segment position expressed through pricing architecture rather than through a marketing page. European coverage is a genuine part of it, with a London base and reviewers pointing European buyers with data protection requirements toward this vendor in preference to the American incumbents. Transcription across more than sixty languages supports international use, and mobile applications on both platforms support field teams.

Held below the top band by limits reviewers state and the vendor does not: revenue intelligence depth and pipeline forecasting are described as thinner than the enterprise alternatives, reporting flexibility is a recurring complaint, and the ecosystem is smaller. Ask what the largest reference deployment is, and what forecasting the platform does not do.