Freshsales

The vendor settles this itself in the price list, which is the most reliable place to test the claim. Freddy AI is marketed across the product as the intelligence layer, and the free tier and the entry paid tier ship with none of it. Lead scoring, deal insights, next best action, duplicate detection and enrichment all begin at the third tier, so two of the four editions are sold complete with the model absent.

What remains without it is a full system of record with pipelines, workflow automation, sequences, telephony, email, chat, reporting and a mobile application. That is the removal test answered by packaging rather than by argument.

Within this product the model advises and a person acts. Freddy scores leads, flags stale deals, surfaces duplicates and drafts email, and in each case the output lands in front of a rep who decides what to do with it. Nothing contacts a prospect on its own initiative and no synthetic voice or persona is involved.

What is missing is any published oversight architecture around the advice: no accuracy statement on the scoring that determines which leads a rep works first, no route to review or contest a score, no administrator control over what the model is permitted to evaluate, and no audit trail of model driven changes to records. The vendor's more autonomous agent products sit in its service and information technology lines rather than here, which is worth stating so the grade is not read across the family.

A published trust framework covers the model layer specifically rather than leaving it to the general security page, and its contents are substantive: privacy protections, content safety filters applied to generative output, redundancy across geographies, and stated alignment with the audit standards the company holds elsewhere.

One disclosure in it is unusually candid and worth naming, because almost nobody publishes it: the vendor states that its team reviews model output performance on an ongoing basis and refines prompts or switches models when performance requires it. That tells a buyer the model behind a feature is not fixed, which is true of every vendor here and admitted by very few.

Off the top band because no provider, model family or version is named for the layer serving this product, and because no accuracy or error rate is published for lead scoring, which is the function that decides where rep attention goes. Independent testing published elsewhere put the scoring error rate in the low single digits on a small sample, which is third party rather than vendor disclosure.

The evidence base here is different in kind from most of this index, because the parent is a publicly reporting company. Customer counts, revenue and product descriptions appear in audited annual filings rather than only in marketing, the figure of more than 74,000 customers across the product family is stated in that context, and a customers page and investor disclosures are both published. Independent review platforms carry large volumes of reviews for this specific product.

Off the top band on a real gap: none of that is product specific outcome evidence. No named Freshsales customer with a quantified result, no case study with a measurement basis and no before and after figure for pipeline, win rate or ramp was located. The corporate record establishes that the company is substantial; it does not establish what the product does for a sales team.

The product sends sequenced email and places calls through built in telephony, so both regulated channels are in scope, and the published position is corporate rather than product level. At the company level there is a route to unsubscribe from marketing, a route for California residents to restrict the sharing of personal information, and a cookie policy.

At the product level nothing states a consent standard for sequenced sending, a suppression or internal do not contact process, calling window enforcement, registry scrubbing, messaging carrier registration or call authentication. For a system of record that also dials, the absence of any telephone specific compliance statement is the gap, and it is the same gap recorded against most of this category.

The privacy programme is real and independently attested rather than asserted. The company holds a certification specifically for privacy information management alongside its security certifications, which is a separate audit against a separate standard and is rarer than the security ones.

Around it sit a published privacy notice, a cookie policy, a route for California residents to restrict sharing of their personal information, a route to unsubscribe under European rules, an accessibility statement and a legal hub. Sub processors are placed under confidentiality agreements alongside their service contracts, and a data processing agreement is available to customers.

Off the top band because none of those documents was read in this pass, so no retention schedule, lawful basis description or product specific processing account has been verified, and because the privacy material is written for the whole product family rather than for this product.

The system of record itself is clean: the contacts, accounts and deals in it belong to the customer and the vendor sells no database and resells nothing. The provenance question arrives through one feature. Contact enrichment is sold as part of the top tier, which means records in the customer's system are being augmented with data the customer did not supply, and no source, supplier, coverage, refresh cadence or accuracy standard is published for it anywhere.

Nor is there a lawful basis statement, a notice, a lookup or a removal route for the person whose details are being added. This is the same unresolved shape already recorded against another system of record in this index, where enrichment was described only as drawing on various sources.

Everything runs on infrastructure the vendor operates or licenses directly. Telephony is its own product within the same family rather than a borrowed channel, email and calendar connect through the sanctioned interfaces of the two major providers, and third party connections run through a published marketplace and a developer platform.

No browser extension harvests another platform's interface, no account credential is held for a service the vendor does not own, and nothing is rotated or paced to stay beneath another company's limit. Off the top band because no conformance position is stated in either direction: nothing addresses the telephony and messaging registration obligations that attach to a product with a dialer in it, and nothing describes what happens to a customer's integrations if a connected platform changes its access terms.

A dedicated trust framework for the model layer is published and enumerates real safeguards: privacy protections, content safety filtering on generative output, redundancy across geographies, and a stated alignment with the security, privacy and health data standards the company is independently audited against. That is materially more than the silence recorded against most vendors on this axis.

It sits at the bottom of the band rather than higher because the question this axis exists to ask is not answered in plain terms. Nothing located states whether customer records, activity history or email content from one tenant contribute to model improvement serving other tenants, and no contractual no training commitment of the kind the highest graded vendor here publishes inside its subscription terms was found. Alignment with an audit standard is a statement about controls, not about whether the corpus is pooled.

The product carries no concealment features, and in this category that is worth saying plainly: there is no local presence, no caller identity rotation, no synthetic voice, no invented persona and no manufactured research. Outbound calls carry the customer's own number and outbound email carries the rep's own address. What is unaddressed is the artificial half.

Model drafted email goes to prospects with nothing indicating a model composed it, a website chatbot converses with visitors and no identity disclosure position for it was located, and the European transparency obligations that took effect in August 2026 are not mentioned anywhere reachable. A clean record on impersonation with no stated position on disclosure is the middle of this band.

A published marketplace carries roughly three hundred applications, a developer platform is available for building against, and native connections cover the two major email and calendar providers. The deeper integration story is inside the vendor's own family, where the telephony, service desk, chat and marketing products share a customer record with this one, so a buyer already holding those gets a genuinely unified surface rather than a synchronisation.

Off the top band on breadth rather than depth: three hundred applications is described by independent comparisons as the thinnest marketplace among this product's direct competitors, so a team running specialist tooling has a real compatibility question to answer before committing, and no object level documentation of what synchronises in which direction was read.

This sits at the top of the band on structure rather than on published detail. A trust portal exists on its own subdomain and is the stated route to compliance documentation, sub processors are placed under confidentiality agreements alongside service contracts, infrastructure is described as running in dedicated segmented private cloud networks, and the model layer trust material refers to redundancy across geographies, which implies more than one region without naming any.

What could not be verified on the public pages read is the part a buyer with a residency obligation actually needs: no region list, no data centre location, no statement of whether a customer may choose where their instance sits, and no published sub processor register. The documentation route exists and the values behind it were not reachable, so this row should be re verified through the trust portal before it is relied on.

This is the most complete security record graded in this index. The certification set is broad and independently audited at least annually: information security and privacy information management under the international standards, service organisation attestations at both type two and the public summary level, a financial controls attestation, the payment card standard, a cloud security registry entry and a United States state cloud programme authorisation.

A dedicated trust portal on its own subdomain carries the documentation. Governance is named rather than implied, with a cybersecurity and privacy steering committee of executive leadership reviewing posture quarterly and a security organisation reporting to a chief information security officer.

Application security assessments including code review, vulnerability assessment and penetration testing are carried out both internally and by independent accredited firms, audit findings are reported to that committee and tracked until closure, and a responsible disclosure route is published.

The detail that pushes this to the top band is the dating: report periods are published with their exact coverage windows, a bridge letter is issued to cover the gap between periods, and the next report period and its expected availability date are stated in advance. Publishing the audit calendar forward is rarer than holding the certifications.

One defect is recorded without changing the grade: the trust page still describes policies as aligned to the 2013 revision of the information security standard, which was superseded in 2022 with the transition period closing in October 2025, so the page text is stale against a certification programme that is demonstrably current.

Four tiers are published and consistent across independent sources: a free edition capped at three users, then roughly nine, thirty nine and fifty nine dollars per user per month on annual billing. The feature gates that decide which tier a team actually needs are documented rather than hidden, including the model layer starting at the third tier, duplicate management at the third, enrichment at the fourth, and a storage allowance of five gigabytes per user at the third.

The add on economics are published too, with the model copilot priced per agent per month and conversational bot sessions priced per hundred after an initial free allowance. Off the top band because the real bill is still hard to size in two places that matter: telephony is metered separately through the vendor's phone product rather than included, so the feature this product is most differentiated on sits outside the seat price, and a buyer on the third tier is paying for a tier that advertises the model layer and can still be sold a separate copilot on top of it. The vendor's own pricing page returned an error in this pass, so the figures are carried from consistent independent records rather than read directly.

A legal hub, site terms and a data processing agreement are all published, so governing documents exist and are reachable, which comfortably clears the bottom band. Their substance on exit was not established in this pass. No export format, bulk download route, post termination retention window, deletion timeline or data return mechanism was located on any public page, and for a system of record that is the axis's central question, since the corpus at stake is the entire pipeline and activity history of a sales organisation rather than a campaign. A published marketplace and developer platform provide a programmatic route to extract records, which is a practical position rather than a contractual one. Re verify through the legal hub.

Both sending channels are first class in this product and neither carries a published discipline position. On email, sequences send from the platform and nothing is published on warmup ramps, bounce handling, complaint thresholds, sender authentication guidance, reputation monitoring or what happens to an account whose sending degrades.

On voice, calls originate from built in telephony and nothing addresses call authentication attestation, number reputation, spam labelling or answer rate monitoring. The vendor's corporate infrastructure quality is not in question and is graded elsewhere; what is missing is any statement about protecting the customer's own sending and calling reputation, which is what this axis measures.

The packaging describes the target more precisely than the marketing does. A free edition capped at three users, an entry tier at single digit dollars and a ladder to a tier with territory management and custom modules maps a clear path from a solo operator through a small team to a structured mid market sales organisation, and independent analysis places the working centre of gravity at roughly five to fifty users.

The product family around it is segmented by function rather than by size, with separate products for service, information technology and marketing, so a buyer can tell what this one is for. Off the top band because the vendor also positions the suite as being for businesses of all sizes, which claims the whole market rather than describing a place in it, and that is the same pattern that has held other vendors here off the top of this band. No vertical or geographic focus is named.