Practifi

Practifi's own commercial structure treats AI as an add on. The client relationship system, pipeline, workflow automation, compliance monitoring and reporting were the product before inference arrived. The intelligence suite shipped in December 2025 as an add on that existing clients activate through an account representative, not as a capability woven through what they already had. Strip it out and the platform runs.

The company says the suite applies an AI first framework across the platform, which sits awkwardly beside its being sold and switched on separately. A second product, described as natively AI based, is announced for release on 25 August 2026, and would change this picture if it ships as described.

The autonomy is described and the oversight is not. Published material says follow ups are generated automatically and workflow actions are triggered without manual intervention. That is the capability being sold, stated as a benefit with no qualification. Nothing published says what a person reviews before a generated follow up reaches a client, or whether generated content can be told apart from staff written content inside the client record.

Nor is there a described approval step before an automated workflow action executes, or a way for a firm to audit what the suite did on its behalf. A client relationship system is the firm's book of record. Content that enters it automatically, with no described gate, becomes indistinguishable from the rest over time.

The claim most in need of measurement is that relationship intelligence surfaces what matters and organizes the information that counts. A summary that omits a material client instruction is wrong in a way nobody notices: the adviser reads a clean summary and never learns what it left out. Nothing published addresses completeness in either direction. No accuracy figure, error rate, benchmark or validation method is published. There is no grounding or citation guarantee linking a generated summary back to the conversation it came from, and no drift statement or revalidation cadence.

Two advisory firms are named as customers, but on a third party software directory, not in the vendor's own material. A directory entry is not the vendor's published claim, and its provenance cannot be established. The company's own material names no client and publishes no case study or quantified result from either the platform or the intelligence suite. The suite shipped in December 2025, so the absence of measured outcomes is expected, not evasive. The vendor's own customer stories material would likely settle the question and is not reflected here.

Nothing published says whether client records, meeting content or communications processed by the intelligence suite are used to train, tune or evaluate any model. Nor does it say whether a model provider retains content, or whether material from one advisory firm can inform output shown to another. The product runs natively on Salesforce, which complicates the question. Nothing published states whose terms govern content passing through the intelligence layer: the vendor's, Salesforce's, or both.

The platform holds the unified client record for advisory firms, spanning demographics, assets, goals and full engagement history. The intelligence suite now adds conversation content to it. No retention schedule, deletion commitment or data processing terms are published, and nothing describes how meeting content is handled once summarized. Salesforce, the platform underneath, carries its own extensive privacy documentation. Whether the vendor relies on that or publishes its own position is unclear.

A certification claim exists, and it has two problems. A third party software directory states that the company adheres to leading security standards, including a named international standard and a named government framework. The first problem is provenance: the claim sits on a directory, not on the vendor's own domain. The second is wording: "adheres to" and "compliance" are not the words a certified organization uses. Neither means the certifications are absent, only that nothing published establishes them. Whether the vendor's own domain carries a trust or security page is not reflected here.

A software supplier to advisory firms has no license and requires none. The platform performs compliance monitoring for its customers, which is a product capability, not any supervised status of the vendor, and the two are separate. The open question is the ordinary one for this category.

Records held in a client relationship system are examinable material for a regulated advisory firm, and nothing published describes how a firm would show an examiner which entries in that record were generated rather than written.

The exposure sits in prioritization, not in any explicit judgment about a person. A system that surfaces insights, decides which client context is worth raising and generates follow up actions is allocating adviser attention across a book. A systematic tendency in what it raises steers service toward some clients and away from others without anyone choosing that. Nothing published explains how the suite decides what to surface.

There is no account of whether its output was examined for systematic effects across client type, account size or engagement history, or of whether a firm can inspect why one client was flagged and another was not.

The risk comes from the autonomy the product advertises. A follow up generated and sent automatically, or a workflow action triggered without manual intervention, is an act the advisory firm performed as far as its client and its regulator are concerned. Nothing published addresses what happens when one of those actions is wrong. There is no public warranty or accuracy commitment, and no service level or correction obligation. Nothing allocates responsibility between vendor and advisory firm, and no disclaimer places it plainly on the firm.

The platform underneath is named and the models are not, and that gap creates real ambiguity, not a simple silence. Naming a platform or host discloses infrastructure, not a model supplier. Salesforce ships its own AI services, so a reader told that the product is native to Salesforce and now has AI capabilities cannot tell whose models these are. They could be Salesforce's, a third party's, or the vendor's own. Product pages, the launch announcement and the blog do not resolve it.

The architecture does most of the work. The platform is built natively on Salesforce Lightning, not as standalone software. It inherits that platform's integration tooling, application ecosystem, and enterprise identity and permission model without the vendor building any of it. Firms already running Salesforce elsewhere in the business gain a common foundation. An email client integration is named, with follow up recommendations surfacing inside it.

No wealth specific connection is named individually. Custodians, portfolio accounting, planning tools and document systems are the integrations an advisory firm evaluates on, and none of them appears in published material.

Because the product is built natively on Salesforce Lightning, its hosting and residency are largely inherited from that platform, which publishes its regional infrastructure. The vendor itself states none of it. Its own material names no hosting region, storage location, tenancy statement or residency commitment. That leaves a buyer inferring where data sits from the platform underneath rather than reading it.

One structural fact is published and no number is. The intelligence suite is activated separately, not included, so the AI carries its own commercial line. The newly announced second product is described as sold alongside the existing platform, not replacing it, which implies a further separate charge. No rate, per seat figure or tier is published. The stated customer range spans firms of ten employees to ten thousand, and the pricing model is the thing a small firm most needs and can least see.

Breadth comes from firm size, not institution type. The buyer is a wealth management or financial advice business and nothing else; no bank, insurer or asset manager segment is addressed. Within that, the stated range runs from roughly ten employees to enterprises of ten thousand. That spread calls for different configuration, permissioning and support models at each end. The company describes serving firms internationally, not in a single market. Two limits matter: the single buyer type, and the lack of published customer evidence behind the international claim.