SAVHN AI Features: Frequently Asked Questions

A lot of software claims AI features that turn out to be templates with variables swapped in — confident-sounding output with no model actually behind it, and no way to tell the difference. SAVHN draws a hard line on this: every AI-touching feature goes through one seam, and every response it produces is explicitly labeled as coming from a real connected model or from a plainly-marked placeholder. This FAQ explains how that works across AI Studio, AI Assistant, Manager Copilot, Business Brain, and the rule-based scoring that runs underneath several modules, and answers some of the practical questions that come up once you start using these features day to day.

The Honesty Pattern

What does "real model or plainly-labeled placeholder" actually mean?

Every feature that touches AI calls the same underlying abstraction. If your organization has connected a real AI provider, that call returns real, contextual output generated by the model. If no provider is connected, the exact same call returns a response explicitly marked isMock: true that echoes back the real context it would have reasoned over — it does not dress up an empty response as if it were genuine AI output. You can always tell which one you're looking at; it's not left ambiguous or buried in fine print.

Why does SAVHN bother with a placeholder instead of just hiding the feature?

Because showing you the real context the feature would have reasoned over — even without a live model — is more useful than a blank screen, and because pretending an unconfigured feature is "working" with fabricated output would be dishonest. The placeholder path exists specifically so nothing ever looks like AI-generated content when it isn't, which matters most in exactly the situations where you'd be tempted to act on AI output without double-checking it.

Do I need to configure my own AI provider to use SAVHN's AI features?

To get real generated output — drafted messages, model-reasoned recommendations — yes, a provider needs to be connected at the platform or per-organization level (Anthropic today, with more providers on the roadmap). Without one connected, AI-touching features fall back to the honest placeholder path rather than breaking outright, and rule-based features that don't require a model at all keep working regardless of whether a provider is connected.

What's the difference between "AI-powered" and "rule-based" in SAVHN?

Some capabilities genuinely require a connected language model to produce their output — drafting a follow-up email in a rep's voice using real deal context, for instance. Others are deterministic and rule-based by design, meaning they don't need any model at all to work, and every result comes with an explicit, human-readable list of the exact rules that produced it. SAVHN doesn't blur these two categories together — a rule-based score is described as rule-based, not dressed up as "AI" for marketing effect, even though both are genuinely useful, automated intelligence.

The AI Features Themselves

How does lead scoring work — is it AI or rules?

Lead qualification is rule-based and fully explainable, not a black-box model call. The engine runs a base score and adds points for real, checkable signals — company name provided, email provided, referral-channel source, deal value above a threshold, traceable campaign attribution — and every rule that fires is recorded as a plain-language reason, like "Came through referral (+20) — referred leads historically convert at a much higher rate." The lead then lands in a Hot, Warm, Cold, or Dormant bucket with a recommended next action. A sales manager can see exactly why one lead outranks another instead of trusting an unexplainable verdict, and can defend that ranking to their own team.

What does AI Studio actually let me do?

AI Studio is where AI-touching capabilities across the platform are configured — connecting a real model provider, and seeing which features will use it once connected. It's the administrative surface behind the honesty pattern described above, not a separate AI product with its own detached feature set disconnected from the rest of the platform.

What does the AI Assistant do?

The AI Assistant is the conversational, in-product AI surface for day-to-day work — drafting communications, answering questions grounded in your organization's real data. Like every other AI-touching feature, it follows the same real-model-or-honest-placeholder pattern: with a provider connected, its output is genuinely generated; without one, it says so rather than fabricating a confident-sounding response you might otherwise act on.

What does the AI Communication Writer draft, specifically?

It drafts things like a follow-up note, a win-back message, or a proposal note, using real deal context pulled from the CRM record it's attached to. If no model is connected, it explicitly refuses to send a placeholder as though it were genuinely generated copy — the honesty pattern applies here just as it does everywhere else, so you never send a client a message that only looks personalized.

How does Manager Copilot avoid false alarms — like flagging someone as idle when they're not?

Manager Copilot's idle-alert logic checks real signals before firing — for example, recent file exports or the current stage of a project someone's assigned to — so an employee who's actually waiting on a client response, rather than genuinely idle, doesn't get incorrectly flagged. The suppression logic is based on real signals present in the system, not guesswork, which matters because a manager tool that cries wolf stops being trusted quickly.

What is Business Brain?

Business Brain is SAVHN's cross-module intelligence layer — it draws on real data already flowing through CRM, Finance, Projects, HRMS, and other modules to surface recommendations and answer questions grounded in your organization's actual records, rather than operating as an isolated AI feature disconnected from what's actually happening in your business. Because it reads from the same modules your team is already using, its answers reflect the real, current state of your operations rather than a stale or separately-maintained data set.

Does the Adaptive Intelligence Layer require a connected AI model?

No — the Adaptive Intelligence Layer's cross-module recommendations (for example, warning Finance before a collections situation becomes a crisis) are rule-based and fire off real event data in real time, the same explainable-reasoning approach used in lead scoring. It works whether or not a model provider is connected, because it's not relying on a model to produce its recommendations in the first place — it's watching for the same kind of checkable conditions a rules engine is well suited to.

Can I tell, looking at the product, whether a given piece of AI output is real or a placeholder?

Yes — that's the entire point of the pattern. Every AI response carries that distinction explicitly (the isMock flag underlying it), so you're never left guessing whether what you're looking at was actually generated by a model or is a stand-in shown because nothing is connected yet. This matters most before you send a communication or act on a recommendation — you should never be uncertain about whether you're looking at real generated reasoning.

Does connecting my own AI provider mean SAVHN can see my conversations with it?

Data sent to a connected AI provider is used to generate the specific response the feature requested — a communication draft, a recommendation — using your organization's context needed for that task. Any specific question about how a given provider handles data you send it, retention, or training use is worth directing to that provider's own policies and to the team through /book-demo if you need clarity for a compliance review, rather than assuming a generic answer covers every provider identically.

Is there human review before AI-drafted content goes out?

AI-drafted content, like a communication draft from the AI Communication Writer, is presented to a real user for review before it's sent — it's a draft, not an automated dispatch. This keeps a human in the loop on anything client-facing, consistent with the platform's broader approach of not letting AI act unsupervised on judgment calls that affect a real relationship.

Will more AI providers be supported besides Anthropic?

Anthropic is the real, currently supported provider; additional providers are on the roadmap rather than available today. If provider choice matters to your evaluation, it's worth confirming current status directly rather than assuming general availability, since roadmap items are explicitly not the same as shipped features, and SAVHN's own honesty pattern would apply just as much to how it describes its own roadmap as to how a feature describes an unconnected model.

Does the AI honesty pattern apply to every module, or just the ones explicitly branded "AI"?

It applies to every call site that touches AI across the platform — lead qualification, communication drafting, Manager Copilot suppression logic, Business Brain recommendations, and anywhere else an AI-generated response could plausibly appear — because all of them route through the same underlying provider seam rather than each module implementing its own separate, potentially inconsistent approach to labeling AI output.

Can I turn off AI features entirely if my organization doesn't want to use them?

Since AI-touching capabilities depend on a provider being connected at all, an organization that never connects a provider is effectively running with those features on the honest-placeholder path by default. Beyond that, module-level enablement through the Marketplace governs whether AI-adjacent modules like AI Studio or AI Assistant are turned on for your organization in the first place.

Does using AI features cost extra beyond the module license?

Whether AI usage carries its own cost component beyond the module itself — for instance, related to the connected model provider's own usage-based pricing — depends on your provider arrangement and is worth clarifying directly through /book-demo or the developer team rather than assumed, since this is a genuine cost variable rather than something the platform can quote generically on your behalf.

How accurate is the rule-based lead scoring compared to a model-based approach?

Rule-based scoring trades the flexibility of a model's open-ended reasoning for full explainability — every point added to a score is traceable to a specific, named rule, which is a different value proposition than a model's holistic judgment call. Whether that trade-off suits your sales process is worth evaluating directly against real leads during your trial rather than assumed from a general description, since the right approach genuinely depends on how your team wants to work with the output.

Want to see a rule-based lead score you can actually defend, or an honestly-labeled AI draft rather than a black box? Start a 7-day free trial — no card required — and configure AI Studio alongside the modules you need through the Visual Pricing Configurator. For questions about provider connections or what's currently shipped versus roadmap, contact the developer team directly.