Predictiv AI
Predictiv AI is a single assistant embedded in your workspace. It answers questions about how the platform works, queries your data with full role enforcement, and creates or updates records with multi-turn confirmation — all in the same widget, with the same context.
We took a deliberately different path from the agent-fleet approach the larger ERP vendors are racing toward. One orchestrator routes between three modes. A small, disciplined set of tools — not forty — keeps the assistant's choices accurate and auditable. Every action runs through the same governance the rest of your ERP uses.
**Note:** Predictiv AI is provider-agnostic. The chat experience runs on the model that fits the job — switching providers is a configuration change, not a rebuild. See **[Open by Standard](/platform/capabilities/ai-openness/)** for the model-choice story.
Challenges We Solve
Common predictiv ai challenges—and how Predictiv addresses them.
Hallucinations and unbacked claims
Most AI assistants confidently invent data. In an ERP, a wrong currency, wrong period, or wrong account is a real-money problem.
Predictiv AI is grounded in your catalogues and your data. Help-mode answers cite the source document. Query-mode answers are bound to the schema and validated. The assistant says "I don't know" when it shouldn't guess.
Loss of control
"Autonomous" agents that read, write, and decide without oversight create audit and compliance exposure.
Three controls, by construction. Every tool call is logged. Every action is bound to the user's role. Document writes use a multi-turn confirm pattern — the assistant proposes, you preview, edit if needed, and confirm. You choose where the assistant acts autonomously and where it must defer to a human.
Vendor lock-in
AI assistants tied to one model provider become a strategic risk when the frontier moves or pricing changes.
An open standard for the tool layer; provider-agnostic at the model layer. Bring the model that fits the task; change it when something better ships. Your prompts, embeddings, and tool definitions stay where they are.
Black-box decisions
When an AI makes a recommendation or change, who can audit how that decision was reached?
Every tool call writes an immutable audit row — tool, parameters, target entity, user, timestamp. The audit trail is queryable like any other Predictiv record. Sample traces are available under NDA.
Three modes, one widget
Help mode
Ask how Predictiv works. Help mode answers from your wiki and configuration documentation, citing the source for each answer. Workspace-aware — the assistant knows what window you are on.
- Wiki and policy search with cited responses
- Click-through to source documents
- Image and PDF input for document questions
- Workspace-context injected automatically (current window, record, role)
Query mode
Ask about your data. Query mode runs through a small set of governed tools that resolve entities through your catalogues, enforce your role permissions, and write every call to the audit log.
- Customer 360, Cash Radar, Morning Briefing, period comparisons
- Generic entity queries with role-bound filtering
- All responses traceable to the database operation that produced them
- Read-only by default; write operations live in Action mode
Action mode
Draft and confirm changes. Action mode uses a multi-turn pattern — propose, preview, edit, confirm. Nothing mutates without an explicit human step on writes, unless you have explicitly enabled autonomy for that workflow.
- Approve from anywhere — documents, requisitions, payments
- Document drop-in — PDF or image to extracted record, with review
- Case management with optional human-in-the-loop
- Compliance obligation actions (mark filed, mark paid, defer, waive)
Workspace integration
The assistant lives where your team works. Pin it to the side of any window; maximise when you need depth; collapse to a bubble when you don't.
- Three states — bubble, sidebar, maximised
- Width and pinned-state persist per user
- Single sign-on with the rest of your workspace
- Mobile-friendly across the same authentication
Comprehensive Capabilities
Full-featured predictiv ai tools built into the platform.
Trust controls
- Every tool call written to an immutable audit log — tool, parameters, target entity, user, timestamp
- Role enforcement bound to the same JWT that authenticates your users
- Multi-turn confirm pattern on writes — propose, preview, confirm
- Per-user rate limiting with configurable thresholds
- Configurable autonomy thresholds — choose where the assistant acts and where it defers to a human
Open foundations
- Open standard for the tool layer — any compatible AI client can drive Predictiv
- Provider-agnostic model layer — OpenAI, Anthropic, AWS Bedrock, Azure, Google, or self-hosted
- Tenant-resident orchestration — chat content does not leave your environment unless you choose a hosted model
- Versioned, auditable prompts in a registry — business owners edit prompts without a developer cycle
- Same data-access path as the rest of the ERP — no separate AI knowledge graph to drift from your data
Built-in capability
- Financial intelligence (period comparison, account analysis, KPIs)
- Customer 360 (single view across orders, invoices, payments)
- Cash radar (current position, near-term forecast)
- Approve from anywhere (any document type, any role)
- Document intelligence (extract from PDF or image, multi-step verify)
- Generic entity operations governed by your role
Use Cases
How businesses use Predictiv AI to drive results.
Document drop-in
A finance user drops a supplier invoice PDF onto the chat widget. The assistant classifies the document, extracts vendor, amounts, line items and tax, validates against the purchase order and supplier records, and presents a draft. The user reviews, corrects any field the assistant got wrong, and confirms. The booked invoice is auditable end-to-end — including the AI extractions and the human edits.
Morning briefing
A controller asks "what should I know this morning?". The assistant returns overdue receivables over a configurable threshold, aging shifts, cash position, and exceptions in the last twenty-four hours, with click-through to source records. No dashboard needed; the assistant fetches what is relevant to the role.
Approve from anywhere
A regional manager receives a notification that a purchase requisition is awaiting approval. They open the assistant on mobile, ask "what's pending for me?", review the requisition context, and approve in chat. The approval is recorded against their role; the audit row shows AI-mediated approval with the conversation reference.
Policy questions
A new joiner asks "how does our purchase approval routing work?". Help mode answers from the policy wiki with citations. If the policy is ambiguous, the assistant says so rather than guessing.
Why Predictiv for Predictiv AI
One assistant, three modes — not forty agents
Most ERP vendors are racing to a multi-agent fleet — one of the largest is now publishing forty-plus named agents. Router accuracy degrades sharply past about fifteen tools. Predictiv took the opposite design: one orchestrator, three modes, six to fifteen disciplined tools. The assistant gets the right answer more often because it has fewer wrong choices to make.
Audited and editable by construction
Every tool call writes an audit row. Every prompt lives in a versioned registry that business owners can edit. Every write is role-bound and confirmable. These are not features bolted on for compliance — they are how the assistant is built.
Open by standard
Predictiv AI pairs three independent layers — a tenant-resident orchestrator, a model-provider-agnostic LLM layer, and an open-standard tool interface. Switch any of the three without rebuilding the others. Bring the AI you want; we provide the trustworthy layer that drives your data.
Same code path as your team
The operations the assistant uses are the same the ERP's UI runs through. Same role enforcement, same business rules, same audit. There is no second source of truth to maintain or drift. When your business rules change, the assistant follows.
Is This Right for You?
Predictiv Predictiv AI works best for organisations with these characteristics.
Company profile
- Mid-market or enterprise running ERP across finance, operations, or both
- A documented audit, governance or compliance requirement
- Multiple model provider preferences across the organisation, or expectation of provider change in the next 24 months
- Data residency or sovereignty requirements
Pain points
- AI assistants that hallucinate or work outside the user's role
- Need to swap or hedge AI providers but locked to a vendor's stack
- Compliance pressure to demonstrate audit trails for AI actions
- AI prompts living in source code instead of a registry the business can edit
- Document-heavy workflows (AP, claims, compliance, KYC) consuming staff time
Scale
- 50+ users on the ERP
- 100+ documents per month flowing through any one AI-eligible workflow
- Existing role-and-organisation structure with at least three role tiers
Ready to Improve Your Predictiv AI?
See how Predictiv can transform your predictiv ai capabilities. Book a personalised demo.