AI Copilots

A copilot trained on your business, sitting where your team works.

Custom AI assistants for Australian businesses. Copilot-style helpers that draft, summarise, search, and decide alongside your team — built on your data, integrated with Slack, Teams, Outlook, and your CRM, deployed with citation-grounded responses you can trust.

2.3×

Value of embedded vs standalone AI

100%

Cited responses

8 wk

Production rollout

Per-seat licence fees

Why a custom assistant beats a generic chatbot

Most teams trying to get value from AI today share the same friction: ChatGPT does not know about their customers, Microsoft Copilot does not know about their custom systems, and the public chatbots forget everything between conversations. The result is staff who copy-paste customer history into a prompt every morning, then copy the response back out. The pattern works — barely — but the friction destroys most of the time saved.

A custom business AI assistant closes the loop. It already has the customer history, it already has the house-style, it already has the project notes. The user types two sentences and gets a draft they can edit and send. McKinsey's 2024 State of AI reportmeasured the gap directly: "copilots embedded in existing workflows deliver 2.3× the value of standalone deployments" (McKinsey & Company, The State of AI in Early 2024).

Where we deploy assistants most often

Sales enablement assistant. Sits in the CRM and Slack. Pre-loads customer history, drafts proposals from a single brief, drafts follow-up emails personalised to the deal stage, summarises every call recording.
Internal knowledge assistant. Answers staff questions about policy, process, product, and price across every internal document — Drive, SharePoint, Notion, Confluence, the wiki — with inline citations.
Account-management assistant. Pre-reads the customer's recent tickets, contracts, and usage data; drafts the agenda for the next QBR; flags renewal risks before the call.
Project-management assistant. Summarises Slack channels, project boards, and meeting notes into a daily standup; drafts status updates for stakeholders; flags blockers before they hit the deadline.
Recruiting assistant. Reviews CVs against the role brief, drafts interview question packs, drafts personalised outreach to candidates, summarises interview feedback into hire/no-hire recommendations.
Finance and ops assistant. Summarises monthly board packs, surfaces variance against budget in plain English, drafts the commentary section of the management report.

The non-negotiables of every assistant we ship

Citation-grounded responses

Every answer includes inline citations to the source documents. Users can click to verify. No citation, no answer — the assistant refuses rather than fabricates.

Refusal-on-uncertainty

When the retrieved context does not contain a clear answer, the assistant says so explicitly and offers to escalate. This single behaviour is the largest driver of user trust.

Per-user permissions

The assistant can only access documents the asking user already has permission to see in the source system. Permissions are checked at query time, not just at ingestion.

Australian or US data residency

All LLM calls routed through region-locked endpoints. Vector database hosted in your tenant or in a region you choose. Source documents never leave your control.

Surfaces where your team works

Slack, Teams, Outlook, your CRM — wherever the work happens. Same backend, consistent context across surfaces.

Full instrumentation

Every query, every retrieved chunk, every response, every thumbs-up / thumbs-down logged for accuracy monitoring and weekly prompt refinement.

Engagement timeline

Weeks 1–3 — Data pipeline. Connect to source systems (Drive, SharePoint, Notion, CRM, helpdesk). Embed and index. Set up incremental sync. Verify retrieval quality on a test set before any LLM is wired in.
Weeks 4–6 — Assistant build. Construct prompts, response logic, refusal behaviour, tool-use for any actions. Build the Slack/Teams/Outlook surface integrations. Wire up SSO and per-user permissions.
Weeks 7–8 — Pilot. Roll to 5–10 power users. Tight feedback loop with the build team. Refine prompts daily based on real questions and real corrections.
Weeks 9–10 — Full rollout. All-staff launch. Weekly accuracy review. Add the next connector and the next capability based on actual usage.
Month 3+ — Hand-off or managed services. Documentation, runbooks, and monthly review cadence. We exit if you want; we stay on a managed retainer if you prefer.

Pricing

Single-team assistant (one surface, focused use case)$9,000 — $18,000 AUD

Whole-business assistant (multi-surface, multi-use-case)$25,000 — $80,000 AUD

Ongoing infrastructure (vector DB + LLM API + hosting)$200 — $1,500 AUD/mo

Managed services (monitoring + tuning + new connectors)$2,000 AUD/mo

All prices ex GST. No per-seat licence fees. Your staff uses the assistant as much as they want; you pay only for the underlying LLM tokens consumed.

Who this is for

Custom AI assistants deliver the strongest ROI for businesses with (a) a meaningful body of internal knowledge that staff currently search manually, (b) a customer-facing or staff-facing function where draft-and-edit is faster than write-from-scratch, and (c) at least 20 active staff who would use the assistant weekly. Typical fit: 50–500 person Australian businesses in professional services, financial services, healthcare, technology, and B2B services.

Poor fit: businesses with very thin internal documentation (the assistant has nothing to ground on), or single-person operations where Microsoft Copilot or ChatGPT Enterprise is already enough.

Frequently Asked Questions

What is an AI assistant for business, and how is it different from ChatGPT?

A general-purpose chatbot like ChatGPT, Claude, or Gemini is a single product that knows everything on the public internet but nothing about your business. A business AI assistant is a custom-built copilot that has access to your CRM, your documents, your project history, and your house-style — and is deployed inside the tools your team already uses (Slack, Microsoft Teams, Outlook, your CRM). According to McKinsey's 2024 State of AI report, "the highest-impact uses of generative AI are not standalone chatbots but copilots embedded in existing workflows, which deliver 2.3× the value of standalone deployments". A business AI assistant is the embedded-copilot pattern, built specifically for your context.

What can a custom AI assistant actually do for my business?

The strongest use cases we ship: (1) drafting customer responses pre-loaded with the customer's history; (2) summarising internal documents (quarterly board packs, contract bundles, RFP responses) in seconds; (3) drafting proposals and quotes from a single brief; (4) searching across all your knowledge sources at once (Google Drive, SharePoint, Notion, the CRM, the helpdesk); (5) acting as a tireless internal helpdesk for staff onboarding ("how do I file expenses?", "what's the leave policy?"); (6) translating data into plain English ("what changed in the Western region last week?"). The pattern is the same: the assistant pre-loads context, drafts the work, the human approves and edits.

How is a business AI assistant different from full AI employee replacement?

A business AI assistant is a copilot — it augments humans, drafts work for human approval, and stays inside the human's workflow. A full AI employee (which we also build, see our /services/ai-employees page) replaces a complete role end-to-end and operates autonomously inside defined boundaries. Most Australian mid-market businesses get faster, less risky ROI starting with assistants and graduating to full employees only for narrow, well-bounded roles. The assistant approach lets your team build trust in the AI gradually and keeps humans in control of every customer-facing or financially material decision.

How do you connect the assistant to our internal data?

We use Retrieval-Augmented Generation (RAG) with a vector database (typically pgvector inside your existing PostgreSQL or a managed Supabase instance). Documents from Google Drive, SharePoint, Notion, Confluence, your CRM, and your helpdesk are ingested on a schedule, embedded, and indexed. When the assistant answers a question, it pulls only the relevant chunks (with citations back to the source document) and feeds them into the LLM. Your raw data never leaves your tenant — only the relevant chunks for a specific query are sent to the LLM, and we route through Australian or US data-residency endpoints depending on your sovereignty requirements.

How do you stop the assistant making things up (hallucinating)?

Three layers. First, every answer the assistant generates is grounded in retrieved source documents and includes inline citations the user can click to verify. Second, we configure the assistant to refuse to answer when the retrieved context does not contain a clear answer ("I do not have enough information about this in your knowledge base — would you like me to escalate to your team?"). Third, we instrument the assistant to log every question, every retrieved chunk, every generated answer, and every user reaction (thumbs-up / thumbs-down / edit) so we can monitor accuracy weekly and refine prompts when patterns emerge. The result is a system where users trust the assistant because they can verify it.

How does the assistant integrate with the tools my team already uses?

We deploy the assistant where your team already works: a Slack bot, a Microsoft Teams app, an Outlook add-in, a sidebar in your CRM, or a private web app — usually a combination. The same backend powers all surfaces, so a question asked in Slack remembers the context if it follows up via Outlook. Single sign-on integrates with Microsoft Entra ID (formerly Azure AD), Google Workspace, or Okta — every assistant interaction is tied to the actual user identity, with role-based permissions controlling what data each user can ask about.

How long does it take to build and deploy a custom AI assistant?

Eight to twelve weeks for a production-ready assistant covering one or two business functions (e.g. sales-enablement plus internal knowledge). Three weeks for the data pipeline (connectors, ingestion, embeddings, retrieval), three weeks for the assistant logic and integrations (Slack, Teams, CRM), two weeks for staff onboarding and prompt refinement, and a final two weeks of monitored production with weekly tuning. Faster timelines are possible for narrower scopes (a single Slack-only assistant for one team can ship in three weeks).

What does a custom AI assistant cost in Australia?

A focused single-team assistant typically costs $9,000 — $18,000 AUD ex GST as a one-off build. A whole-business assistant covering multiple departments and surfaces typically ranges from $25,000 to $80,000 AUD. Ongoing infrastructure (vector database, LLM API, hosting) sits at $200 — $1,500 AUD per month depending on data volume and query rate. Optional managed-services tier at $2,000 AUD per month for monitoring, prompt refinement, new connector additions, and quarterly retraining. No per-seat licence fees — your team uses the assistant as much as they want.

Why hire Iverel rather than buy an off-the-shelf assistant like Microsoft Copilot or ChatGPT Enterprise?

Off-the-shelf assistants (Microsoft 365 Copilot, ChatGPT Enterprise, Glean) are excellent for generic knowledge-worker productivity. They struggle when you need (a) deep integration with your specific business systems beyond the major SaaS platforms, (b) control over the prompts, the routing logic, and the response style, (c) per-user data permissions that match your existing org chart, or (d) the ability to extend the assistant with custom tools (e.g. "draft a quote in our format using current pricing"). We build assistants that solve all four — and you own the codebase, the prompts, and the data pipeline outright. No annual per-seat fee escalation.

See what a custom assistant could do for your team

Book a free 30-minute scoping call. Pick the two highest-friction tasks your team currently does manually. We'll tell you how an assistant would handle them, what it would draw on, what would stay human-only, and what it would cost — written down, in numbers, before you commit to anything.

Book a Free Scoping Call →