Drag
loader

Contact info

Email
hello@ferressystems.com.au
Based in
Melbourne, Victoria

Follow us

Talk to the team
Talk to the team
All services // AI AND WORKFLOW AUTOMATION

AI that does a real job, with the guardrails on

Half your team is already pasting company information into a chatbot, and nobody has decided whether that is allowed.

Purpose-built agents for processes you actually run, Microsoft Copilot deployed where it genuinely fits, a security review of the AI your staff are already using, process automation and integration, and the governance to go with all of it.

Scope this with us
 Free: AI acceptable use and data handling policy
// DEMONSTRATION

A shared intake inbox, sorted before anyone logs in

A demonstration drawn from the intake agent written up in our field notes, which is a composite with identifying details changed. Every step and figure below matches that write-up.

  1. 01 What arrives

    About 120 messages a day

    New enquiries, client correspondence, supplier invoices, recruiters and newsletters, all in one shared inbox a senior person used to sort for two hours every morning.

  2. 02 What the automation does

    Classifies, tags and routes

    Each message is sorted into one of seven categories with a confidence score and routed to the right person. It reads the message and its metadata, and nothing leaves the firm's own tenant.

  3. 03 What a person checks

    Anything it is unsure of

    Low-confidence messages go to a human review queue instead of being guessed at, and new client enquiries always get a human check. It cannot send anything, and it does not open attachments from unknown senders.

  4. 04 What gets saved

    A reason for every decision

    Every classification is logged with its reasoning and kept for ninety days, so when someone asks why a message went where it went, there is an answer.

  5. 05 How success is measured

    Ten hours a week down to one

    Agreed before the build: time spent on triage, and new enquiries not answered within a day. Triage fell to about an hour of review, and missed same-day responses went to zero.

// SOUND FAMILIAR

If two of these sound familiar, we should talk

The rest of this page covers what the work includes, what it produces and how it is priced.

01 Staff are already using AI tools you have not approved and cannot see
02 You bought Copilot licences and almost nobody uses them
03 A repetitive process eats hours a week and everyone has stopped noticing
04 Somebody has asked what your AI policy is and there is not one
05 You want to use AI but cannot answer where the data goes, so it keeps not happening
// THE SCOPE, BOTH SIDES OF IT

What this is, and what it deliberately is not

What this is

  • Agents scoped to one job you can name, built to do that job properly
  • Microsoft Copilot deployment and adoption where Copilot is genuinely the right tool
  • Purpose-built agents where it is not, integrated with the systems you already run
  • Access controls and data boundaries decided before deployment, not retrofitted
  • An AI security review of what your business is already doing, sanctioned or otherwise
  • A written acceptable use position your staff can actually follow

What this is not

  • A licence resale with a training deck attached
  • An agent sending data anywhere you have not approved
  • A pilot with no agreed measure of success
// THE ENGAGEMENT, START TO FINISH

What actually happens after you say yes

Typically three to six weeks for a first agent

A Copilot rollout or a review of what your staff are already doing is shorter than that. We would rather deploy one thing that earns its keep than six that get switched off in a month, so a first engagement is deliberately narrow whichever of these it is.

01

Find the job

We sit with the people doing the work and look for the process that is repetitive, rule-based, high volume and low judgement. That is where AI actually pays. On a Copilot rollout this is the same question asked of teams rather than of one process, and on a review it becomes finding what people are already doing unsupervised. Anything needing real judgement stays with a human, and we will say so.

02

The data question

Before any build or rollout: what information does this need, where does that information live, where would it be sent, and who is allowed to see the output. If we cannot answer all four, we do not proceed. These questions are answered again for every new agent, because a different job reaches different information and needs its own permissions.

03

Guardrails first

Access boundaries, retention, logging and the human checkpoints get configured before anything does useful work. It is much harder to add these to something already in production and much easier to justify to an auditor. This is also where an acceptable use position gets written down, if you do not have one.

04

Build or configure, then test

A purpose-built agent gets built against real examples from your business. A Copilot deployment gets configured and its access scoped. Either way it is tested on cases we know the answer to, including the awkward ones, and you see it working on your own data before anyone depends on it.

05

Supervised rollout

It goes live with a human reviewing the output, so mistakes are caught by your team rather than by your client. Supervision relaxes as the evidence justifies it, not on a schedule.

06

Measure and decide

We agreed at the start what success looks like. Now we check. If it did not deliver, you are told, and we either fix it or stop before anything else is built.

// WHAT YOU WALK AWAY WITH

Things you can hold, forward and act on

The artefacts this work produces, named so you know what you are buying.

Written document

Process assessment

Which of your processes suit automation, which do not, and honestly why

Written document

Data flow and access map

One per agent: what it can reach, what it cannot, and where information travels

The thing itself

The deployed agent

Configured, integrated, logged and running against your real work

Written document

Governance position

Acceptable use, retention and review, ready to hand to staff or an auditor

Live session

Outcome review

Did it do what we said it would, measured against what we agreed up front

// NOT A MOCK-UP

The clause that does the real work

A page straight out of the AI acceptable use and data handling policy, shown as it is. Read the page, then decide whether the whole document is worth an email address.

AI Acceptable Use Policy SECTION 03 // INFORMATION TIERS
ORGANISATION Your business name goes here

Information is sorted into four categories. The category decides which tools it may be entered into, and it applies to anything you type, paste, upload or connect.

NEVER CRITICAL

Credentials and access material of any kind

No AI tool, ever. Section 5 approval does not create an exception to this row. If any of it is entered by mistake, treat the credential as compromised and rotate it immediately. Deletion by the vendor is not a substitute for rotation.

  • Passwords, passphrases and PINs
  • API keys, access tokens, session tokens, refresh tokens and connection strings
  • Private cryptographic keys, certificates with private keys, and signing keys
  • Multi-factor recovery codes and break-glass account credentials
TIER 3 / RESTRICTED, NOT PROHIBITED

Regulated and sensitive information, including personal, health, financial and legal records. Permitted only where every one of the nine conditions in section 5 is met, documented and approved. De-identification is not a loophole: removing a name from a clinical note frequently leaves enough context to re-identify the person.

TIER 2 / APPROVED TOOLS ONLY

Internal business information, non-identifying operational data and draft work products. Approved tools only, on an organisation account. Not a personal account of any tool, including the free tier of an otherwise approved product.

Editable policy template · 18 pages Read the rest of this page
Send me the template
// WHAT IT COSTS

Fixed price for the first agent, then per agent

The first engagement carries the groundwork that is done once: the acceptable use position, logging and retention arrangements, and the review process. So it usually costs more than the ones after it. Each new agent still gets its own data flow and permissions assessment, priced into that agent, because what one agent is allowed to reach says nothing about the next. Licences, where you need them, are yours directly rather than resold through us.

Get a written scope
WHAT MOVES THE NUMBER
How many processes you want automated, and how different they are from each other
Whether the agent has to integrate with systems that have a usable interface
Whether you already have a governance position or are starting from nothing
How much change support your team needs to actually adopt it
// FROM THE FILE

One of these, in the field

DECLASSIFIED

The agent that reads the intake inbox, and the two things it is not allowed to do

An agent that triages a shared intake inbox and saves about nine hours a week. The interesting part is not what it does, but the two capabilities we deliberately did not give it, and the week spent on data boundaries before anything was built.

Read the case file
// QUESTIONS, ANSWERED

Asked before signing anything

No. A good share of these engagements build nothing at all. Scoping and governance before a rollout, or an answer about whether a process is worth automating, are real outcomes. If the right answer is that you should not deploy AI at this yet, you will hear that, and it is a cheaper answer to get early.

No, and it would be a poor answer to every problem if we did. Copilot is one tool and it is genuinely good at some things, particularly if you already live in Microsoft 365. Plenty of the work we do is purpose-built agents integrated with the systems a business actually runs, and often the right answer is both, with each doing the part it is suited to.

That is decided before anything is built, written down, and shown to you. Depending on the design, data may never leave your tenant at all. Where a third-party model is involved we tell you which one, what it receives, what it retains, and you approve that before deployment.

Then the honest first step is finding out what your staff are already doing with AI tools, because in most businesses the answer is more than management expects. An AI security review plus a written acceptable use position is a legitimate place to stop.

Start with the policy template

We agree the measure before we build: hours returned, error rate, turnaround time, whatever genuinely matters for that process. At the end of the engagement we check against it and report the result, including any category that underperformed.

// OFTEN SCOPED TOGETHER

What usually comes with this

Custom Software and Integrations The tool nobody sells, because it is yours Custom Software and Integrations, scope and pricing
Cybersecurity Governance and Compliance Evidence is a folder, not a scramble Cybersecurity Governance and Compliance, scope and pricing

Tell us the problem, not the solution.

Get in touch
Shapes Shapes