A review of the Claude you already run.
For small and mid-sized teams in New Zealand and Australia that are already on Claude Team or Enterprise, often already connected to NetSuite, Xero or Microsoft 365, and want to know what it costs, what is actually used, whether the set-up is sound, and what to build next.
The short answer
Claude optimisation is Sentry AI's review of a Claude Team or Enterprise set-up that a business already runs. It covers four things: what each seat uses and costs, which skills and MCP connections are actually used and by whom, whether the way Claude reads and writes your systems is sound, secure and able to grow, and which workflows to automate next. It ends in a written report the team can act on itself, and, where the team wants it, a day in your office building the next automations together.
What the review covers
Five areas, read from your own usage data and from the connections and skills as they are set up today. Nothing is estimated; where a figure cannot be read, the report says so.
Usage and cost per seat
Who is active, which models they use, what each seat costs, and whether the seat types fit the usage. Light users on premium seats and heavy users hitting limits both show up here.
Skills and connections in use
Which skills are loaded and by whom, which MCP connections are used and for what, and which were set up and never touched. A skill at zero is a finding, not a gap in the data.
Security and access
Per person and per department: who can reach which connection, what each may read and write, where a write is held for a named person to approve, and where it is not.
Reports and dashboards
The reports your team already prepares with Claude. Are the numbers right, which entity and period does each one read, and does it read posted or unposted figures?
What to build next
The jobs still done by hand, ranked by the time they take and by how close they are to a connection that already exists. Each with its week-one baseline, so the gain can be measured rather than claimed.
If one of your connections is to an ERP, the review reads it the way a finance lead would: which entity, which period, posted or unposted, and who can post. What a finance team can build on that connection is set out in Claude connected to NetSuite or your ERP through MCP, and how a connection works at all is on MCP explained.
What we need from you
Three things, and none of them is a project. The people who own the systems are in the room for the walkthrough and nowhere else.
Admin access to your Claude organisation
To read the usage and cost figures and the plan you are on. Your organisation, billing and data stay in your name throughout.
A walkthrough of each connection
With the person who owns that system, so we understand what it is for and what it is allowed to do, not just what it can do.
The three reports you rely on
The ones your team prepares with Claude and acts on. We trace each one back to what it actually reads.
What you get
Three documents your team keeps, written so the next step does not depend on us.
A written report
Plain words, one page per area, with what we found and what to do about it. Written so your own team can act on it without us.
A goals register with a week-one baseline
Each job still done by hand, timed as it runs today, with a unit and a frequency. The register is what the next measurement is read against.
A ranked build list
The next automations in order, each named with the connection it reads, the person who checks its output, and whether your team can build it or a developer is needed.
The build day
When to pair it with a training day
When the build list is longer than the time your team has, or when the people who should own the next automations have not built one yet. The training day runs in your office. Each person takes one job from the build list, builds the skill or automation on your live connections with us beside them, and leaves able to change it without a developer. Every job is timed before and after against the baseline in the register, which is how the day is judged.
A team that has not yet connected Claude to anything usually starts with the day and reviews later. A team that is already connected usually wants the review first, so the day is spent building on a set-up it trusts. How the wider rollout works is on Claude for business, and the plan controls that matter are compared on Claude Team vs Enterprise.
What we do not do
The usage data the review reads is a short allowlist: tokens, model, cost, which tool was used, which skills were loaded, and the workspace paths. Everything else is dropped before it is stored. We never capture:
- Prompt text
- Claude's answers
- Tool results
- The arguments a tool was called with
- File contents
- Reselling Claude seats or taking over your organisation. It stays in your name.
- Changing a connection's permissions. Those sit with your administrator; the report says what to change and why.
- Promising a saving. The register holds the baseline; the next measurement says what moved.
How we hold that line, and how Claude's use is governed on either plan, is on AI governance and security.
Who it is for
Small and mid-sized teams in New Zealand or Australia that are already on Claude Team or Enterprise. Usually the person asking is the finance lead, the operations lead or the owner, and the question is one of three: are we paying for the right seats, is the connection to our systems safe, or what should we build next. If you have not started on Claude yet, start with Claude for business instead.
Frequently asked questions
What does a Claude usage and cost review cover?
Four things: what each seat uses and costs, which skills and MCP connections are actually used and by whom, whether the way Claude reads and writes your systems is sound, secure and able to grow, and which workflows to automate next. It ends in a written report your team can act on itself.
What do you need from us to run it?
Admin access to your Claude organisation, a walkthrough of each connection with the person who owns that system, and the three reports or dashboards you rely on most.
What can you see in our usage data, and what can you not?
We see tokens, model, cost, which tool was used, which skills were loaded and the workspace paths. We never capture prompt text, Claude's answers, tool results or the arguments a tool was called with. That line is held by an allowlist before anything is stored.
Does the review include our connection to NetSuite, Xero or another ERP?
Yes. For each MCP connection we check what it may read and write, who can use it, which entity and period each report actually reads, whether writes are held for a named person, and what the connection costs in usage.
Do we need to be on Claude Team or Claude Enterprise?
Either. The review reads the plan you are on and says whether the seat types fit the usage. If a control you need is Enterprise only, the report says so; our Claude Team vs Enterprise page lists the current controls of each.
Can our own team do the building afterwards?
That is the aim. The report ranks what to build next in plain words, and the optional training day is spent in your office with your finance, admin and ops people building those automations themselves, so a small change no longer waits for a developer.
Already on Claude and connected?
A short call to hear what you have built and what you want from the review. If a training day is the right next step instead, we will say so.
Ask an AI about this page
Opens the assistant with this page loaded: read it, summarise it, cite it.