A view of the business built from the reports you already have
Ask which runs cost the most per delivery, which accounts are holding rental cylinders longest, or where debtor days are drifting, and get an answer traced to the report it came from. Signal builds that view from the exports and reports your team already produces in Swift, Easy Routes, Xero and HubSpot.
- Picks up new exports and reports from one agreed shared folder, such as Microsoft SharePoint, as they land
- Extracts each figure once and records where it came from
- Maps every figure to one agreed definition
One view you can question
Every measure against target across every depot, run and customer segment, on one set of definitions, with commentary, alerts and plain-English answers.No integrations to start. Signal runs on exports and reports your team already produces, so a working view is in front of your leadership team within four weeks. Direct connections to Swift, Easy Routes, Xero and HubSpot come later, and only where the business case holds.
From the exports you already pull to answers you can trust
Signal reads an agreed set of exports and reports, turns their figures into consistent measures and keeps the supporting documents linked. Your leadership team sees cited answers and alerts, with access rules applied throughout.
Start with your reports
Your team keeps saving the agreed Swift, Easy Routes, Xero and HubSpot exports and reports to one shared folder, such as Microsoft SharePoint. Signal picks up new and changed files from that scope.
Turn files into trusted figures
Signal tags documents by domain and sensitivity, extracts KPIs and maps each figure to an agreed definition. Source documents remain searchable evidence with access rules attached.
See the view and ask questions
The leadership dashboard shows consistent KPIs. Ask Signal gives cited answers, while each refresh can trigger alerts and explanations of what moved.
Keep control of every answer
Answers are logged, citations are checked and only people with the right permissions see sensitive material.
Forecasting, such as cylinder reordering by bottle type and area, and management pack preparation with human sign-off are options for a later stage, not part of the initial reports release.
Scope matters. Signal processes new or changed files in the agreed report set, not every company document. It is internal only: it never answers a customer, takes an order or dispatches a truck.
Try asking Signal
Click any of the four sample questions below to see how Signal answers, with figures traced to their source.
Answers and figures on this page are illustrative sample data, not drawn from SpeedGas or any other organisation.
What your leadership team sees
The dashboard opens instantly because commentary is prepared after each data refresh. Figures are extracted once and stored with their source, and every sentence of commentary carries a citation.
Every measure against target, by depot, run and customer segment, on definitions you agree once.
Flags measures moving off track using fixed, tested rules.
Plain-English commentary on what moved and why, with a citation on every statement.
Answers questions about the data it holds, without waiting on someone to pull a report.
A monthly management scorecard drawn from the same figures as the dashboard.
Every figure marked actual, calculated, target or not yet measured.
Where Signal saves money and where it makes money
Signal saves money by cutting the hours spent pulling exports, rebuilding spreadsheets and checking figures across Swift, Easy Routes and Xero. In our experience the bigger gain comes from seeing delivery cost, cylinder holdings and debtor days early enough to act on them.
Saves money
- Less rebuilding
- Exports from each system land on common definitions, so nobody rebuilds the same spreadsheet each month.
- Less checking
- Calculations run in tested code and every number links to its source, so reviewers check exceptions only.
- Faster month-end reporting
- Commentary and scorecards are drafted as soon as the data lands.
- Time back for customers
- Released time goes to customer conversations and decision support, not to another spreadsheet.
Makes money
- Delivery cost seen by run
- Cost per delivery and run length for every truck, so a heavy run and a light one beside it show up in the same week.
- Cylinders accounted for
- Rental cylinders tracked by account and age, so idle stock is recovered before more is bought.
- Cash in sooner
- Debtor days by segment and account, so slow payers are followed up while the amount is still small.
- Accounts kept
- Order patterns watched for every trade account, so a customer who goes quiet gets a call before they are lost.
of the time AI saves is lost to correcting its output.
Many AI tools produce answers quickly, but the output then has to be checked by hand. In Signal the AI model does no calculations. Figures come from tested code with their source attached, so far less needs checking.
Workday research, via Accounting Today
What AI changes. What Signal delivers.
These figures show where time and money can be lost, not what SpeedGas will save. For each one, here is the specific role of AI and the work Signal puts in front of your team.
Explains movements in delivery cost run by run, using sourced run sheets and vehicle costs, so the team can see where to look first.
Cost per delivery, drops per run and run length by depot and truck, with alerts linked to their source. Any improvement is measured from SpeedGas data, not inferred from the 41% figure.
Explains which segments and accounts sit behind a movement in debtor days, using sourced receivables reports, so follow-up starts sooner.
Debtor days and overdue balances by segment and account, refreshed with each Xero export. The follow-up call stays with your accounts team.
Answers leadership questions in plain English, reducing the need to recut the same figures for different audiences.
A governed view with cited answers and exception alerts. Monthly value reporting separates cash savings from released capacity, after checking effort and running costs.
Published external benchmarks. None is drawn from SpeedGas data or is a projection of SpeedGas results.
How we measure the return
We take a baseline before anything is built and publish the method before any result. Every case is stated net, over three years.
| Cost test | What it means for you |
|---|---|
| Cost to benefit | Running cost is a small fraction of the benefit, stated before build. |
| Predictability | A cost ceiling is agreed for each workflow at design. If AI model prices change, the cost forecast is updated with no rework to the solution. |
| Stoppability | Any component can be switched off without stranding cost. |
| Scaling | Cost does not grow faster than benefit as more reports and users come on. |
| Bankability | A benefit counts as a saving only where the freed time can be released. Otherwise it is reported as capacity. |
Signal reports its own running cost against the value it delivers each month, during the engagement and after it.
Reporting effort: try the numbers
Illustrative starting values, not SpeedGas data. Move the sliders to test the logic.
Illustrative inputs. The calculation follows the method above: hours released, times bankable share, less checking effort, less run cost. The fully loaded rate assumes about 28% on-costs (superannuation, payroll tax, leave and workers’ compensation) over base salary. Run cost starts at zero because it is agreed at design against a cost ceiling. This counts reporting effort only; delivery cost, cylinder and debtor gains are measured separately from SpeedGas data. Measured baselines replace these inputs at Gate 1.
Six guardrails, each enforced in code and tested
Consistent figures. Each measure renders the same way every time, so everyone sees the same value for the same metric.
Labelled commentary. Anything the AI writes is marked as generated.
Sourced numbers. Every figure is shown with the document it came from. Where a figure is missing, it is left blank.
The model does not calculate. Numbers are computed by tested code and the model explains them.
Access enforced in code. Results are filtered to each person's permissions before the model sees them.
Documents for narrative only. Figures are taken from tables and spreadsheets. Written documents are used for commentary.
Controls you can tighten for more sensitive work
AWS secures the infrastructure and the model service. The application controls around it are configured to suit each organisation, and Signal can be tightened for sensitive domains such as customer pricing, payroll or people data.
- Inference pinned to Australian regions
- Customer-managed keys in AWS KMS
- Private connectivity via AWS PrivateLink
- Bedrock Guardrails for sensitive data
- Hosting in an AWS account in your name
- Tighter clearance tiers and approvals
We work through these with whoever looks after your IT in weeks 1 to 4, then document the agreed settings and who owns each control before sensitive data is connected.
A staged roll-out, starting with your reports
The architecture is staged. Each step switches on more of what is already built, and each is a separate decision with its own business case.
Reports
Two or three domains built from existing exports and reports. Dashboard, alerts, commentary and questions live for the leadership team.
Connections
Read-only links to Swift, Easy Routes, Xero and HubSpot where they remove manual effort or add daily visibility.
Forward view
Forecasting, such as cylinder reordering by bottle type and area, what-if scenarios and management pack preparation with a human approval step.
Already running for one of Victoria’s most recognisable brands
Bosley designed and built Signal for the executive team of a Victorian not-for-profit with a complex mix of service, fundraising and operational reporting.
It runs on the organisation's own board-approved KPI framework, and question and answer capability shipped in the first release. The organisation has since funded the next phase, taking Signal into day-to-day use by its executive team.
Source: Bosley AI engagement records, June to September 2026. Client not named.
A 12-week path to a working release
Indicative plan, based on the not-for-profit engagement. Final timing depends on data readiness.
Every engagement runs on the Bosley Agent Delivery Lifecycle, with test sets and a quality scorecard written before testing begins. Spend is gated. At each gate you decide whether to continue, and nothing further is payable if you stop.
About Bosley AI
Bosley AI designs, builds and runs AI agents for Australian organisations, and every agent we take on has to either save money or make money. We work from the value case through to production, and we can run what we build after go-live, with monthly reporting on cost and return.
Getting started
- A working session with your leadership team to agree the questions Signal should answer first.
- A short review of the exports and reports behind those measures, and a baseline of the effort they take today.
- A first release on your existing reports, with a working view in front of your leadership team within four weeks.
The questions worth asking before you start, and how the approach answers them
The staged approach is built around these questions. Each one is answered by how Signal starts, not by a promise about how it ends.
We promise customers a person, not a machine.
Signal is built to protect that promise. It is not customer-facing. It never answers the hotline, replies to a customer or changes an order. It reads internal reports and answers questions from your own team.
The time it gives back is time for the conversations that need a person: the regular driver who knows the site, the call to an account that has gone quiet, the follow-up on an overdue invoice.
We already have Swift and routing software.
Signal replaces neither, and it does not dispatch a truck. Swift and Easy Routes keep doing what they do. Signal reads their exports, alongside Xero and HubSpot, and shows how the whole business is performing on one set of definitions.
It is not a software purchase either. Bosley designs, builds and runs it on AWS in Australia, so there is nothing to select, license or roll out. Read-only connections to your systems come at stage 2, as a separate decision with its own business case.
Our data isn't good enough.
Signal does not need clean, connected data to start. Stage 1 works from the reports you already produce and already run the business on. Every figure is shown with its source document, and where a figure is missing or inconsistent it is left blank rather than estimated.
That makes the first release a working map of where the data is weak. The data readiness check in weeks 1 to 4 turns those gaps into a prioritised fix list, so data quality improves through use instead of a clean-up project that has to finish before anything ships.
How do we know it is paying its way?
The effort behind today's reporting is baselined before the build starts, so the return is measured against a number agreed up front, not claimed afterwards. Spend is gated at each stage of delivery: you decide whether to continue at each gate, and nothing further is payable if you stop.
After go-live, monthly reporting shows cost and return side by side. The same baseline gives any later agent, such as the dispatch prototype we walked through with your team, a measured starting point to be judged against.