The Proactive Report Agent: Reports That Write...

Every year the same pattern. Someone realises on 20 October that the NGER report is due in eleven days. The sustainability lead opens the ledger, runs the totals, and the numbers don't reconcile with what the CFO briefed the audit committee three months ago. The last week of October gets written off. Not because the work is hard. Because nobody generated the draft early enough for the disagreements to surface with time to fix them.

The Proactive Report Agent exists to break that pattern. Not by taking humans out of the loop. By making sure the draft that a human eventually needs to read has already been generated, quietly, on a schedule, before anyone has to remember to ask for it.

The panic-late-week pattern most Australian reporters fall into

We can describe the pattern in three sentences.

The lodgement deadline is fixed. The data that goes into the report is not. So the report gets generated once, in a rush, in the final week, and any disagreement between the sustainability team's numbers and finance's numbers becomes a live crisis instead of a settled question.

You can see it in the reporting calendar of every Australian NGER filer. October 31 is the deadline. Nobody generates a draft in July. Draft generation happens somewhere between 15 and 25 October, which gives finance, the auditor, and the CFO between six and sixteen days to reconcile a document that took months of source data to assemble.

The reason this happens is not laziness. It's that generating a report is expensive when it's a manual exercise. Someone has to pull the ledger, sanity-check the totals, format the tables, write the narrative, chase the missing months, and produce the deliverable. Doing that once a year is painful enough. Doing it every month, as a background hygiene practice, is impossible if it costs three days of a senior analyst's time.

So teams don't do it. And the reconciliation debate happens in the last week of October, every October.

What automation actually changes

The Proactive Report Agent solves the timing problem by making draft generation cheap enough to run on a schedule. The reporter configures how often they want a draft to arrive. The platform generates it against the live ledger, packages the current picture into an email, and drops it into the sustainability lead's and CFO's inbox.

Nothing changes about the underlying data. The ledger is what it is. What changes is that the numbers get looked at ten times before lodgement instead of once. And the ten looks land in inboxes that already read reporting emails, at times the reporter chose, rather than requiring anyone to open the platform.

The behaviour we want to enable is small and boring. Read the draft on the tram home. Notice that the September Scope 2 total looks wrong. Open the ledger the next morning. Find the underlying invoice. Fix it four weeks before the deadline, not four days after it.

What the auto-generated draft covers

Five report types are wired in.

The weekly summary looks at the last seven days of ledger activity. It surfaces the totals, the record and document counts, and any open anomalies that appeared in the window. It's built for teams that want a rolling pulse rather than a compliance artefact.

The monthly summary does the same for a rolling 30-day window and adds a Scope 1, 2, and 3 breakdown with a mix-shift commentary against the prior month. This is the report we see teams standardise on for internal management review.

The anomaly digest pulls from the anomaly queue rather than the raw ledger. It lists open anomalies with recommended investigation priorities. On months where nothing is misfiring, it says so, and gets skimmed in ten seconds. On months where fifteen new anomalies appeared, it becomes the sustainability team's Monday agenda.

The target progress report runs against the Targets module. It calculates year-to-date emissions against the target baseline and pathway, using an Australian financial-year cut-off (1 July to 30 June). It's the report a CFO reads when they want to know whether the FY26 interim target is at risk.

The compliance readiness assessment is the anchor report. It sizes up the ledger against the NGER lodgement and AASB S2 disclosure requirements. Scope coverage. Temporal completeness. Documentation attachment rate. Anomalies that would compromise a report if lodged today. If it looks like the NGER report generation or the AASB S2 climate disclosure would fail an internal review, this report says so before the reporter opens the report generator.

Each draft is grounded in a data snapshot. Total emissions in tonnes CO2-e. Scope 1, 2, and 3 subtotals. Record count. Document count for the period. Open anomaly count. Those are the numbers that anchor the narrative. The AI commentary layered on top is bound to the snapshot, so if the snapshot is empty, the draft is empty. The agent does not invent numbers.

The schedule the reporter configures

The reporter chooses the cadence. There are four sensible defaults we recommend and one anti-pattern we discourage.

Monthly is the default cadence for the monthly summary and the anomaly digest. It matches the finance close and gives the sustainability lead one predictable email per month to read.

Quarterly matches the ASX-listed reporting rhythm and the audit committee cycle. This is where target progress reports usually sit for public companies.

Pre-lock-date is the useful one. If the period lock and restate workflow locks the Q1 ledger on 15 October, we schedule a compliance readiness draft to land 14 days before the lock. That gives the reporter two weeks to close any gaps the draft flags.

Pre-submission-date follows the same logic for the actual lodgement. Two weeks before 31 October, a compliance readiness draft lands. If nothing is red, the reporter proceeds to generate the final report. If something is red, they have time.

The anti-pattern is daily. We turned this off by default. A daily draft trains the reader to ignore the email, and the whole point of the scheduled cadence is that the email keeps being read.

What each draft contains

Every draft is built on the same skeleton.

At the top: the period, the total emissions figure in tonnes CO2-e, and the Scope 1, 2, 3 breakdown. All grounded in the live ledger via the same aggregation the Carbon Ledger uses, so what the reporter sees in the draft matches what they see in the dashboard.

Below: the current forecast for the full reporting year with a confidence band. This is the Emissions Forecasting output rendered inline, so a reader who opens the email on their phone can see the year-end projection without opening the platform.

Then the queues. Open anomaly count. Compliance gap count. Missing scope data. Any of these three going non-zero triggers a warning line in the draft. The reader knows immediately whether the draft is telling them everything is fine or telling them to look at a specific queue.

Then the methodology label distribution. What percentage of the total is calculated from measured data, what percentage from supplier-provided data, and what percentage from spend-based or estimated inputs. This is the number an auditor asks for on day one of a limited assurance engagement, and having it visible on a monthly cadence prevents surprise.

Then the AI-generated narrative. Two paragraphs on what moved. Two to four actionable recommendations. All bounded by the data in the snapshot.

The draft is factual first, opinionated second. If the data is thin, the narrative says so.

How the draft is delivered

We send the draft by email. Inline HTML with the summary tables and chart images embedded, so a phone reader gets the picture without downloading anything. A PDF attachment for the audit trail. A Word attachment for the sustainability team member who inevitably needs to paste sections into a board pack.

Recipients are configurable per report type. The compliance readiness assessment usually goes to the sustainability lead and the CFO. The monthly summary usually adds the head of operations and the COO. The anomaly digest goes to the person who actually resolves anomalies, which in most teams is not the same person the compliance report goes to.

The email itself carries the draft's status. Right now that status is either "generated" or "fallback" (the second means the AI narrative was unavailable and the draft is a data-only summary, which is still fine to read, just less prose). We're extending this to include "review requested" and "acknowledged" so the sender knows whether the report was actually opened.

Integration with the Executive Summary generator

The monthly and quarterly drafts include an Executive Summary block. That block is generated by the same engine that produces the board-pack version of the Executive Summary (chart set, headline metrics, top-three variance drivers, top-three upcoming risks).

For quarterly drafts specifically, this means the board pack material is essentially pre-drafted a week before the audit committee. The sustainability lead doesn't build the Executive Summary from scratch two days before the meeting. They open the email, read the draft, edit the narrative if needed, and hand it to the CFO. What used to be a 40-hour build becomes a 90-minute review.

We're honest that this only works if the underlying ledger is clean. Garbage in, garbage out. If the source documents aren't in the platform, or if half the ledger is spend-based estimates with no supplier data behind it, the Executive Summary will render but the narrative will be caveated. That's the correct behaviour. It's also the behaviour that most spreadsheet-based processes suppress, which is how boards end up briefed on numbers that don't survive audit.

Integration with the Compliance Gap Fixes agent

The compliance readiness draft cross-references the Compliance Gap Fixes agent output. Any gap the fixes agent has queued gets flagged inline with the specific records affected and the suggested remediation.

The point is that the reporter reading the draft sees the compliance state and the compliance fix in the same paragraph. They don't need to open the platform, navigate to the gap queue, review the suggested fix, and cross-reference which report it affects. The email carries all of that in one place.

For teams that trust the fixes agent enough to run it on Auto or Trusted mode, the flagged gaps are already resolved by the time the draft lands. For teams that keep it on Shadow or Co-pilot mode, the draft is the review surface where the reporter approves the fix and sends it back to the ledger.

MCP and Copilot integration

For teams that connect ChatGPT or Claude to the platform via MCP, the CFO can ask a natural-language question about the latest draft without opening the email. Prompts like "what did the last proactive draft show for Scope 3 in the September window" and "what are the top three anomalies in the last compliance readiness draft" return grounded answers pulled from the persisted report bundle, not from a hallucinated summary.

This matters more than it sounds like it does. CFOs read a lot of email. A CFO who can ask a question of the latest draft in the same window they use to draft the audit committee response is a CFO who actually acts on the draft. The delivery mechanism only counts if the reader engages with what was delivered.

What the agent does NOT do

Three important limits.

It does not auto-lock a period. Locking is a governance action that changes what auditors can rely on. The agent generates the draft that shows the reporter whether locking is safe. It does not perform the lock itself. That stays a two-click confirmation by a user with the Reporter or Admin role.

It does not auto-submit to the Clean Energy Regulator or lodge with ASIC. The compliance readiness assessment is a diagnostic, not a lodgement. When the draft says everything is green, the reporter still opens the formal report generator, reviews the assembled document, and lodges through the normal path with a human signature.

It does not modify the ledger. The agent reads the ledger, produces a summary, and stores the summary in a separate table. If the compliance readiness draft flags a gap, resolving that gap is a human action taken through the Compliance Gap Fixes queue. The proactive draft is a mirror, not a hand.

These are deliberate boundaries. Automated report generation is safe. Automated lodgement is not. The regulatory penalty for a bad NGER submission is real (up to two years imprisonment for dishonest behaviour under the NGER Act, and civil penalties calculated in penalty units currently valued at $330 each). Nobody is asking for an AI to accept that risk on their behalf.

What we're still working on

We'll be honest about what the current shape of this doesn't yet solve.

The first is business-unit filtering in scheduled drafts. Right now the reporter configures the schedule at organisation level. A construction group with 40 project sites wants each project manager to receive a draft scoped to their site. The agent supports the business-unit filter as an input parameter, but the scheduling UI treats the schedule as one-per-org. Extending that to one-per-BU is on the roadmap.

The second is Recipient acknowledgement. Sending an email is not the same as making sure it got read. We can tell you the draft was generated. We can't yet tell you the sustainability lead opened it. This matters because the value of the whole system depends on the read happening.

The third is variance narrative integration. The Variance Explanation Agent produces a ranked driver list for month-over-month movement. The proactive draft doesn't yet pull that ranked list inline. It surfaces the totals and flags material movement, but the "why" narrative is still a separate report. Combining them into a single monthly draft is next.

None of these gaps prevent the current shape from doing what it's built to do. They just mean the surface is going to keep expanding.

FAQ

What happens if the ledger is empty when the schedule fires?

The agent generates a data-only draft that notes the ledger is empty for the window. It doesn't fabricate numbers. If the compliance readiness draft finds no Scope 1 data at all, it says so and flags Scope 1 coverage as a compliance gap. This is different from the Scope 2 being small; it's Scope 1 being missing, which is a data-collection problem the reporter needs to know about.

Can we send drafts to external stakeholders like our auditor?

Yes, though we recommend the compliance readiness assessment for this and not the raw monthly summary. Auditors want the diagnostic view (scope coverage, methodology label distribution, anomaly state) more than they want the emission totals. The email delivery supports arbitrary recipient lists. What we don't yet do is watermark the draft as "unaudited internal draft". That's a policy the reporter enforces in the email body.

Does the agent work if we're on the SME tier?

Yes. Every tier gets the Proactive Report Agent. The report types available and the recipient list size scale with the tier. The core mechanic, scheduled generation, email delivery, grounded snapshot, is baseline platform behaviour, not a premium add-on. Pricing sits at $100/month plus a per-project component; specifics at hello@carbonly.ai.

Is the AI narrative something an auditor will accept?

The narrative is opinion. The numbers under the narrative are the ones an auditor evaluates. Both are stored on the draft, along with which model generated the narrative and how many tokens were used. If the auditor wants to disregard the narrative and audit the underlying snapshot, they can. If they want the narrative in the trail as evidence of what management believed at the time of drafting, that's there too.

Can we turn a specific report type off?

Yes, per organisation and per business unit. Teams that don't yet have targets configured usually turn off the target progress report until they have a baseline to measure against. Teams running a hundred document uploads a day usually turn off the weekly summary and rely on the monthly summary plus the anomaly digest. It's a settings screen, not a support ticket.

If you want to see what a proactive draft against your real ledger looks like, send us the last three months of source documents, however messy the pile is, and we'll set up the schedule against a live tenant. Email hello@carbonly.ai.

The Proactive Report Agent: Reports That Write Themselves and Land in Your Inbox