Marcus Webb is the CFO of Fenwick Route, a Series C logistics-software company doing $22M ARR with a seven-person board that meets every quarter. Ten days before each meeting, Marcus and one FP&A analyst rebuild the board deck from a blank template, because there is no other version of it anywhere. Slide four this quarter shows a pipeline coverage ratio of 3.1x, pulled from the VP of Sales's CRM export, where pipeline means every opportunity past stage two. Slide nine, built by Marcus from the finance spreadsheet, shows 1.8x, because finance nets out anything not realistically closing this quarter. Two slides in the same deck disagree about the same word. Worse, a board member asks why net revenue retention dropped from 112% to 104% since last quarter, and nobody in the room can produce the explanation, because the analyst who wrote that footnote in Q1 left in March and took the reasoning with her. This is not a slide-design problem. It is a systems problem, and it is exactly where AI agents for board reporting change the math: not by writing better prose around the same disconnected numbers, but by giving the numbers one definition and the narrative a memory.
Why ChatGPT and Cursor don't fix quarterly board prep
Paste this quarter's numbers into ChatGPT or Claude.ai and it will write you a clean board-deck narrative. What it cannot do is remember that last quarter's deck defined pipeline coverage using only late-stage opportunities, or that the NRR footnote from Q1 already explained a churn spike in the mid-market segment. Every session starts cold. Marcus would have to re-paste last quarter's deck, re-explain the metric definitions, and re-summarize every prior board decision each time he wants continuity, which defeats the purpose of asking for help in the first place. The generalist tool is fluent at the ten percent of the job that is sentence construction and blind to the ninety percent that is reconciling what pipeline means across two departments and remembering what the board was already told.
Cursor and GitHub Copilot are not the right comparison at all, but they get asked about often enough that it's worth being direct: they autocomplete code in an editor. Board reporting is not a coding problem. There is no function to complete that produces a reconciled NRR figure or a documented board decision. Even the best autocomplete tool has no access to a CRM export, no notion of what a healthy net revenue retention trend looks like, and no mechanism for flagging that this quarter's deck contradicts what the CEO told the board in the meeting notes three months ago. Applying an editor tool to a governance and financial-reporting problem is a category error, and it's why teams that try it end up right back in the blank-template scramble every quarter.
The deeper issue is that board reporting is a recurring, cross-department, cadence-driven job that depends on three things no generalist tool is built to hold: a single canonical definition for every metric that appears in the deck, a persistent record of what the board already decided and why, and a reporting pipeline that produces the same numbers the same way every single cycle. A chatbot session has none of that memory. An autocomplete tool never had it to begin with. Fixing this requires agents built for the finance, analytics, and institutional-memory roles specifically, not one tool stretched across all three.
There's also a compounding cost to the blank-template approach that boards rarely name directly but always feel: every quarter the deck is rebuilt from scratch, the risk of a new inconsistency goes up, not down. A new analyst joins finance and builds the variance section slightly differently than the departed one did. A board member forwards last quarter's deck to a new investor who asks a follow-up question referencing a number that was never defined the same way twice. None of this is anyone's fault in isolation, it's what happens when the artifact of record for a company's financial story is a document instead of a system. A document gets rebuilt. A system persists, accumulates definitions, and answers the same category of question the same way whether it's asked in Q2 or Q4.
Meet Mint, with Lens defining the metrics and Atlas holding the record
Mint is Tonone's finance engineer for Claude Code: P&L, runway, unit economics, fundraising materials, and board reporting. The mint-recon skill runs first, auditing the current P&L, burn rate, and unit economics so the board package starts from where the business actually is, not from whichever spreadsheet was open last. From that baseline, mint-board assembles the monthly or quarterly financial package itself, P&L, cash position, key metrics, and variance against plan, and mint-report keeps the underlying management reporting and variance analysis running on the same cadence between board meetings, so the quarterly deck is a rollup of a reporting pipeline that already exists rather than a document invented from scratch every ninety days.
Tonone's Mint runs a financial recon before assembling any board package, so the P&L, cash position, and variance figures reflect the current state of the books rather than the last version someone happened to save.
Lens gives every department the same definition of the same metric
The pipeline coverage discrepancy on Marcus's slides four and nine is a metrics-definition problem, and it's Lens's problem to fix, not Mint's. Tonone's data analytics and BI engineer, Lens, runs lens-metrics to produce a complete metrics definition document for a given area: metric name, exact formula, data source, segmentation, and what a good or bad reading actually looks like. Once pipeline coverage has one formula that both Sales and Finance point to, slide four and slide nine either agree or the disagreement becomes a real conversation about which definition the board should adopt, not an accidental contradiction nobody noticed until a board member did. lens-report then keeps that metric flowing on a schedule, scheduled queries, delivery, and historical comparison, so the same canonical number feeds every quarterly package without anyone rebuilding the query by hand.
Tonone's Lens produces a single metrics definition document, formula, data source, and segmentation, so departments stop disagreeing about what a number like pipeline coverage or NRR actually means.
Atlas remembers what the board already decided
The NRR footnote nobody could explain is a different failure: institutional memory that lived in one departed analyst's head instead of anywhere durable. Tonone's knowledge engineer, Atlas, closes that gap with atlas-adr, which documents a board-level decision the way an engineering team documents an architecture decision: what was decided, why, what alternatives were considered, and what trade-off was accepted. When the board approved delaying an EU expansion last quarter given softening NRR, that decision and its rationale becomes a durable record instead of a line in someone's meeting notes. atlas-changelog then tracks how metric definitions and the numbers behind them changed release over release, or in this case quarter over quarter, so when NRR moves from 112% to 104%, the explanation is already logged and citable instead of requiring someone to reconstruct it under board-meeting pressure.
A worked example: Fenwick Route's Q3 board package
Ten days before the Q3 meeting, Marcus starts differently. He runs mint-recon, which pulls the current P&L, confirms burn against the general ledger, and flags that NRR moved from 112% in Q1 to 104% in Q2. Instead of guessing at why, atlas-changelog already has the answer logged from the Q2 cycle: a mid-market cohort churned after a pricing change, flagged and explained the moment it was noticed rather than three months later. lens-metrics has already defined pipeline coverage as late-stage opportunities only, the definition the board adopted in Q2 after the Q1 contradiction, so Sales's CRM export and Finance's spreadsheet are now computing the same formula. mint-board assembles the package against that shared baseline, and atlas-adr surfaces the standing board decision on EU expansion so this quarter's narrative doesn't quietly contradict what was already agreed.
Mint, Board Package: Q3 2026, Fenwick Route
Recon: ARR $22.1M (+$1.4M QoQ), burn $380k/mo, cash $9.6M.
NRR 104% (down from 112% in Q1, see logged explanation below).
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1. P&L and cash
ARR $22.1M. Gross margin 76%. Net burn $380k/mo. Runway 25 months.
2. Pipeline coverage (lens-metrics definition, adopted Q2)
Formula: opportunities stage 3+ only, weighted by historical
close rate. Sales CRM and Finance spreadsheet now agree: 2.4x.
Prior quarter's 3.1x vs 1.8x discrepancy traced to Sales counting
stage 2 opportunities, corrected under the shared definition.
3. NRR movement (atlas-changelog entry, logged Q2)
112% -> 104%: mid-market cohort (18 accounts) churned following
a Q1 pricing change. Flagged same cycle, not discovered at
board meeting. Enterprise segment NRR unaffected at 121%.
4. Standing board decision (atlas-adr-2026-04)
Decision: delay EU expansion to Q1 2027 pending NRR stabilization.
Status: still in effect. This quarter's plan does not include
EU headcount, consistent with that decision.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Prep time this cycle: ~3 hours review, down from ~4 days rebuild.Nobody in the room asks why two slides disagree, because there is only one definition of pipeline coverage now, cited by both departments. Nobody asks why NRR dropped and gets silence, because the answer was logged the quarter it happened, not reconstructed under pressure. And the EU expansion question, which comes up in nearly every board meeting for a company this size, resolves in one line because the standing decision is on record rather than living in whoever remembers the March conversation. The four days Marcus and his analyst used to spend rebuilding a template from scratch became roughly three hours of reviewing a package that was already assembled against a recon, a shared metrics definition, and a documented decision history.
The same pattern holds beyond the boardroom. Fenwick Route's Series C investor sends a monthly update request that used to require a second, parallel scramble to answer, because the investor update and the board deck were built by different people at different times with different source numbers. Once mint-report and lens-report are running on a schedule, the investor update pulls from the same reconciled pipeline the board package does. When Fenwick raises its Series D, the data room's financial section draws from the same recon rather than requiring a third rebuild under diligence pressure. The work Marcus does once, defining pipeline coverage, logging the NRR explanation, recording the EU decision, pays for itself every time someone downstream asks a question that used to require reconstructing the answer from memory.
If your board deck gets rebuilt from a blank template every quarter, start with /lens-metrics on the two or three numbers that cause the most board Q&A, then /mint-recon before the next package assembly. Getting one shared definition per metric eliminates the contradiction before it ever reaches a slide.
Mint, Lens, and Atlas vs the alternatives for board reporting
None of this is a knock on ChatGPT as a writing tool or Cursor as an editor. The comparison below is specific to what quarterly board reporting actually requires: a reconciled financial baseline, one definition per metric across departments, and a durable record of what the board has already been told.
Tonone's Mint, Lens, and Atlas turn a four-day, blank-template board scramble into a three-hour review by starting from a recon, a shared metric definition, and a documented decision record instead of a fresh document each quarter.
| Capability | Tonone | Generalist chatbot | Cursor / Copilot |
|---|---|---|---|
| Quarterly board package assembly | Yes, mint-board produces P&L, cash, metrics, and variance from a recon baseline | No, writes narrative around whatever numbers you paste in each session | No, no board-reporting or financial-data concept |
| One definition per metric across departments | Yes, lens-metrics documents formula, source, and segmentation once | No, has no memory of which definition was used last time | No, autocomplete has no metric-governance function |
| Recurring reporting pipeline vs one-off answer | Yes, lens-report and mint-report run on schedule between board meetings | No, every session is a one-off with no persistence | No, not a reporting or scheduling tool |
| Durable record of board decisions and rationale | Yes, atlas-adr documents what was decided, why, and what trade-off was accepted | No, no persistent memory across sessions | No, out of scope entirely |
| Explaining metric movement quarter over quarter | Yes, atlas-changelog logs why a number moved when it moved, not months later | No, has to be re-briefed on prior quarters every time | No, no historical tracking capability |
Install and try
Tonone is free and MIT-licensed. Install it once and Mint, Lens, Atlas, and the rest of the agent roster are available in the same Claude Code session. You pay only for Claude Code token usage during the work itself.
1. Add to marketplace
2. Install Mint
Frequently asked questions
What does Tonone's Mint do for board reporting?+
Mint is Tonone's finance engineer for Claude Code. It runs mint-recon to baseline the P&L, burn rate, and unit economics, then uses mint-board to assemble the quarterly financial package and mint-report to keep management reporting running between meetings.
How do AI agents for board reporting fix disagreements between departments about a metric?+
Tonone's Lens runs lens-metrics to produce a single definition document for each metric, formula, data source, and segmentation, so every department references the same number instead of computing it a different way.
Can an AI agent remember what a board already decided?+
Yes. Tonone's Atlas uses atlas-adr to document board decisions with the rationale and trade-offs considered, so the reasoning is on record rather than living in one person's memory of the meeting.
How does a board explain a metric that moved significantly since last quarter?+
Atlas's atlas-changelog logs why a key metric changed at the time it changed, so the explanation for something like an NRR drop is already documented instead of being reconstructed under pressure at the next meeting.
How is this different from just using a smarter spreadsheet template?+
A template does not resolve two departments using different formulas for the same metric name, and it has no memory of prior board decisions. Lens standardizes the metric definition and Atlas records the decision history, neither of which a template can do on its own.
How long does quarterly board prep take with Mint, Lens, and Atlas versus manually?+
In the worked example, a four-day manual rebuild of the board deck dropped to roughly three hours of review once the recon, shared metric definitions, and decision record were already in place.
Is Tonone free to use for board reporting agents like Mint?+
Yes. Tonone is MIT-licensed and free. You pay only for Claude Code token usage during the work itself.
How do I install Mint, Lens, and Atlas?+
Install Tonone via the get-started guide at tonone.ai/get-started. Mint, Lens, Atlas, and the rest of the agent roster are all available in the same Claude Code session once installed.