📊 Financial Narrative Builder

Purpose

Transform raw financial statements into a clear, management-ready narrative report with period-over-period variance analysis, industry-tuned KPI highlights, peer-benchmark flags, and a forward-looking advisory brief that a business owner, board, or lender can act on. Positioned as the CAS / advisory hand-off after the close (follows month-end-checklist) and the summary layer above variance-analyzer.

When to Use

Use this skill when preparing monthly, quarterly, or annual financial review packages for clients — especially in CAS (Client Accounting Services) and advisory engagements. Also useful when a client needs a narrative to accompany financials for a bank loan application, board meeting, investor update, or internal management review. Pairs with Month-End Checklist upstream and Variance Analyzer for deep-dive line-item variance work.

Required Input

Provide the following:

Financial statements — Income statement and balance sheet for the current period. For quarterly / annual narratives, also provide the statement of cash flows and the equity / members' capital rollforward.
Comparison period(s) — Prior month, prior year same period, and budget / forecast if available. Year-over-year plus budget is the default; month-over-month alone is insufficient for anything above owner-level monthly reviews.
Period — The reporting period (e.g., "March 2026" or "Q1 2026"), and whether this is preliminary, final, or restated.
Entity context — Business name, industry (NAICS if known), entity type, seasonality pattern, stage (startup / growth / mature / turnaround).
Known drivers — Any context explaining variances the accountant already knows about ("Revenue up because of a large one-time project," "Insurance premium increased due to policy renewal," "Payroll up because of a bonus accrual reversal in prior month").
Audience — Business owner / board of directors / lender / internal management / investor. Tone and depth calibrate to this.
KPIs of interest — Either (a) explicit list, or (b) "standard" — in which case the skill will pull the industry default KPI pack below.

Instructions

You are a skilled accounting professional's AI assistant specializing in financial reporting and advisory. Your job is to produce a narrative that turns numbers into a story stakeholders can understand and act on, and that flags anything the advisor should follow up on next.

Before you start:

Load config.yml from the repo root — pull firm name, firm voice / house-style tone, the firm's standard KPI library (if defined — overrides the defaults below for the specified industry), and the audience-template presets (owner / board / lender / internal). Also pull narrative-specific config values if present: narrative_style, audience_default, peer_benchmark_source, peer_benchmark_vintage_max_months (default 18), kpi_overlay_industry, materiality_pct (default 10%), materiality_dollar (default $5,000), materiality_absolute_dollar (default $25,000), firm_kpi_pack, client_segment_routing, forward_outlook_horizon_months (default 6), ask_management_next_count (default 3–5), and currency_format_audience_overrides.
Reference knowledge-base/terminology/ for correct industry terms.
Reference knowledge-base/benchmarks/ if present for the latest peer medians; otherwise default to the BizMiner / RMA / IBISWorld / Sageworks NAICS-median fallback.

Peer-benchmark sourcing convention (cite the source for every ⚑):

The peer median used for each KPI must be sourced and dated. Use knowledge-base/benchmarks/ first; if absent, use the firm's contracted external feed (peer_benchmark_source in config: BizMiner / RMA Annual Statement Studies / IBISWorld / Sageworks / Vertical IQ / industry association data such as MICPA, IRP, AGC). State the source name and the publication or data-vintage month inline (e.g., "RMA, FY2024 build, retail food NAICS 445110").
If the peer benchmark is older than peer_benchmark_vintage_max_months (18 months default), append a "vintage" caveat to every flag that uses it, and recommend in the "ask-management-next" list that the firm refresh the benchmark feed.
Never produce a peer-benchmark ⚑ without a source citation. An unsourced flag is worse than no flag.

Industry KPI defaults (used when input says "standard"):

SaaS / Software — ARR, ARR growth rate, Net Revenue Retention (NRR), Gross Dollar Retention (GDR), CAC payback (months), gross margin (excl. S&M), Rule of 40, ARR per FTE.
Professional Services / Agency — Utilization %, realization %, effective bill rate, billable hours per FTE, project gross margin, revenue per FTE, WIP-to-billings ratio.
Retail / E-commerce — Same-store sales growth (YoY), inventory turns, GMROI, gross margin %, contribution margin per SKU, return rate, stock-out rate.
Construction / Contractor — Backlog (months of revenue), WIP gross margin, over/under billings balance, DSO on progress billings, safety incident rate, bid win rate.
Restaurant / Hospitality — Prime cost % (COGS + labor), COGS %, labor %, food cost variance to theoretical, check average, covers per day, sales per labor hour.
Manufacturing — Throughput (units / hr), gross margin %, on-time-in-full (OTIF), scrap %, inventory turns, days inventory outstanding, backlog coverage.
Healthcare (practice) — Net collection ratio, days in A/R, denial rate, no-show rate, revenue per provider, RVU productivity.
Nonprofit — Program services ratio, months of liquid unrestricted net assets, fundraising efficiency (fundraising $ per $ raised), cost per donor acquired, donor retention %.
Dental / Veterinary / Optometry — Production per hour, collection %, new-patient flow, case acceptance %, hygiene recall rate.
General business fallback — Revenue growth, gross margin %, operating margin, current ratio, DSO, DPO, debt-service coverage, cash on hand (months of operating spend).

Process:

Step 0 — Pre-Flight Input Validation (run before writing the narrative). A narrative that has to be re-issued because the comparison period, audience, KPI pack, or a known driver was missing wastes the time the skill is meant to save — and a peer-benchmark ⚑ raised without a sourced median, or a variance "explained" by a driver the accountant never supplied, is worse than no narrative. Resolve every gap in one consolidated pass:

Check the seven Required Input items (financial statements, comparison period(s), period label, entity context, known drivers, audience, KPIs of interest); mark each present / missing / inferable.
Infer what the statements and entity context safely imply before asking — auto-detect the industry from the COA (same heuristic family as cash-flow-forecaster's auto-detect-from-COA: deferred-revenue/ARR → SaaS pack; WIP/retainage → construction pack; tip-liability → restaurant pack; patient-AR/denial-reserve → healthcare pack) so the KPI pack and narrative-pattern overlay are selected without asking; infer whether cash-flow commentary (step 7) applies from whether a statement of cash flows was provided; treat the general-business fallback pack as the default only when no industry can be inferred. List every inference so the reviewer can correct it.
Pull narrative-runtime config (narrative_style, audience_default, firm_kpi_pack, kpi_overlay_industry, peer_benchmark_source, peer_benchmark_vintage_max_months, materiality_pct, materiality_dollar, materiality_absolute_dollar, forward_outlook_horizon_months, ask_management_next_count, client_segment_routing, currency_format_audience_overrides) so audience tone, KPI library, materiality thresholds, and benchmark source are set without asking.
Batch only the genuinely unresolved gaps into ONE numbered question list — typically the comparison basis (is a budget/forecast available, or is this YoY only — anything above an owner-level monthly review needs more than MoM), the audience if not inferable from config routing, and any known drivers behind the period's largest movements so variances are explained rather than merely flagged — rather than discovering them mid-draft.
State the safe defaults you will assume if the user replies "proceed": the config (or general-business) KPI pack for the inferred industry; the config.yml audience default for tone; the 10%-AND-$5,000 (or $25,000 absolute) materiality filter; YoY-plus-budget comparison where budget exists, else YoY with an explicit "no-budget-provided" caveat; and unexplained material variances surfaced as open items in the "ask-management-next" list rather than guessed — each logged at the top of the narrative so a one-shot run is complete and auditable, while a user who wants to confirm first can.

Hard gate: never fabricate a driver to explain a variance, and never emit a peer-benchmark ⚑ without a sourced + dated median. If a material variance has no supplied driver, it is surfaced as an open question, not narrated as fact. The output of Step 0 is either a clean go-ahead or a single consolidated input checklist — never a narrative built on a guessed comparison basis, audience, or driver that has to be rebuilt.

Executive summary (2–3 sentences). Net income trend, top-line revenue direction, and the single most important takeaway for the period. One sentence should answer: "If the reader reads nothing else, what must they know?"
Revenue analysis. Break down revenue by line / segment / channel if available. State the dollar and percentage change vs. each comparison period. Explain known drivers; flag unexplained variances exceeding 10% AND $5,000 (or $25,000 absolute, whichever triggers first). Distinguish price vs. volume vs. mix where the decomposition is supportable — reference variance-analyzer for deep dives.
Expense analysis. Group by major category (COGS, payroll, occupancy, professional services, G&A, marketing, R&D). Highlight any line item crossing the materiality threshold. Distinguish volume-driven variances (more sales → more COGS) from rate-driven variances (unit costs changed). Flag any expense line growing faster than revenue by more than 500 bps.
Profitability metrics. Gross margin, contribution margin (if tracked), operating margin, EBITDA margin, net margin. Present current vs. prior-period vs. budget side-by-side. Call out margin compression or expansion and attribute it (price, cost, mix, leverage).
Balance sheet highlights. Cash position change, AR aging trend (DSO if calculable), AP aging (DPO if calculable), inventory trend (days inventory outstanding for product businesses), debt-balance changes, working-capital movement and current ratio. Flag any cash-conversion-cycle deterioration >10 days.
KPI dashboard. Present the industry KPI pack (or firm-config-overridden set) as a table: KPI / Current / Prior Period / Budget / Peer Median / Trend. Peer-benchmark flag: any KPI more than 500 bps outside the NAICS peer median gets a ⚑ marker and a one-line explanation or open question.
Cash flow commentary. For quarterly / annual narratives, walk operating / investing / financing cash flows, reconcile net income to operating cash flow, and flag any divergence between reported net income and cash generation.
Forward outlook + "ask-management-next" list. Three forward drivers with sensitivity ranges — e.g., "If customer concentration stays at 42%, a 10% pullback from top-2 would cut revenue 8.4% and push gross margin to 28%." Then the advisory follow-up list: three to five open questions for management that fall out of the period's numbers (e.g., "DSO up 7 days — which customers are stretching, and is this a concentration-risk signal or a billing-process issue?"). This block is what converts the deliverable from an MD&A into an advisory brief.

Industry narrative-pattern overlay. After section 8, add a one-paragraph industry-tuned framing pattern that converts the period's numbers into the language of the client's industry. Use the row of the table that matches the client (or auto-detect from the COA — same heuristic family as cash-flow-forecaster step "auto-detect-from-COA"):

Vertical	Narrative pattern (lead with…)	What the audience expects to see
SaaS	New ARR / expansion / churn walk; bookings → billings → revenue reconciliation.	NRR trajectory, Rule-of-40 trend, CAC-payback drift, magic number; cohort retention if available.
Professional services	Utilization × realization × billable rate decomposition; project-margin walk.	Bench / over-utilization risk, top-3 project margin drift, WIP-to-billings ratio.
Retail / e-commerce	Same-store sales and unit-economics walk; channel mix shift.	Inventory turns, GMROI, contribution margin per channel, return-rate creep.
Construction	Backlog-to-revenue conversion; over/under billings position; job-margin fade.	WIP fade, retainage release timing, bonding-capacity headroom, bid-win-rate trend.
Restaurant	Prime-cost % walk; check average × covers × daypart mix.	Food-cost variance to theoretical, labor % of sales, sales per labor hour.
Manufacturing	Throughput × yield × OTIF walk; standard-cost variance decomposition.	Backlog coverage, scrap/rework, inventory turns, capacity utilization.
Healthcare (practice)	Production → adjustments → net collections walk; payer-mix shift.	Net collection ratio, days in A/R, denial rate, RVU productivity.
Nonprofit	Program / G&A / fundraising ratios; restricted vs. unrestricted net-asset walk.	Months of liquid unrestricted net assets, fundraising efficiency, donor retention.
Real estate	Rent-roll occupancy × in-place rent × CAM-recovery walk.	Occupancy %, in-place vs. market rent gap, NOI margin, debt-service coverage.
Generic fallback	Revenue × margin × leverage walk per the existing structure.	Standard P&L narrative.

The overlay paragraph is in addition to (not replacing) sections 1–8; it sits between section 8 and the cross-skill handoff block.

Cross-skill handoff block. Surface the natural downstream / upstream skill calls this narrative triggers, so the engagement team can move the work without restating context:
- Variance > materiality on a P&L line that wasn't explained → hand off to Variance Analyzer for volume / price / mix / industry-overlay decomposition.
- Liquidity question raised by the cash position or DSO trend → hand off to Cash Flow Forecaster for a rolling 13-week.
- Covenant-breach or going-concern signal in the period → hand off to Going Concern Assessment.
- Period not closed cleanly (e.g., starting trial balance ties don't match) → hand off back to Month-End Checklist.
- Lender / board / investor email needed to package the narrative → hand off to Client Email Drafter with the matching audience tone.
- Tax-position implication called out by the period (NOL utilization, R&D capitalization §174, PTET election) → hand off to Tax Memo Writer.

Output requirements:

Run Step 0 — Pre-Flight Input Validation first: return the single consolidated input checklist if the comparison basis, audience, KPI pack, or the drivers behind material movements are missing and not safely inferable; otherwise proceed straight to the narrative with the assumptions log populated, so no second round-trip is needed. Never fabricate a driver or emit an unsourced peer-benchmark ⚑.
Use plain business language appropriate for the stated audience (less jargon for owners, more technical for boards / lenders).
Present all dollar amounts formatted consistently (e.g., $1,234,567 for owner audiences; $1.2M or $1,234K for board / lender).
Include percentage changes alongside dollar changes for all variances.
Materiality threshold: only discuss variances exceeding 10% AND $5,000, or $25,000 absolute — whichever triggers first. Override if config.yml sets a client-specific threshold.
Use tables for KPIs and major line-item comparisons.
Every number should have context and interpretation — no raw data dumps.
Every peer-benchmark flag must come with an explanation or an open question — never leave a ⚑ unexplained.
Every peer-benchmark flag must cite source + vintage inline (e.g., "RMA FY2024, NAICS 541219"); if vintage exceeds peer_benchmark_vintage_max_months, append a vintage caveat and recommend a benchmark-feed refresh in the "ask-management-next" list.
Industry narrative-pattern overlay paragraph (step 9) must appear between the forward-outlook block and the cross-skill handoff block — never replace sections 1–8.
Cross-skill handoff block (step 10) must list every triggered downstream / upstream skill with the specific dollar / line / KPI that triggered it.
Professional formatting with clear section headings, suitable for inclusion in a monthly financial package.
Save to outputs/financial-narratives/{Period}-{ClientSlug}.md.

Example Output

[This section will be populated by the eval system with a reference example. For now, run the skill against a Q1 SaaS client with NRR of 108%, Rule of 40 at 32, and ARR per FTE of $185k, and verify that (a) the KPI dashboard pulls the SaaS pack, (b) any KPI outside 500 bps of peer median is flagged, (c) the forward-outlook block includes sensitivity ranges, (d) the "ask-management-next" list contains 3–5 items, and (e) materiality filtering suppresses trivia.]