RPP Sales Plan: Deep Dive and Landing Plan

TL;DR

The four answers

1 · How do we get version control?

There is none today, in either the sheets or the app. The "(UNLOCKED)" in the filenames is just a name: zero sheet protection exists on any of the 6 workbook files (verified via Sheets API and xlsx export). In RPP, every edit overwrites the single plan row in place; nothing is snapshotted. Fix: a status + lock field this week, then immutable plan versions (Budget locked, Forecast rolling, Reforecast snapshots), which is also the industry-standard model.

2 · Why is December missing for HOUZN?

It is not missing in the 2026 Sales Goal workbook. All 12 months are populated in all three properties' workbooks (cells pulled 2026-07-19), and all HOUZN scenarios in the RPP database carry December too. Two real things hide behind the impression: HOUZN's December RevPAR cells were hand-typed over broken formulas (the only such patch in all 3 files), and the separate file "HOUZN - Sales Team Goals - 2026" is completely blank, every month $0. If you saw an empty December, it was almost certainly that second file.

3 · What's up with the fonts?

It is all one font (Arial), zero exceptions in 4,900+ cells. What reads as a font mess is size and color: the pasted STR "Response" tab renders its dot-plot with 18pt bullet characters next to 10pt data tables, plus hardcoded blacks and grays that ignore the theme. The GM-facing Sales Goals tab is actually clean and identical across all three properties.

4 · What does each tab do, and is it clunky?

The app is a superset of the sheet's math but is missing the sheet's one irreplaceable job: entering the budget. The in-app budget editor was deleted 2026-07-17 as dead code and never replaced, so budgets exist only for the 3 seeded pilots and 3 of the 5 tabs dead-end everywhere else. Full tab-by-tab and a ranked clunk audit below.

Ground truth

The original: where it lives and how it works

The paper ancestor of the RPP Sales Plan is one Google workbook per property, built during Ronald's December 2025 budget cycle (GM deadline 2025-12-20, email 2025-12-12, Spark msg 13967).

Property	File	Created	Last edited	Drive path
HOURP	2026 Sales Goal HOURP - v1 (UNLOCKED)	2025-11-25	2026-02-16	Drive (THM) / HOURP - SpringHill Suites Houston Med Center / HOURP - Sales and Marketing / Sales Performance / Sales Goals and Targets
HOUUS	2026 Sales Goal HOUUS - v1 (UNLOCKED)	2025-11-24	2026-02-16	.../ 01 - Sales Performance / 01B - Sales Goals and Targets
HOUZN	2026 Sales Goal HOUZN - v1 (UNLOCKED)	2025-11-20	2026-02-16	.../ 01 - Sales Performance / 01B - Sales Goals and Targets

File IDs: HOURP 1fCx4Jwp...wy_4, HOUUS 1C_xQkVl...eUQQ, HOUZN 1hLyourB...27Wg (drive metadata via gog, pulled 2026-07-19). Each folder also holds the frozen 2025 edition. All three 2026 files were last touched within 7 minutes of each other on Feb 16, 2026, and not since: one batch data-entry session, then 5 months of dormancy.

Three tabs per workbook

Sales Goals (43 rows): the actual model. 2025 actuals block on top, 2026 projection block below.
Comp: a pasted STR "Tab 4 - Competitive Set Report" (December 2025 edition). Reference data, not formula-linked to the projections.
Response: a pasted STR "Tab 22 - Response Report", same vintage. Also reference-only.

The mechanism: three orange inputs, everything else cascades

Row 30 Market Growth % (monthly)

Row 33 Occupancy Adj % vs 2025

Row 36 ADR Adj % vs 2025

→

Projected Occ / ADR = 2025 actual × (1 + adj)

→

RevPAR & Revenue = RevPAR × days × rooms

→

Projected RGI = TTM RGI × (1 + own growth − market growth)

The punchline of the design: the GM's plan is forced to declare a market-share position (projected RGI), not just a revenue number. HOURP's 2026 plan landed at $6,534,417 projected on $6,093,340 TTM 2025 (+6.8%), projected RGI 101.2 vs 96.0 actual (Sales Goals rows 9, 28, 40, 41).

Known quirks in the sheet math: the Growth row computes 1 − (2025 ÷ 2026), which understates growth (Feb shows 10.7% where the conventional calc is 12.0%: 523,340 / 467,268 − 1). Revenue is STR-derived (occ × ADR × rooms × days), so it never ties exactly to QBO account 4100. And the whole workbook is static: STR pasted once, no live actuals, frozen since Feb 16.

Downstream: the plan's monthly Growth row is copied into "HOURP - Sales Team Goals - 2026" (same folder), which sets the group-sales team's monthly revenue goals. HOURP's tracker has 2026 actuals through April; HOUUS has no tracker file at all; HOUZN's exists but is empty (see question 2).

Question 1

Version control: what exists, what to build

Current state, sheets

"(UNLOCKED)" is decorative. No sheet protection or protected ranges exist on any of the six Sales Goal files (2025 + 2026, all three properties), verified two ways: Sheets API metadata shows no protectedRanges, and the xlsx exports show protection.sheet = False on every tab.
Version control is a filename convention ("v1", a 2025 file next to a 2026 file) plus whatever Google's built-in revision history holds. Nobody has named or pinned versions; our tooling cannot even read revision history (gog has no revisions command), only whole-file modified times.
Anyone with edit access can silently change the "locked" 2026 budget today; the only reason it has not drifted is that nobody has touched the files since 2026-02-16.

Current state, app

One row per property + year in sales_plans (unique key, migration 037). Every save is an upsert that overwrites in place (src/hooks/useSalesPlan.ts:613). Yesterday's numbers are gone the moment you edit a cell.
The version column is a schema-shape marker (1 = V1 layout, 2 = V2), not version history. It never increments and no UI reads it.
No draft / submitted / approved / locked status. No audit trail of who changed what. The pattern exists next door (deal_assumption_versions, migration 259) but was never applied here.

How to get it

Option	What it is	Gets you	Effort*	Verdict
A. Status + lock	Add status (draft / submitted / approved / locked) + approved_by/at to sales_plans; gate writes when approved or locked; status badge in the UI.	The plan stops silently changing. A real "this is the committed budget" state.	0.5-1 day	Do this week
B. Immutable versions	New sales_plan_versions table: on approval, the full plan (scenarios + the three adjustment maps + targets) is frozen as a labeled snapshot ("Original Budget", "Q2 Reforecast"). The live row stays the working draft. Version picker in the top bar; the existing Scenarios tab becomes the compare view.	Exactly the industry model: Budget locked, Forecast rolling, Reforecasts as approved snapshots, all comparable side by side.	2-4 days	The real answer
C. Full change log	Append-only log of every field change (who, when, old, new), point-in-time reconstruction.	Forensic audit trail ("who moved June ADR").	3-5 days	Only if lender/partner scrutiny demands it

*Effort figures are engineering estimates from the code dissection (assumption, not measured). Detail: the snapshot payload reuses the JSONB shapes that already exist in sales_plans, so option B is mostly a migration + an approve RPC + a version picker.

Once B ships, the Google workbooks stop being the system of record: the December budget cycle happens in RPP (GM types the same three levers, Ronald approves, the plan locks), and the sheet becomes an export, not a source.

Question 2

HOUZN December: found it

Checked three places: the Google workbook, the sibling files, and the RPP database. December exists in the workbook and the database. The blank December lives in a different file.

Where	December state	Evidence
2026 Sales Goal HOUZN workbook	Fully populated	All 12 months + totals present in both the 2025 actuals and 2026 projection blocks (cells pulled 2026-07-19): Dec 2025 occ 56.23%, Dec 2025 RevPAR $62.72, Dec 2026 projected ADR $113.79, RevPAR $64.63. Comp/Response tabs carry the full December 2025 STR report (created Jan 17, 2026, STR ID 63698, 9-member compset).
RPP database (sales_plans)	Fully populated	All four HOUZN scenarios (2025 Actual, GM Budget 2026, SMART Forecast 2026, Fair Share 2026) carry months 1-12; GM Budget December: $278,475 revenue, 56.8% occ, $113.79 ADR (live query 2026-07-19).
"HOUZN - Sales Team Goals - 2026" tracker	Entirely blank	Every row Jan-Dec, all years 2022-2026, is empty; the sheet's own total row computes $0 (cells pulled read-only 2026-07-19 from file 1CiAgtwC...IE4k). Last touched 2026-06-30, the most recently edited file in the whole investigation. HOURP's equivalent tracker has actuals through April 2026 (same-day pull of file 1kPj1UrS...-sfw); HOUUS has no tracker file in its folder.

The forensic detail worth knowing: in the HOUZN workbook, the two December RevPAR cells (Sales Goals O19, O20) are hand-typed constants (62.72 and 53.7) where every other month, and both sibling workbooks, use live formulas (=O15*O11). It is the only formula break in all three files. Someone manually patched HOUZN's December during the Feb 16 batch session; the typed value is 1 cent off what the formula would produce, consistent with copying STR's own rounded number. December needed manual help once, but it is there.

So the answer: if you saw a missing December for HOUZN, you were most likely in the Sales Team Goals tracker (which is 100% empty and was touched as recently as June 30), not the Sales Goal workbook. If you did see it blank in the workbook itself, that state no longer exists and predates nothing we can reconstruct: revision history is not readable through our tooling, so pre-Feb-16 states are unverifiable. Flagging that honestly rather than inventing a cause.

Question 3

Fonts: one family, chaotic sizes

All three workbooks were exported to xlsx and every populated cell's font inventoried (4,900+ cells total, ~1,600-1,800 per workbook).

Font family: 100% Arial. Zero non-Arial cells anywhere. The "pasted STR data brings its own fonts" theory is disproved.
The real noise is size and color, concentrated in the pasted STR tabs:
- The Response tab renders STR's dot-plot chart as 18pt bullet characters (149-221 cells per workbook, scaling with compset size) directly above 9-10pt data tables. This is the single biggest "what is going on here" visual.
- Header rows in the Comp tab use hardcoded black instead of theme black (90 cells, invisible now, breaks if theming ever changes), an 11pt label column against the 10pt body, and light-gray footnote text.
The GM-facing Sales Goals tab is clean: consistent formatting, identical cell-for-cell across all three properties (one stray explicitly-formatted cell in HOURP, P40, trivial). 18-19 distinct font/size/bold/color combos per workbook, and nearly all of the variety is the two pasted STR tabs.

Practical takeaway: nothing to fix by hand. The mess is an artifact of pasting STR report renders into Sheets, and it disappears entirely once RPP is the surface, because RPP already ingests the same STR data into compset_metrics and renders it natively (the Compset Intel tab). The pasted Comp/Response tabs are exactly the part of the workbook RPP has already made obsolete.

Question 4

The app, tab by tab

Route /app/sales-plan, five tabs. Property picker defaults to the most recently updated property that has a plan; properties are tagged "plan" / "no plan".

Tab	What it does (plain English)	Data source	You can edit	State
Forecast	The heart, and the direct descendant of the Google sheet. Auto-pulls last year's STR actuals for you and your compset, gives the GM the same three levers (Market Growth, Occ Adjustment, ADR Adjustment), instantly projects occ/ADR/RevPAR/revenue and projected RGI per month. Closed months show real actuals. A rank-goal bar lets you set a target compset rank, and "Wynn" auto-suggests the lever values needed to hit a target RGI.	STR via compset_metrics (auto), room counts via budget_assumptions, math in salesPlanCalc.ts	3 adjustment rows, target rank	Works, best-in-class core
Pace	"Will this month make budget?" Budget vs month-to-date actuals vs on-the-books, plus the pickup still needed and the ADR that implies.	night_audit_current + booking_pace + the GM Budget scenario	Nothing (read-only)	Blank without a pace feed + budget
Actuals vs Plan	Monthly report card for closed months: actual vs GM Budget, green/amber/red variance, YTD summary, STR RGI column where it exists (never fabricated).	night_audit_current + GM Budget scenario + compset_metrics	Nothing (read-only)	Dead-ends without a budget
Scenarios	Side-by-side comparison of stored scenarios (GM Budget, SMART Forecast, prior-year actual): weighted annual cards + a 12-month revenue matrix.	scenarios JSONB on the plan row	Nothing (read-only)	Redundant today; becomes the version-compare view
Compset Intel	18-month STR trend table (occ/ADR/RevPAR with MPI/ARI/RGI and ranks) plus the compset member roster.	compset_metrics + compset member tables	Nothing (read-only)	Duplicates the Compsets page

The clunk audit, ranked by user pain

	Problem	Why it hurts	Where
1	You cannot create or edit a budget in the app. The budget editor (BudgetBuilderTab, 467 lines) was deleted 2026-07-17 as already-dead V1 code and never replaced.	Budgets exist only for the 3 seeded pilots (HOURP, HOUUS, HOUZN, verified by live DB query). For any other property, "Create Sales Plan" makes an empty shell and Pace, Actuals vs Plan, and the Forecast's budget rows all dead-end. The single biggest blocker to landing the feature.	commit f57e8a0; SalesPlan.tsx:71
2	Two "plans" with an unexplained relationship. Editing the Forecast levers changes the SMART Forecast scenario, but Pace and Actuals vs Plan always grade against the GM Budget.	A GM will assume the numbers they edit are "the plan." Nothing on screen says otherwise except a footnote.	ForecastTab.tsx:349
3	Projected revenue silently shows $0 when budget_assumptions is missing for the property/year.	RevPAR rows populate while the revenue KPI reads $0. Looks broken, no warning shown.	ForecastTab.tsx:73-81
4	RGI is averaged three different ways on one screen (revenue-weighted KPI, arithmetic-mean grid total, arithmetic TTM in the rank bar).	The numbers visibly disagree with each other, eroding trust in all of them.	salesPlanCalc.ts:264; ForecastGrid.tsx:38
5	Occ adjustment is points in the app, was % in the sheet.	A GM trained on the sheet typing "+3" gets +3 occupancy points, not +3%.	salesPlanCalc.ts:165
6	The brand forecast is stored with no provenance. The SMART forecast is Marriott's own number (Ace, 2026-07-19), but the app stores it as an anonymous scenario, indistinguishable from something we computed.	Who produced a forecast, when, and via what feed is the whole ballgame for a multi-brand operator. Fixed by the forecast-lane model below.	Spark msg 179768; sales_plans.scenarios JSONB
7	Wynn's goal-seek is a black box (magic weighting constant, hard caps, unexplained per-month suggestions).	GMs will blindly accept or entirely ignore it.	salesPlanCalc.ts:352-453
8	No role scoping. RLS is tenant-only; every role that sees the nav, including front desk, can edit any property's plan.	Fine for internal pilot, untenable the day GMs get logins.	migration 259:15; nav-config.ts:131
9	Half the data model is dead (target_revpar_rank, rgi_goal, fair_share_formula, market_context, actuals_through: no UI reads them), plus dead V1 hooks including the orphaned GL-4100 budget sync.	Cruft, and the market-events overlay everyone wants is already modeled but invisible.	useSalesPlan.ts:129-292
10	Five tabs where three carry the load; invisible save state; heavy horizontal scroll on mobile.	Feels bigger and clunkier than it is.	SalesPlan.tsx:118

Research

What normal hotels do for budgeting

Web research across industry guides, tool vendors, and asset-management sources. Full source list in the footer note; key claims linked inline.

The standard process

Budget season runs roughly August to November, locking before year-end: history pull in August, first draft September, review rounds October-November, ownership approval by late November (Lighthouse, Otelier). Expect 2-4 iteration rounds. THM's own cycle (Ronald's kickoff late Nov, GM inputs due Dec 20) runs about a month later than the industry norm.
Rooms revenue is built from prior-year same-month actuals, adjusted by a seasonal index, a special-events overlay (a strong citywide is worth 4-6 occupancy points and 10-15% ADR per Gourmet Marketing), and segment stacking. For select-service, 3-5 segments is plenty; the 12-segment matrix is convention-hotel overkill.
Best-in-class anchors the budget to STR fair share: set next-year RGI targets and back-solve the occ/ADR needed (Docyt, Mews). Benchmarks: 100-110 healthy, sub-90 triggers asset-manager review (RevPARGenius). This is exactly what the THM sheet and the RPP Forecast tab already do, and almost nothing affordable does it.
Expenses are driver-based on three drivers (% of revenue, fixed monthly, per-occupied-room), which is also exactly THM's budget template convention (columns D/E/F with a MAX formula). Select-service labor guardrails: ~1.44 hours per occupied room total, ~$28 housekeeping cost per occupied room (hoteldata). Rooms-department flow-through should run 60%+ (prostay).

The version model everyone converges on

"The budget is a commitment made once a year, the forecast is the truth updated monthly, and when they diverge, the forecast wins with the variance log explaining why" (Model Reef). One model, three views: Budget (locked at the assumption layer), Forecast (rolling, updated monthly), Reforecast (approved snapshot when the plan is no longer the plan). Monthly cadence, every variance over 5% gets a one-line written explanation, forecast accuracy targets of ±5% at 30 days out.

Brands do ask for the numbers

Correction to the initial research (Ace, 2026-07-19, confirmed in the inbox): franchise revenue management requests budgets. Marriott's SMART program runs on HOUZN (Amy Wickliff's SMART Recaps to Sam Xie and ownership, Spark msg 179768) with One Yield as the yield system (registration approval msg 63402), and IHG's equivalent covers HOUUS. So the module needs a brand-RM-ready output: monthly occupancy / ADR / RevPAR in the shape the revenue manager loads into One Yield or Concerto, on the brand's calendar. What it does not need is franchise-compliance workflow beyond that.

Tools reality check

Tier	Tools	Cost signal	Who uses it
Enterprise budgeting suites	Actabl ProfitSword, Otelier TruePlan, Aptech Targetvue, Fairmas	"Contact us" pricing	20-30+ property groups, PE/asset-manager mandates
RMS forecasting	IDeaS G3, Duetto	~$1,200-3,500 / mo / hotel*	Full-service and better-performing select-service
Accounting-platform budget modules	M3, Inn-Flow	Bundled with accounting	The select-service mainstream
The actual default	Excel / Google Sheets + STR STAR	$0	Most 10-30 property operators, including THM today

*Directional third-party figures (exploretech for IDeaS, Duetto from vendor marketing), not quoted price sheets. The gap in the market: nothing self-serve and affordable does STR-anchored budgeting for a 15-hotel select-service operator. That gap is RPP's wedge at $299/mo.

Adversarial review, 2026-07-19

The review board

Seven independent reviews of this plan: the four W5 lenses, two cross-vendor advisors on non-Claude models, and the Hormozi offer lens. All seven returned REVISE. The plan below already incorporates their accepted findings and Ace's round-2 decisions.

Reviewer	Lens	Sharpest finding
Wynn	Revenue science	Compute projected RGI per month as an exact ratio (not the flat TTM approximation); grade the budget on attainment but prediction lanes on accuracy at matched snapshot horizons, or GMs learn to sandbag.
Warren	Money math	Measure THM's actual STR-to-GL revenue gap and calibrate the 4100 sync with it; store integer rooms-sold as the budget primitive; a reforecast approval must never silently re-baseline the locked budget.
Walton	Operations	Brand-recap feeds need expected-cadence freshness alerting like every other ingestion pipeline; the monthly ritual needs the same reminder machinery as the annual season or it decays like the trackers did.
Woz (fresh eyes)	Engineering	The "SMART Forecast" scenario in code is the GM's own computed projection, not Marriott's number (two-lane split required); snapshots must freeze computed outputs, not input levers; the orphaned 4100 function is broken as written.
Andrej (GPT-5.4)	Systems rigor	No cutover acceptance bar: freeze a golden corpus from the pilot workbooks and assert month-level parity before the sheet loses system-of-record status; approval needs atomicity and a read-back reconciliation assertion.
ChaseAI (Gemini)	Scope discipline	Ship P0 completely before any P1; the auto-suggest black box must be fixed or removed, not shipped as-is; gate the GL sync.
Hormozi	The offer	Flat $299/mo breaks the pricing model for a portfolio product; the accuracy-grading proof can be built from THM's existing history now; nothing in the plan was designed to make an operator switch.

Adopted from the board (engineering calls)

Snapshots freeze outputs: a locked version stores the materialized monthly numbers plus the compset roster, STR data vintage, and room counts. STR restates history; a locked budget that recomputes on read is not locked.
Two-lane split for the brand forecast: the lever-driven computed lane becomes "GM Model Forecast" (source: rpp_model); a "Brand RM Forecast" lane exists only when the parsed Marriott/IHG feed writes it. No fabricated provenance.
One RGI definition: ratio-of-sums, computed monthly against monthly compset RevPAR, used identically in the KPI, grid totals, rank display, and the suggest logic.
Grading discipline: lanes snapshot at fixed lead times (90/60/30 days out) and are graded only at matched horizons; the GM budget is graded on attainment (beating it is good), prediction lanes on closeness; a month's grade finalizes on a defined close, with restatements handled explicitly.
Money-path hardening: integer rooms-sold stored on promote; the sheet's understating growth formula corrected in the rebuild; the last-year seed defaults to last year times market growth and cannot lock without an implied-RGI checkpoint; approval is atomic with a read-back assertion (snapshot total = budget scenario = GL 4100 sum).
Ops layer: recap lanes register expected cadence in the existing freshness monitor; monthly reforecast gets reminders and per-property completion tracking; tabs without a live feed say so instead of rendering blank.
Honest sizing and sequencing: P1 splits into three slices (versions + backfill, roles, season workflow), each estimated separately; a backfill mints "Original Budget" snapshots for anything approved during P0; golden-corpus parity against the pilot workbooks is the cutover acceptance bar.
Deferred or dropped: market-events overlay deferred; the IHG parser is sized only after a real IHG recap is captured and inspected; dead schema columns cleaned up; the full change log (option C) is an explicit non-goal.

Round-2 decisions (Ace, 2026-07-19)

Decision	Call
Occupancy lever	REVERSED to points (stored and computed), displayed unmistakably as "+3.0 pts" with a translation tooltip for sheet-trained GMs. No pilot data migration, no suggest-logic rewrite.
4100 sync guardrail	Calibrated auto-sync: per-property trailing-12 STR-to-GL ratio applied and shown at approval; residual gap over ~5% flags a data-integrity review instead of posting as a grade (threshold is Warren's convention, tuned after measuring).
Pricing	Per-property base plus premium tier (accuracy grading, multi-brand aggregation, portfolio rollup are the premium). Warren models the per-door number before the first external conversation.
Proof artifact	Build the accuracy-grading case study on THM's own history during P0/P1, as both the sales proof and the design spec for the live P2 feature.
Design partner	Hard gate: when P0 proves on a non-pilot property, one external management company onboards (founder-priced) so P1/P2 are shaped by two operators.
Auto-suggest	Rebuilt as Ace's manual process: per-month, raise-only against the hotel-set RGI target, each suggestion with its one-line reason, capped, never lowering months already above target. Stays in for the 2027 season.
Compset Intel tab	Dropped from the module; a link jumps to the Compsets page with the property preselected.
Portfolio rollup	Built during the 2027 season work; aggregates only approved locked budgets, everything else shows "no plan yet". Ready as the design-partner demo.
The offer	Free graded-budget diagnosis ("if we don't show you something you didn't know, you pay nothing") as the opening offer, reusing the case-study machinery.
Support path	Ronald is first line for GM data questions with in-app explainers ahead of him; genuine data or bug issues escalate to Woz.

Recommendation

The landing plan

The Forecast tab's core (STR-anchored, three levers, projected RGI, rank goals) is already better than anything in the affordable tier. What is missing is the boring institutional half: a budget you can enter, lock, version, and report against. Three phases.

Locked decisions (Ace, 2026-07-19 Q&A)

Decision	Call
Budget entry	Promote forecast to GM Budget (tune the three levers, freeze as budget, inline month tweaks allowed before approval). Plus a one-click "Set budget = last year" seeding option, since the non-pilot properties budget off prior year today.
Roles	GM drafts own property, ownership approves and locks; Ronald/Junaid get portfolio-wide staff edit.
Occupancy lever	Percent, matching the workbook. Superseded Reversed to points in round 2 after the engineering review (see Review Board section).
First in-app season	2027 budget season starting September 2026, lock by early December; dry run = a 2026 mid-year reforecast on the pilots.
Rollout scope	The 3 pilots stay the real users for now. Wider THM rollout later; last-year seeding covers the rest meanwhile.
IHG feed	IHG sends an email recap like Marriott's (Ace), so the IHG parser follows the SMART parser on the same pattern.
Accuracy grading	Visible to ownership and GMs, every lane including the brand RM's.
Segments	One group/negotiated lane only in P2, replacing the Sales Team Goals sheets. No full segmentation yet.
P&L scope	Rooms revenue only. Approved budget auto-syncs to GL 4100 in the budget/BVA system (revive the orphaned sync); expenses stay in the budget module.
Connector timing	After P1 lock ships: SMART parser then IHG as the first P2 items.

Brand-agnostic by design (locked with Ace, 2026-07-19)

The SMART forecast comes from Marriott directly; IHG has an equivalent that is not wired in; many management companies use brand revenue management. So the architecture treats forecasts as lanes with provenance and brand programs as connectors:

Every forecast lane carries: source (brand_rm / gm / rpp_model / ownership), brand, program (display metadata only: SMART, RMCC, Concerto), as-of date, and feed type (email parse, upload, manual). The model never hardcodes a brand concept; "Marriott SMART" is a badge, not a schema.
A brand integration is just a parser normalizing that brand's recap/forecast into the lane shape. Marriott SMART is first (the recap emails already sit in the inbox, e.g. Spark msg 179768); IHG follows when wired. Email is the universal API; portals have none for franchisees.
Decisions locked: planning lanes normalize to monthly (near-term daily brand detail flows to Pace, raw recap file attached to the lane); the GM budget stays the single plan of record that everything grades against, with brand RM as an input lane that can inform but never silently become the budget.
The payoff for management companies: one pane across flags (GM budget vs brand RM forecast vs RPP model vs actuals in one shape), plus forecast accuracy grading per source once actuals land, something no brand will ever build across its competitors' flags.

Make it whole

~1.5-2 weeks (re-estimated after review: the RGI and suggest items are math changes, not display fixes)

Rebuild budget entry (decided): a "Promote to GM Budget" action that freezes the current working forecast as the budget scenario, inline month editing for tweaks before approval, and a "Set budget = last year" one-click seed for properties without a modeled plan. Kills clunk #1.
Status + lock (version-control option A): draft / submitted / approved / locked with a visible badge.
Trust fixes: one RGI definition everywhere (ratio-of-sums, monthly-exact), occ lever stays in points labeled "+3.0 pts" (round-2 reversal), a visible warning instead of silent $0 revenue, a save indicator, the two-lane forecast split ("GM Model Forecast" for the computed lane; "Brand RM Forecast" reserved for the real parsed feed), and the auto-suggest rebuilt per-month raise-only with visible reasons.

Version control + the annual cycle

re-estimated ~3-4 weeks, split into three independently-shipped slices: versions + backfill, roles, season workflow

Immutable plan versions (option B): approval freezes a labeled snapshot; reforecasts open a new draft; the Scenarios tab becomes the version-compare view. Matches the industry "one model, three views" standard. The forecast-lane provenance fields ride along in this same migration (source, brand, program, as-of, feed) so nothing brand-specific calcifies.
Monthly reforecast ritual in-app: variance over 5% requires a one-line note that persists (institutional memory, the thing Excel never keeps).
Role scoping RLS before any GM gets a login (decided): GMs edit their own property's draft, ownership approves and locks, Ronald/Junaid hold portfolio staff edit, locked plans read-only for everyone.
Budget-season workflow (decided): the 2027 season runs in-app starting September 2026 with a lock by early December; dry run is a 2026 mid-year reforecast on the 3 pilots. Ronald's email cycle becomes an RPP flow with deadlines and per-property completion status.
Revive the 4100 sync (decided): on approval, the locked rooms-revenue budget pushes automatically to GL account 4100 in the budget/BVA system. Sales Plan stays rooms-only; expenses stay in the budget module.

Differentiators

sequenced by demand

Brand connectors (decided: start after P1 lock ships): the Marriott SMART recap parser first (built against the real emails in the HOUZN thread history), then the IHG parser, which follows the same pattern since IHG also delivers an email recap (Ace, 2026-07-19). Plus forecast accuracy grading per lane, visible to ownership and GMs, once a few months of actuals accumulate: "your brand RM's forecast has run N% hot" is the feature that sells this to management companies.
Brand-RM export: one-click monthly occ / ADR / RevPAR budget sheet in the shape each brand's RM program asks to receive (One Yield for Marriott, IHG's equivalent). One canonical budget, per-brand rendering.
Market-events overlay: the schema already stores events (market_context: Rodeo, World Cup, the Apple/Foxconn facility for HOUZN) with zero UI. Surface them on the Forecast grid as month annotations with suggested bumps.
Group lane (decided, instead of full segmentation): one group/negotiated revenue lane with monthly goals for the DOS, replacing the half-dead "Sales Team Goals" sheets (one of which is HOUZN's blank file).
Portfolio rollup: all 15 properties, revenue / RGI / variance vs plan on one owner screen. This is the screen that sells the product to other operators.

Bottom line: the hard, differentiated part (STR-anchored forecasting with share targets) already exists and works. Landing the feature is mostly plumbing: budget entry, lock, versions, roles, and a clean annual workflow. That set turns the December Google-sheet ritual into the product, and it is the same set that makes it sellable beyond THM.