Platform/Forecasting
Forecasting

Forecasts you can defend —
and measure.

Manual control + AI assist + monthly accuracy report. Built for revenue managers who'd rather verify than trust.

Forecast = OTB + Pickup. The math is simple. The defensibility is the hard part — and it's where Peaqplus is built differently.

app.peaqplus.com/forecast/2026-03
FORECAST · MARCH 2026
v3 · DRAFT
v1v2v3
DayOTBLY · SPBudgetSmart F.Final
Mar 05142 / €18.4k128 / €16.2k140 / €18k146146
Mar 06156 / €21.1k149 / €19.8k155 / €21k162160
Mar 07168 / €24.5k158 / €23.1k170 / €23.5k171171
Mar 08124 / €19.2k142 / €22.4k155 / €22k138138
Mar 09118 / €17.8k138 / €21.2k150 / €21.5k132132
Change reason required: Weekend pickup underperformance, lowering Mar 8–9.
01
Three layers, not one model

Statistical engine. AI correction. Accuracy report.

Most forecast products are one number out of one model. Peaqplus runs three independent layers and lets you turn each one on or off (per-hotel setting). The statistical engine — Layer 1 — builds the per-hotel baseline every night. The AI correction — Layer 2 — adjusts the baseline using business context. The accuracy report — Layer 3 — measures, monthly, how well each layer did.

All three run on the 60-day horizon, on your data, refreshed every morning at 5:00. The hotel decides which tier to use; the choice is a setting, not a contract change.

The three layers
  • Layer 1 — statistical baseline (always on for Tier 1+) — bucketed by month, weekday, and booking-window position; recency-weighted; outlier-filtered; trend-corrected — rebuilt every night without downtime
  • Layer 2 — AI correction (Tier 2 only) — a frontier reasoning model reads OTB, last-year same-point, budget, pickup trend, and the multi-country event calendar; produces a bounded adjustment with a per-day explanation in your language
  • Layer 3 — accuracy tracking (always on for Tier 1+) — error metrics published across 7, 14, and 30-day horizons; raw vs AI side-by-side; the log is never pruned, so the long-term trend stays honest
  • Refresh cadence — every morning at 5:00, per hotel; 60-day horizon; the accuracy report runs on day 1 of every month
FORECAST PIPELINE · DAILY 5:00 AM
L1
Statistical engine
Per-hotel baseline, rebuilt every night
Tier 1+
  • ·Bucketed by month, weekday, booking-window position
  • ·Recency-weighted · outlier-filtered · trend-corrected
  • ·Rebuilds overnight without downtime
L2
AI correction
A frontier model reads context and adjusts
Tier 2
  • ·Inputs: OTB · LY same-point · Budget · pickup trend · events
  • ·Per-day reasoning in your language · bounded to ±30%
  • ·Daily refresh · 60-day forecast horizon
L3
Accuracy tracking
Monthly retrospective — how well did it actually work?
Tier 1+
  • ·Error metrics across 7 / 14 / 30-day horizons
  • ·Raw vs AI side-by-side
  • ·Log never pruned — long-term trend stays honest
60-day horizon · refreshed every morning · accuracy report monthly
02
The forecast you actually edit

Three modes. Versions. Audit trail.

The auto-engine is the foundation, but the committed forecast is human work. Choose how you forecast: segment-level (pickup nights + ADR → revenue auto-calculated), daily (per-day pickup directly), or combined (segment × day matrix). On daily mode, the Smart Forecast Enhanced pre-fills every pickup cell — the team starts from the AI's best read and finishes by hand.

Every forecast has versions — v1, v2, v3 — and from v2 onward, every change requires a reason. "Pickup underperformance." "Price adjustment." "Group booking change." Pick from the dropdown or write your own. The full version history stays attached to the month, every reason searchable.

The editor workflow
  • Five statuses — DRAFT (editable) · COMMITTED (visible everywhere) · LOCKED (month-end frozen) · SUPERSEDED (newer version replaces) · EXPIRED_DRAFT (no commit before month-end)
  • Autosave every 30 seconds — nothing lost; optimistic locking surfaces conflicts as a 409 dialog rather than silently overwriting
  • Smart Forecast Reload button (on daily DRAFT) — one click refreshes the AI pickup baseline while preserving manual overrides
  • View switcher — Table or Calendar; the calendar makes 30-day pickup edits read at a glance
  • Stale snapshot banner — when fresh OTB lands during a session, the banner offers to refresh the DRAFT to today's data
app.peaqplus.com/forecast/2026-03
FORECAST · MARCH 2026
v3 · DRAFT
v1v2v3
DayOTBLY · SPBudgetSmart F.Final
Mar 05142 / €18.4k128 / €16.2k140 / €18k146146
Mar 06156 / €21.1k149 / €19.8k155 / €21k162160
Mar 07168 / €24.5k158 / €23.1k170 / €23.5k171171
Mar 08124 / €19.2k142 / €22.4k155 / €22k138138
Mar 09118 / €17.8k138 / €21.2k150 / €21.5k132132
Change reason required: Weekend pickup underperformance, lowering Mar 8–9.
03
Forecast Context Bar

Eight tiles. While you edit.

Manual forecast editing without context is gambling. The Forecast Context Bar — a sticky header above the editor on daily mode — surfaces every signal the decision needs: how much time is left, how much is on the books, what last year did at the same point, where the booking pace is, the budget gap, the current competitor rate, what previous versions said, and how your manual number compares to the Smart Forecast pickup.

The team doesn't have to leave the editor to check anything. Every tile is live; every tile refreshes when the snapshot updates.

The eight tiles
  • Time pressure — days left in the month, days to check-in
  • On the books — current OTB for the month
  • Last year, same point — week-position-aware YoY
  • Booking pace — vs last year at the same point in the curve
  • Budget gap — current forecast vs plan, live what-if
  • Competitor rates — public-data refresh + your relative position
  • Version movement — how many pickup cells changed vs prior version
  • Smart pickup — the AI pickup baseline vs your manual entry
app.peaqplus.com/forecast/edit · daily · v2 DRAFT
CONTEXT BAR · LIVE
While you edit, the signals refresh.
Time pressure
11 days
remaining in month
On the books
684 RN
−12 RN since yesterday
LY same-point
712 RN
−3.9% vs LY
Booking pace
−4.2%
vs LY at this point
Budget gap
−32 RN
forecast vs plan
Competitor avg
€152
your position: −4.6%
Version movement
7 cells
changed vs v1
Smart pickup
+9
AI vs your manual entry
Stale snapshot: fresh upload landed at 09:14 — refresh the DRAFT
04 — AI assist, with brakes

A frontier AI model, calibrated on your data.

For hotels on the AI tier, Smart Forecast Enhanced corrects the daily 60-day pickup forecast each morning. The AI never works alone — it works on top of the per-hotel statistical baseline, with bounded adjustments and a written reason for every day.

Tier
What it uses
Best for
T0
Baseline
Quarterly statistical average
Default; no setup needed
T1
Enhanced statistical
Recency-weighted, outlier-filtered, segment-aware
Hotels with strong historical data, no AI
T2
AI-corrected
Tier 1 + a daily 60-day AI correction
Hotels on the AI add-on
Built-in guardrails
±30% deviation clamp
The AI cannot move a forecast more than 30% from the statistical baseline
Room-count cap
Forecasts can never exceed your physical inventory
AI in your subscription
AI usage is part of your Pro tier — no per-call billing, no metered usage charges
Structured output
The AI can only respond through one bounded contract — no free-text path
No-downtime refresh
The nightly rebuild swaps in atomically; the app never blinks
Guardrail audit log
Every safety trip is logged for review
05
Numbers, not promises

Most AI forecasts promise. We measure.

Every month, on your data, we publish a MAPE / SMAPE / MAE accuracy report. For each forecast horizon — 7 days, 14 days, 30 days — you see the AI's accuracy and the statistical baseline's accuracy side by side. You decide whether the AI earns its place.

No magic claims. No average-of-averages benchmarks from someone else's hotel. Just the numbers, on your numbers, every month.

What's in the monthly report
  • MAPE / SMAPE / MAE for AI corrected vs statistical baseline
  • Per-horizon breakdown (7d / 14d / 30d)
  • Days counted, zero-day exclusions
  • Per-day comparison if you want to dig in
app.peaqplus.com/forecasting/accuracy
MAPE · 12-week rolling
3 tiers · your hotel
T0T1T2 · AI
10%8%6%4%2%W1W2W3W4W5W6W7W8W9W10W11W12target ≤ 5%
May 2026: Tier 2 MAPE 6.1% · Tier 1 baseline 8.4% · improvement +27%
06
Forecast Advisor

Ask the forecast. In your language.

Pulse Chat carries a dedicated Forecast Advisor agent — a mentor persona, not a number-fetcher. It explains why the AI raised a day, where the next risk is, what event lines up with a spike, and where the manual forecast diverges from the AI baseline. The reasoning lands in the user's language on every answer.

Four dedicated tools give the Advisor access to forecast-specific data: the 60-day AI window, per-day correction reasoning (single-day deep, multi-day detailed, or multi-week summary), accuracy trends with raw vs AI side by side, and the event calendar. The Advisor's tool-routing was measured at 100% on a 13-question evaluation — meaning when you ask, you get the right data, not the wrong screen.

The Advisor's tools
  • Smart Forecast window — daily AI snapshot, raw-vs-AI delta, freshness timestamp
  • Correction reasoning — three depths: single day (full text) · 2–7 days (detailed) · 8–30 days (summarised)
  • Accuracy trends — raw vs AI side by side, choosable 7 / 14 / 30-day horizons
  • Event lookup — multi-country holidays, festivals, and in-house events on the date range
  • Terminology discipline — "Budget" = the plan only · "forecast" = your manual commitment only · "Smart Forecast" = the AI output only · the Advisor never blurs the three
  • Freshness disclosure — every answer cites when the data was generated; no false certainty
app.peaqplus.com/pulse-chat · Forecast Advisor
🔮
Forecast Advisor
preview · AI-assisted
Why is the forecast so high for August 20?
checking the AI's reasoning for Aug 20…

2025-08-20 · 41 room nights (raw: 32, AI: +28%)

National holiday. Last year on this date the pickup was +18; the long weekend pulls neighbouring source markets to a seasonal peak. The Budget is conservative for this day (32 RN); the AI raises it based on the historical pattern.

Freshness: based on the forecast refreshed this morning at 05:00 · generated 2025-04-12 05:03 UTC
Show me the whole week around itWhat other events are coming in August?How accurate was the AI in April?
terminology guard · "Smart Forecast" = AI output (not Budget, not manual commitment)
What you see on every forecast

Four reference points on every cell.

Every cell on the forecast tells you four things at once. The decision is yours.

01
Actual · OTB

The current on-the-books from your PMS. Live, snapshot-aware.

02
Last year · SP

Week-position-aware YoY — not date-aligned. Thursday week 17 to Thursday week 17.

03
Budget

Your plan, broken to the same granularity as the forecast (segment, daily, or combined).

04
Smart Forecast

The AI suggestion (add-on), with the same change-reason audit available if you accept it.

07 · A day in the forecast

Six conversations. Six forecasts.

The forecast isn't a once-a-month deliverable. It's a daily companion — the AI refreshes overnight, the team reads the Context Bar in the morning, the Advisor answers the why questions during the week. Here's what that feels like.

01
Morning sanity check

Open the Pricing Calendar at 08:00. The Smart Forecast pickup is loaded on every row. Red rows = forecast lags budget. Three days to pay attention to.

Calendar · 30-second scan
02
Mid-month edit

Open /forecast/edit. Smart Forecast pre-filled every cell. Context Bar shows LY same-point, current OTB, pace, competitor rate. Five minutes, two clicks to commit v2.

Manual + AI hybrid
03
Why is Aug 20 high?

Ask the Advisor. Response: "National holiday, LY same-day was +18 pickup, long weekend pulls neighbouring source markets to a seasonal peak. The AI suggests 41 vs the budget's 32."

AI reasoning, in your language
04
Accuracy check

"How accurate was the AI in April on the 14-day horizon?" Raw forecast 9.8% error, AI-corrected 6.1% — over a 30-day sample. The team's confidence in the AI is now calibrated.

Monthly accuracy report
05
Strategic prep

"What events are coming in August?" → multi-country holiday list, local festivals, in-house events. The team prices the affected dates ahead of the curve.

Calendar foresight
06
Smart Forecast Reload

After a fresh PMS upload, the stale banner offers a refresh. One click and the manual DRAFT picks up today's AI baseline — manual overrides preserved.

Hybrid reload
Signal → Decision → Action → Outcome

Forecast. Verify. On real-shaped numbers.

In our 45–60 minute walkthrough (length depends on how deep you want to go), we run Peaqplus on our live demo environment — a simulated property with data that moves day to day — and walk through a forecast end-to-end. We run a sample AI correction and show you what the monthly accuracy report would look like on your own data once you're live.

No setup fee. No commitment.