ai platform

forecasts,
in plain english.

scoop's ai is not a chatbot. it's a planner with ten years of operational data, that learns every night, and that answers in the language of the council meeting.

the ai platform

ask questions. get routes.

scoop's ai combines sensor history, weather and behavioural patterns. it answers in plain english. it plans in scenarios.

scoop ai · planner
live · v1.4
ask scoop something…
forecast per container
every point gets its own model, based on history, weather, season, events.
anomaly detection
unexpected fillings or dead sensors are flagged automatically.
conversational planning
frame scenarios in plain language. scoop answers with a map and a plan.
reporting & accountability
automatic monthly reports, co₂, kilometres, pickups, for the council and the board.
under the hood

six kinds of data in.
four kinds of plan out.

scoop's ai isn't one model, but a combined pipeline of time-series forecasting, gradient boosting and rule-based optimisation. what makes it special isn't the algorithm, it's what goes into it.

inputs
sensor history
every 15 min, 30+ days
weather & climate
forecast 7 days ahead
calendar events
markets, festivals, holidays
seasonal patterns
2 years history per container
driver reports
broken lid, side-placements
behavioural patterns per district
flat vs. terraced house vs. centre
scoop model v1.4
relearns daily
70% time-series30% boosting
outputs
7-day fill-level forecast
per container, with confidence
dynamic routes
automatic every morning
anomaly alerts
deviations detect themselves
plain-english answers
to scenario questions, with a map
under the hood

a dedicated model per container ,
and an uncertainty band that's honest.

scoop's forecasts are not averages. every point gets its own time-series model, trained on local history and adjusted with weather, calendar events and behavioural patterns. the confidence band shows where we're sure, and where we're not.

peak on sat·94%
crosses 75% this week
mon
tue
wed
thu
fri
sat
sun
0%25%50%75%100%58%64%71%78%86%94%88%saturday marketmontuewedthufrisatsun
75%, time to plan
per-container training
at least 30 days of history, every 15 minutes. the model remembers which containers behave differently from their neighbours.
retrained daily
every night the model rolls again. if a district changes (new flat, market moved), the model follows within a week.
confidence, not fables
the shaded band around the line is the 80% confidence interval. narrow = certain. wide = be careful.
scenario planning

frame a scenario in plain language.
get a full plan back.

no more model training, no scripts, no excel. type a question the way you'd ask a colleague, scoop combines sensor data, weather and planning rules and gives you a concrete plan, with kpi's, within 4 seconds.

scoop ai · scenario
live · v1.4
what if it rains next week?
rain lowers the fill level by an average of 8%. i'll shift 6 stops from thursday to monday and skip the centre district on wednesday.
plan ahead of the rain
proposed changes
  • mon 07:30kerkplein 12shifted
  • mon 08:15westwijk-ashifted
  • tue 09:00havenstraat 8unchanged
  • wed ,centre ringskipped
  • thu 08:45dorpsweg 8shifted
−12fewer km
3avoided stops
−34 mintotal time
anomaly detection

something off?
scoop notices it before you do.

sensors can die. districts change. events disrupt patterns. scoop continuously runs a layer of anomaly detection over every measurement, and acts on it before you see it.

critical
sensor is silent
detectionno update in 4h where there should normally be 16 → flagged as 'possibly dead'.
actionback-fill with modeled estimate + create maintenance ticket.
warning
unexpected rapid filling
detectionfill growth > 2× expected for 2 consecutive measurements.
actionadd to risk queue + propose an extra pickup within 24h.
informational
slower-fill than forecast
detectionactual fill level < 80% of forecast for 3 consecutive days.
actionretrain model; possibly propose reduced frequency.
informational
behavioural shift
detectionday-of-week pattern shifts structurally (new traffic? new flat?).
actionai adjusts seasonality weight automatically within 7 days.
a dead sensor is not a blind spot
scoop keeps computing, with confidence marked.

while a sensor is being replaced, scoop estimates the fill level based on neighbours, weather and historical patterns. in the dashboard you see a dashed line, you know it's an estimate, but you can plan with it.

ai governance

an ai you can rely on ,
and that you can switch off if you have to.

the public sector can't absorb ai mistakes. scoop is built with accountability as a starting point: daily retraining, no hallucinations and full traceability. plus: you can always fall back on fixed routes.

24h
cadence
retrained every night
at 02:00 the model rolls again with the latest 24h of data. no 'frozen' ai, no stagnation. drift is corrected before the morning briefing.
0
hallucinations
only from hard data, no fabrication
scoop only answers from measurements and verified reports. when it doesn't know, it says so. no creative filling-in, no 'probably'.
100%
traceable
every forecast has a story
click on a forecast → you see which data went in, which sub-model decided the outcome and with what confidence. no black box.
manual override, always.
one button switches the ai planner off and falls back to your fixed routes. no 'we need to call an engineer'.
what if we told you this is already reality?

measurably smarter.
proven at reinis.

0–0%
fewer kilometres
less
CO₂ emissions
faster response
0×
faster response to filling containers vs. fixed planning.
side-placements
−0%
fewer complaints and resident reports.
proven at reinis, waste collector for the municipality of nissewaard
plan a demo

let us show you the kilometres
you'll save.

a 30-minute live demo on your own district data. we bring a planner and a driver back-seat.

rotterdam · nl