Locata
← All casesCase study · Deposit return

From 1,000+ candidates to a 100-location shortlist. In three weeks.

[ numbers pending Statiegeld Nederland confirmation ]

Statiegeld Nederland is rolling out the next generation of bulk deposit return points — large-format machines deployed at non-retail sites where volume justifies dedicated infrastructure. Scouting the first hundred locations would have taken a small team six to twelve months of fieldwork. Locata delivered it in three weeks, with per-location reasoning that survived host outreach and internal review.

Candidates evaluated
1,000+

Gas stations & transfer stations

Shortlist size
~100

Ranked, with reasoning

Time to shortlist
3 weeks

End-to-end, kickoff to handover

Scouting reduction
~10×

Versus manual fieldwork

01 · The challenge

Bulk machines need more than retail floor space.

The original deposit return network was built around supermarket-installed RVMs. Volume kept up while bottle returns flowed through grocery checkouts. Bulk machines change the geometry: they live at non-retail sites — gas stations, transfer stations, amusement parks, schools — where surface, power, and 24/7 access matter as much as catchment.

Scouting them by hand is slow, inconsistent, and prone to missing the commercial reality of each host. A site can look viable in BAG and Street View and still fail because the host won't commit to a 10-year contract. The team needed a screening layer that pre-qualified candidates on physical, regulatory, and commercial criteria in one pass.

Hard constraint

Surface ≥ 400 m²

Truck access + manoeuvring space

Hard constraint

Three-phase power

Within 50 m of the proposed location

Hard constraint

24/7 accessibility

No fenced-off industrial parks

Hard constraint

Host alignment

Operator footprint, lease length, brand fit

02 · The approach

The four-step methodology, applied to bulk return points.

Nothing bespoke. Same methodology we apply to grid substations, EV concessions, and retail expansion — different prompt, same scoring backbone.

  1. 01

    Defined the candidate universe

    Every gas station and transfer station in the Netherlands, sourced from public operator data and BAG. ~1,000 candidates after de-duplication.

  2. 02

    Enriched per location

    BAG, Kadaster, BGT (surface), CBS catchment, NDW traffic, operator footprint, Street View frontage. Standardised per candidate, no manual lookups.

  3. 03

    Scored with three models

    Claude, GPT, and Gemini ran the same scoring prompt — hard constraints first, then weighted scoring against catchment, host alignment, accessibility, and public-space sensitivity.

  4. 04

    Handed over a ranked shortlist

    GeoJSON + PDF report per location with the top three reasons, top two risks, and per-model confidence. Used directly for host outreach and internal sign-off.

03 · The results

Ranked, reasoned, ready for outreach.

The handover was a GeoJSON file with all enrichment fields, plus a PDF report per shortlisted location. The team went straight from delivery to host outreach — no re-formatting, no follow-up data requests.

  • Defensible to stakeholders. Every score traced to its inputs. No “the AI said so” conversations.
  • Wider funnel, sharper top. All ~1,000 candidates pre-qualified — not just the obvious ones a manual scout would have visited first.
  • Methodology reusable. The prompt and pipeline have since been adapted for glass containers, textile collection, and underground waste.
Top of shortlist · top 5 of ~100schematic
Gas station 0421 · Utrecht943/3
Transfer station 18 · Almere893/3
Roadside hub · Eindhoven873/3
Service area · A1 east of Apeldoorn813/3
Petrol plaza · Breda south732/3

Indicative names — real shortlist names withheld pending host outreach.

Quote pending approval

We're holding this space for a quote from the Statiegeld Nederland team, currently in approval. We'd rather show an empty frame than invent one.

Statiegeld Nederland

[ role + name TBD ]

What this means for your rollout

Different output, same backbone.

The Statiegeld engagement defined a pattern that now runs across five verticals. Different data sources, different scoring prompts, same auditable trail.

    Adjacent collection categories

    Glass · textile · underground · parcel lockers

    Same methodology applies — different scoring prompt. Bulk machines, glass igloos, and textile bins share most of the constraints (power, surface, access, host alignment).

    See deposit return →

    Bigger funnel, same shape

    EV charging concession scouting

    Scoring thousands of plankaart locations against a CPO's business case is structurally the same problem: hard constraints, weighted scoring, ranked output for outreach.

    See EV charging →

    Higher stakes, same pipeline

    Grid operator site selection

    Substation and transformer siting under congestion pressure is the most demanding application of the same methodology — public opposition risk in the first screening round.

    See grid operators →

Your rollout, scored

Bring a candidate set. We'll score it live.

30 minutes, online. The same methodology Statiegeld used, applied to whatever you're scaling.