Impact Calculator and rollout planner · PackBack Network
From a spreadsheet and a guess to a proposal in two days.
An honest impact estimate on the website for prospects, and an internal planner where AI suggests where machines should go and turns it into a proposal. Sales went from weeks of spreadsheets to a two-day turnaround.
- Role
- Product Lead, lead designer
- Company
- PackBack Network
- Timeline
- Dec 2025 to Feb 2026
- Team
- 1 product designer, 2 engineers, sales lead, ops lead
- Platforms
- Website, Internal web tool, PDF proposals
1Estimates come from similar real locations
2Every assumption can be opened and edited
3A low to high range sits next to the single number
15 → 2 days
Time to send a proposal
2.3×
Demo requests from the website
74%
AI placement suggestions kept by ops
+9 pts
Proposal win rate
The problem
Every prospect asked the same three questions: what will it cost, what will we save, and where do the machines go. Sales answered them in Excel after a site visit, which took two to three weeks per proposal.
What I did
I designed a public calculator with editable assumptions and ranges, and an internal planner where sales upload a floor plan and AI suggests machine positions from footfall at similar locations. Both feed one proposal template.
What changed
Proposals now go out in two days instead of fifteen. Demo requests from the website went up 2.3×, ops kept 74% of the suggested positions, and win rate rose nine points.
01 · Context
Selling a system people can't picture yet.
PackBack sells a subscription that includes machines, cups, washing and software. For a campus or a restaurant chain, it replaces a cost they know well, disposable cups, with one they've never seen. Every sales conversation started with the same questions, and answering them well took time we didn't have.
- 01Website visitGeneric 'book a demo'
- 02Intro callCollect numbers by email
- 03Site visitWalk the space, take photos
- 04Excel modelCopied from the last deal
- 05Proposal15 working days on average
Where most of the time went
02 · Problem
Slow answers, and numbers nobody could check.
The Excel model had grown over three years. Each salesperson had their own copy, the assumptions were hidden in cells, and prospects didn't trust a single 'you will save €48,000' number. Machine placement was decided by whoever did the site visit, and ops often moved machines after installation.
Before
- ✕One savings number, no range
- ✕Assumptions hidden in Excel
- ✕Placement guessed on site
- ✕Proposal written by hand each time
After
- ✓A likely range, with low and high
- ✓Every assumption visible and editable
- ✓Placement suggested from real footfall data
- ✓Proposal generated from the plan
The question we worked on
How might we give prospects an estimate they believe within minutes, and give sales a plan good enough to send without a site visit?
03 · Research
Listening to sales calls.
I listened to 20 recorded sales calls, interviewed 6 customers about how they decided, and compared planned versus actual machine positions at 40 existing locations with our ops lead.
Buyers need to defend the number internally
The person we talk to rarely signs. They need a number they can explain to finance, with assumptions they can change.
“If I can't show how you got there, my CFO won't read past page one.”
Placement decides return rate
Locations where SmartBins were within 15 metres of the main exit had return rates nine points higher than the rest.
“Put the bin where people leave, not where you have space.”
Similar sites behave alike
Campus food courts, station kiosks and office canteens each had clear patterns. We had the data to use them, just not the tool.
“Every campus I visit looks like the last one.”
04 · Process
Nine weeks, two tools, one model.
The key decision was building one shared impact model first. The public calculator, the internal planner and the proposal all read from it, so a prospect never sees one number on the website and a different one in the PDF.
Gated PDF after a form
Marketing's first idea. It collects leads, but prospects told us they distrust numbers they can't play with.
Single savings number
Clear, but people anchored on it and felt misled when the real proposal differed.
Range with editable assumptions
A likely range, with every assumption visible and changeable. Fewer surprises later.
05 · Design decisions
Honest numbers, suggested placement.
Every assumption is visible and editable
The calculator shows where each number comes from: the prospect's input, the average of similar PackBack locations, or an industry figure. Each can be changed. A small sensitivity chart shows which assumptions matter most, so buyers know what to check first.
AI suggests where the machines go
Sales upload a floor plan or a phone photo of one. The planner marks exits, counters and seating, then suggests positions for dispensers and SmartBins using footfall patterns from similar locations. Each suggestion has a reason and a confidence level, and ops can accept or move it.
The plan becomes the proposal
Once placement is accepted, the proposal writes itself from the same model: machine list, monthly cost, expected savings and a site plan. Sales can edit the contract length and deposit, and the numbers update everywhere.
06 · Edge cases
When the model shouldn't guess.
| Situation | What the tool does |
|---|---|
| Venue type with fewer than 10 similar locations | Shows a wider range and says the estimate is based on few sites |
| Floor plan photo is unclear | Asks sales to mark exits and counters by hand before suggesting |
| Prospect enters very high volumes | Flags it and suggests a pilot at one location first |
| Assumption edited beyond realistic limits | Accepts it, marks it in the proposal as 'set by customer' |
07 · Outcome
Faster proposals, and buyers who trust them.
“I changed the return rate to something pessimistic and it still made sense. That's what convinced our finance team.”
“I can send a proposal the same week as the first call now.”
08 · Reflection
What I learned.
Showing a range felt risky to marketing. It turned out to be the most persuasive part of the calculator, because it showed we weren't hiding anything.
Next time I'd connect the planner to PackPulse from the start, so the model learns from every new location automatically. We added that link in the spring.
Tools on this project
Figma, FigJam, React prototypes, Claude API, Google Sheets for the first model, Gong call recordings
Names of the people quoted have been changed to protect their privacy.
Atlas Copilot →