HoldOff

Four research reports, each on one subject

1. The evidence on what predicts smoking, which set the data model.
2. Interface and visual design, which set the second release’s design.
3. The psychology of smoking and cessation.
4. The use of AI in smoking cessation and in cessation-plan design.

The first two shaped the second release. The last two shape the AI features now in design.

A quick, low-cost validation of the logging engine

I took a spec to a tested, deployed app in under two days and released it the same day to 18 regular smokers recruited as testers. It was a deliberately quick, low-cost check of whether they could engage with the logging engine and whether it addressed their problem, and it was never meant as a designed product. The same 18 testers stayed on for the second release.

Designed before the build, with a rebuilt data model

Five days after the first release, the second was the first meant to be used as a product, so I designed every screen before the build and rebuilt the data model from the research.

Logging

One tap logs an entry with its time and its place in the waking day. Four quick questions follow, one per card, and any can be skipped by closing the sheet. Optional questions sit behind a separate link.

Questions built from evidence

Each question earns its place from the research. Urge strength was the single strongest predictor; triggers, others smoking nearby, and coffee, food and alcohol followed. Where the two research reports disagreed, I settled the conflict on the record. One report wanted the previous context pre-filled silently and the other said never to preselect answers, so a one-tap “Same as last time” card settled it and stale context is never carried forward unnoticed.

Rules for the questions

Each question answers one distinct thing, each option points to a different action, nothing recorded automatically is asked, and no question has more than six options, so the options fit a phone screen as thumb-sized tiles.

Metrics that point to an action

Hold-off time after waking is the headline. Craving-driven against autopilot smoking, shown as counts of the day’s total, tells a user whether the fight is with cravings or with routine. The clinical Heaviness of Smoking Index appears as a weekly trend and never as a label, and hold-off is measured against clinical bands.

Tone

A warm and honest coach. Real wins are named specifically, setbacks get empathy and a next step, and nothing is shamed or faked. There is no confetti, no badges and no streaks that reset. The model is Jürgen Klopp’s way with players. The research supports it, since rising confidence predicts quit attempts, and shame and a drop in confidence after a slip predict relapse.

Scope

Behaviour-change mechanics, resisted-urge logging and a structured reduction plan were held back to a later release, so that the effect of each change could be attributed. To save time, no compatibility with earlier data is kept during the MVP releases; it begins with a production product.

Privacy and security

A strict security policy, nothing sent over the network after loading, damaged or wrong import files refused with the data unchanged, and a rule for any third-party asset: secure, free to use, legal, needing no disclosure to users, and respecting privacy.

Testing the second release with the test group

I planned validation around testers’ exports and answers. Testers gave feedback instead, so I moved the method to collated tester feedback as the input. Every completed day in the exported data reviewed carried both its start and end markers. Validation closed with no fix release needed, and four findings carry forward to the next release:

• The progress ring’s colours read as go and no-go, and the day-start and day-end keys were tapped by accident.
• Half logging was unused across the test group (0%), so it is being removed.
• History should show the last seven days, with each past day marked as met or missed.
• Charts need axes and counts, so they can be read at a glance.