Product Requirements Document
Your triggers, your stakes, your tools — an AI companion that knows your story and gets sharper every time you use it.
Prepared for
Conrad Tracey
Addiction Coaching Australia
Prepared by
Rossco Paddison
A Player Labs
Date
16 March 2026
01
People struggling with addiction need support in real-time crisis moments — when cravings hit, when triggers fire, when the war in their head is loudest.
Your coaching methodology works one-on-one, but it doesn't scale. Your frameworks — the Not-To-Do List, the 5R Strategy, Anchors & Accelerators, the Filters — live in your head and in Loom videos.
When a client is at 2am staring at a decision that could destroy their life, they can't call you.
But recovery isn't only crisis moments. Between the cravings, there's reflection, pattern recognition, learning, and reinforcement. Your best clients don't just survive the cravings — they build an ever-deepening understanding of their triggers, their stakes, and their decision patterns.
The Gap
There is no self-guided tool that learns a person's specific triggers, provides living personal tools they actively maintain, walks them through personalised crisis intervention, and gets smarter about them over time.
02
Layer 1 — One-Time
A guided 15–25 minute setup that teaches your frameworks and seeds the user's personal toolkit.
5 steps · Interview · Not-To-Do List · 5R Strategy · Anchors & Accelerators · Filters
Layer 2 — Persistent
A set of living, user-managed tools that grow richer over time through use and reflection.
5 tools · My Not-To-Do List · My Triggers · My Filters · Cravings Log · Triggers Log
Layer 3 — Repeatable
An AI-guided crisis intervention that reads from the toolkit, follows your protocol, and writes back into it.
8-step protocol · AI-driven · Personalised to each user
Layer 4 — Anytime
Video-based educational content that deepens understanding of the frameworks outside of crisis.
3 modules · Breaking Cravings · Identifying Triggers · Win the War
The app is a learning system. Every crisis session enriches the toolkit. Every toolkit update makes the next crisis session more personalised. The user's recovery journey is captured, tracked, and reinforced — not just survived.
03
A guided sequence that teaches frameworks and captures personal data. The user learns your system while simultaneously populating their toolkit.
Understands the person — substance/behaviour, history, goals, situation.
Seeds → User profile context
Surfaces the negative outcomes of poor choices — more psychologically motivating than positive goals.
Seeds → My Not-To-Do List
Teaches the craving-breaking framework. Video, interactive, or AI-delivered — format TBD.
Installs knowledge — no data captured
Identifies personal triggers — people, places, situations that anchor them to the behaviour or accelerate relapse.
Seeds → My Triggers
Installs the decision-making matrix: "Does this serve my mental/physical health? My finances? My relationships?"
Installs knowledge — no data captured
~15-25 minutes. Conversational, not clinical. Progress saved at every step — if they drop off at step 3, steps 1-2 data is retained. Can be resumed anytime.
04
These are not just database records — they are first-class product surfaces the user interacts with directly. Living tools they own, review, edit, and grow over time.
Personal list of negative outcomes the user wants to avoid. The consequences of poor choices — jail, divorce, job loss, health collapse — in the user's own words.
Created at onboarding · Editable anytime · Read by Cravings Killer
Personal catalogue of anchors and accelerators — people, places, situations, emotions — categorised by type.
Created at onboarding · Updated by CK write-backs · Fed by Triggers Log patterns
Personal decision-making matrix: "Does this serve my mental/physical health? My finances? My relationships?"
Learned at onboarding · Viewable anytime · Used by Cravings Killer
Historical record of every Cravings Killer session — date, severity in/out, reduction %, context, duration. Shows trends over time.
Auto-populated by every CK session · Shows on home screen
Raw observational record from CK sessions — what trigger fired, when, where, who. Surfaces patterns that feed My Triggers.
Auto-populated by CK · "3 cravings on Friday nights alone" → prompts trigger update
Key Distinction
My Triggers = the curated list. "I know Dave's house is a trigger."
Triggers Log = the raw data. "March 5th, 9pm, home alone, severity 8."
The Log feeds the List. The system connects the dots.
05
AI-guided crisis intervention. Reads from the toolkit, follows your protocol, writes back into it.
Free text + severity score (1-10)
→ Writes to Cravings Log
Surfaces the user's negative outcomes. Reminds them what's at stake in their words.
← Reads from My Not-To-Do List
→ May prompt: "Add to your Not-To-Do List?"
Guide through the craving-breaking strategy. Check: Hungry? Angry? Lonely? Tired?
Check environment against known triggers. If trigger detected, suggest an environmental change.
← Reads from My Triggers
→ Writes to Triggers Log · May prompt: "Add to your triggers?"
Mental/physical health? Finances? Relationships? Binary yes/no for each.
← Reads from My Filters
Compare to step 1. Target: 50% reduction.
→ Writes final severity + reduction % to Cravings Log
Affirm the choice. Build confidence. Reference progress trend if available.
← Reads past sessions from Cravings Log
Collect feedback. Rating (1-5) + optional comment.
→ Writes feedback to Cravings Log
Write-Back Loops
These are what make the app a learning system. Every session captures new data — new triggers discovered, new not-to-do items surfaced — and writes them back into the toolkit. The app gets smarter about the user with every use.
06
Week 1
Toolkit has 5 not-to-do items, 3-4 triggers. Thin but functional. First CK sessions feel somewhat generic.
Week 4
8 sessions in. 3 new triggers and 2 new not-to-do items added through write-backs. AI conversations are noticeably more personalised. Severity trending down.
Month 3
"You've had 4 cravings on Friday evenings recently." The app is now proactively defensive, not just reactive.
Month 6+
Rich toolkit. Long-term trends — frequency declining, severity declining, sessions shorter. The user can see their own progress.
07
Mode 1
"I'm having a craving right now."
One tap to Cravings Killer. Zero friction.
Must be reachable in under 2 seconds from app open.
Mode 2
"I'm doing well. I want to strengthen."
Toolkit management, progress tracking, learning.
Reflection, growth, reinforcement.
The crisis button is always visible and fast. But it's not the only thing on the screen. Below it lives progress, quick access to personal tools, and learning content. Both modes served without one blocking the other.
08
These are specific content deliverables required before the corresponding features can be built.
Full list of what you ask in the initial session. Numbered list with expected answer types.
Onboarding step 1
Prompts, structure, example walkthrough. How do you guide someone through this?
Onboarding + Toolkit
What are the 5 R's? Full teaching content. Written framework or video recording.
Onboarding + CK
Categories, example triggers, capture format. How do you categorise these?
Onboarding + Toolkit
Beyond health/finances/relationships — is there more nuance?
Onboarding + CK
Full definition and how it integrates with 5R. Sequential or user-chosen?
CK step 3
Existing or new recordings for the 3 Learning modules. Video files or hosted links.
Learning layer
Colours, fonts, logo, visual identity. Whatever you have — PDF, style guide, or just direction.
Design system
09
Key decisions that need your input before or during the build.
Is the Cravings Killer a free-flowing AI chat conversation, or a structured wizard with UI controls — sliders, big buttons, yes/no taps? Someone at 9/10 with shaking hands — are they typing into a chat, or tapping large buttons? This is the single biggest architectural decision.
Severity rating could be a slider. Filters could be three yes/no buttons. HALT could be four checkboxes. Or all of it could be conversational. The split between structured UI and AI conversation shapes the entire build.
"I'm having a craving" is one mode. But what about "I just used"? That's a completely different conversation — shame, harm reduction, next steps. Does the app handle post-relapse, or only pre-decision?
Generic supportive voice, or trained on your actual speech patterns and phrases? If it's meant to feel like talking to you, we need transcripts of you coaching to train the tone.
At 9/10 severity, reminding someone "you could lose your kids" might tip them over the edge rather than pull them back. You said there are times you would and wouldn't surface it — that's a safety-critical design question.
Does a user set up for one substance or behaviour, or could they have multiple? A user might be fighting alcohol and gambling. Changes the data model, onboarding, and how the AI personalises.
At 2/10 do they really need all 8 steps? At 9/10 do they have the patience for 8 steps? The linear protocol might be wrong for both extremes.
Individual user journeys? Efficacy rates across all users? Most common triggers? Session completion rates? What would change how you coach if you could see it?
10
You review this document. Fill the gaps. We answer the open questions together. Then we start building.
Send through the content gaps — interview questions, 5R strategy, HALT, A&A categories
Send through your brand guide — colours, fonts, logo, whatever you have
Answer the open questions — GHL API, pricing, tier split
We jam on the plan, refine, and start building
Web app first (PWA) · App store later · 4-6 weeks to v1 staging