What a pilot looks like
Fruitlah: a group-buy agent that fills boxes and reconciles payments, while people make the calls.
See how it works →For Business
Paper & Pixel builds tailored AI solutions for small and medium businesses, starting with one recurring workflow.
We help you find one useful opportunity, map how the work actually happens and build the smallest solution that makes it easier — a focused assistant, a guided workflow or a light integration, built around your context, rules and existing tools.
Quick facts
- Offer
- Starter Workflow Pilot
- Price
- From S$1,200
- Scope
- One recurring workflow, done with you
- First step
- A short workflow call
Starter Workflow Pilot · done with you
From S$1,200
A done-with-you implementation for one recurring workflow — deliberately small enough to map, build and test properly before either of us commits to something larger.
Best for a clear, repeated workflow with a small number of people and straightforward tools.
- Opportunity and workflow mapping for one recurring task
- One focused assistant job or similarly narrow solution
- One simple tool connection, sometimes two where the workflow is straightforward
- Permissions and approval points designed into the workflow
- Testing with representative examples, including incomplete and risky cases
- Operating notes and handover, including limits, what to do when something goes wrong and a short support window to fix obvious issues and refine instructions
- First step
- A short workflow call to understand the task and decide whether a pilot is a sensible fit.
- Scope
- Confirmed in writing before the build begins. More complex permissions, tools or team needs are quoted separately.
You do not need a technical brief. Bring the recurring task, how it works today and a few representative examples.
Probably a fit if
- The task happens often enough to matter
- You can show us how the work is done today
- The context and rules can be made clear
- You can provide representative examples for testing
- A person can review important outputs and actions
- You are willing to begin with one narrow job
Not a fit — and not what this is
- The process changes completely every time
- There is no clear owner for the workflow
- Several workflows at once, or a large transformation programme
- Unsupervised decisions about customers, money or official records
- A promise to replace people or professional judgement
- A black box handed over without operating guidance
Prefer to learn the method yourself? Explore the workshops →
From opportunity to working first version.
The process stays close to the people who know the work. Your examples and judgement help us make the solution useful; clear boundaries keep it dependable.
- 01
Identify the opportunity
The repeated task, the friction around it and the outcome that would genuinely help.
- 02
Map the workflow
Inputs, context, rules, tools, approval points and common exceptions.
- 03
Scope the smallest version
One clear job, and what the first version will — and will not — do.
- 04
Build
Only the connections needed, the least access required, approval on important actions.
- 05
Test real cases
Normal, incomplete, ambiguous and stop-and-ask examples before relying on it.
- 06
Hand over clearly
Operating notes, known limits and guidance for monitoring the solution.
The simplest structure that fits — with clear limits.
The right format depends on the job and its level of risk.
Help with one well-defined job
An assistant works with approved information and instructions to organise incoming work, summarise what needs attention, draft updates and responses for review, and flag missing details or exceptions.
Connect useful steps where it makes sense
A light integration reads from a clear source of truth, prepares structured records and drafts, surfaces a review queue and records approved actions — with logs and checkpoints where they matter.
Who does what.
Permissions follow the consequence of an action: preparing work is low risk, acting for the business needs a much higher bar.
AI prepares
limited and reversible
- Read approved information
- Organise and summarise
- Draft options
- Flag missing context
People review
a person stays accountable
- Resolve ambiguity
- Handle exceptions
- Correct the output
- Decide what happens next
Approval required
never on an AI suggestion alone
- Customer messages
- Spending or moving money
- Official records
- Actions affecting reputation
Questions before we begin.
Do we need to know which AI tool we want?
No. Start with the workflow and the outcome you need. We recommend a suitable structure only after understanding the context, rules, tools and level of risk.
Do we need to know how to code?
No. You need someone who understands the workflow and can test the solution with realistic examples. We handle the agreed technical implementation.
Can a solution work with our existing tools?
Often, yes, but we do not assume every connection is useful or safe. We first check what access is possible, what information is needed and whether a simpler approach would work better.
How do you handle access and sensitive information?
We start with the least access the workflow needs and avoid unnecessary data. Read-only or draft-based steps may come before write access. Work involving sensitive information, several staff members or higher-consequence actions needs additional scoping.
Will it act automatically?
Not by default. A first version usually prepares, drafts or flags work for review. Anything involving customers, money, official records or reputation requires explicit human approval.
What happens after the pilot?
You can continue using the first version, refine the same workflow or discuss another layer. We only recommend expanding after the pilot has proved useful in real work.