How we work
A five-stage delivery method, a scoring model for deciding what to automate first, a deliberate level of human control for every step, and engineering standards that keep workflows running for years.
The Shapes Method
01
Five stages, from problem to working system.
Durations are indicative for a single workflow. Your proposal sets the real timeline once the scope is understood.
- 01
Days
Discover
Understand the problem, the current workflow, and the result you want.
Outputs
- Problem statement
- Stakeholders and owners
- Success measures, agreed
- 02
1–2 weeks
Map
Identify the steps, systems, exceptions, permissions, and opportunities for improvement.
Outputs
- Current-state process map
- Systems and data inventory
- Scored opportunities
- 03
Days
Design
Agree on the proposed solution, scope, deliverables, and success measures.
Outputs
- Solution design
- Human-review points
- Fixed-scope proposal
- 04
2–6 weeks
Build and test
Implement the workflow and test normal cases, errors, edge cases, and human review.
Outputs
- Working workflow
- Test results on real examples
- Runbook
- 05
Ongoing
Improve
Document the solution and discuss monitoring, adjustments, and ongoing support where agreed.
Outputs
- Handover and training
- Before-and-after measurement
- Next-step roadmap
Prioritisation
02
How we decide what to automate first.
Every candidate workflow is scored from one to five on five criteria. High totals with clear rules and accessible data go first; low totals are left alone, however exciting the technology.
Worked example · Supplier invoice intake
Total 22 / 25 · build first
- 01
Frequency
How often does it run? Daily work repays automation faster than a yearly task.
5 - 02
Time per run
How long does each instance take a person, including checking and correcting?
4 - 03
Cost of errors
What happens when it goes wrong: a typo, a missed customer, or a compliance issue?
4 - 04
Rule clarity
Can the steps be written down? Clear rules automate reliably; judgement calls need review.
5 - 05
Data access
Is the information digital and reachable through an API, export, or inbox?
4
Levels of autonomy
03
Every step has a deliberate level of human control.
We decide, step by step, how much the workflow does on its own. Most start at Review or Exception and move up only when the results justify it.
- L1
Assist
AI prepares drafts, summaries, or suggestions. A person does the work and decides.
e.g. A drafted reply waiting in the inbox.
- L2
Review
The workflow does the work. A person approves before anything leaves the business.
e.g. Invoice data extracted, approved by finance.
- L3
Exception
Routine cases run automatically. Anything unusual, uncertain, or high-impact goes to a person.
e.g. Standard enquiries routed; unusual ones flagged.
- L4
Automate
Fully automatic, with monitoring and logs. Used only for low-risk, well-understood steps.
e.g. Records synchronised between two systems.
Engineering standards
04
Built to keep running.
What separates a workflow that runs for years from one that breaks quietly in a month.
Separate environments
Development, staging, and production, so changes are tested before they reach live work.
Version control
Every workflow change is recorded, reviewed, and can be rolled back.
Tested on your data
Normal cases, edge cases, and failures tested with real examples before launch.
Errors reach a person
Every workflow reports failures to a named owner with the context to fix them.
Least-privilege access
Each connection uses only the permissions it needs, stored in encrypted credentials.
Documented handover
A description, runbook, and owner for every workflow, so nobody depends on us.
Next step
What would you like to improve?
Tell us what is taking too much time, what is difficult to manage, or what you would like your systems to do differently. You do not need a technical specification to start.