More time for customer decisions
95% less processing time across the five delivered workflows, freeing specialist time for customer and commercial decisions.
AI workflow design
AI-enabled specialist workflows
~18,000 projected banker hours freed up annually
Across the document and analysis workflows.
95% less processing time
Across all five delivered workflows.
5 workflows in 7 months
Designed and built independently.
$0 incremental token spend
Built on existing enterprise AI tooling.
Usage spend, rather than total delivery cost.
- Problem
- Manual extraction, checks and calculations consumed hours before specialists could act for customers.
- Solution
- I turned domain expertise into reusable engines and AI workflows, with checks, review triggers and content rules built in.
- Outcomes
- 95% less processing time across five workflows, with checkable outputs and a document analysis pattern reused in another workflow.
- My role
- AI product ownership and consulting; workflow, product and content design; engineering, change and capability building.
What changed
Outcomes across the workflow
Consistent, checkable analysis
The extraction, processing and calculation pattern was reused across workflows, with standard methods and traceable review.
AI confidence
2× increase in users rating themselves very confident with AI tooling.
Pilot self-assessment.
Intent to use
100% likely or very likely to use the tool.
Reported pilot intent.
Pattern 1 / Document analysis
From documents to collaborative outputs
My five use cases centred on two patterns: document analysis and a content design system for agents. The analysis pattern extracts and processes documents, classifies information and performs calculations. Its outputs are structured for people and AI to inspect, question and continue working with.
Extract and structure
Pull information from documents and structure it for analysis.
Classify and calculate
Apply rules and code, with AI interpretation where rules do not cover the case.
Validate and flag
Run checks, highlight exceptions and surface items for review.
Human + AI collaboration
Specialists inspect, question and refine the structured output with AI, keeping the final decision.
The decisions behind the patterns
The work before the decision.
Specialists had to extract, categorise, calculate and check information before using it in a customer conversation. The aim was to reduce that preparation while preserving a clear path for review.
Size the problem and choose the right AI solution.
I assessed what enterprise had already built, what the business was willing to afford and what we could deliver. I reused existing AI tooling, then built the engines and workflow harness needed for the task. One starting instruction could run the lead analysis through to a completed workbook, followed by human review.
Reduce AI risk by assigning AI the right jobs.
Known cases used explicit rules. Numerical checks supported the extracted information. AI interpretation helped where rules did not cover the case, with review guidance showing where attention was needed.
Make the result inspectable.
I designed a path from summary to review priority, method and supporting detail. A specialist could challenge an output, trace how it was produced and keep refining it in Excel.
Pattern 2 / Another AI use case
A guardrailed content design system for people and AI.
I designed a content system for crafting AI-generated messages and content for an agent. People defined each scenario’s templates, structure and editing mode in advance; the AI retrieved the matching content and followed those rules.
People defineSet the content rules
Define the scenario, message structure, templates and how much AI may change.
- Content system
Retrieve the template
Match the agent’s situation to the content and editing mode people defined.
AI followsCraft within the mode
Use wording verbatim, adapt within guidance, or draft within the required structure.
Agent contentProduce the message
Give the agent content shaped for the situation, within human-defined boundaries.
Conceptual flow. Piloted with real users. Internal templates and messages are not shown.
- Verbatim
- Use the specified wording unchanged.
- Guided
- Adapt within the scenario’s defined limits.
- AI draft
- Create a draft within the required content structure.
My role
From commercial opportunity to working product.
How the work took shape
- Feb–Apr 2026
- Lead analysis workflow developed and scaled.
- By Sep 2026
- Other workflows built in parallel and rolled out.
Typical rhythm per use case, with work happening in parallel across workflows.
About 1 day
Frame value
Value framed
I framed the problem, commercial value and where AI could help.
About 2 days
Build requirements
Requirements
I defined the workflow, content rules and human review points.
About 1 week
Prototype and test
Prototype tested
I built and tested the prototype with specialists.
About 1 month
Pilot and refine
Pilot refinements
I piloted the workflow, iterated on feedback and prepared it for rollout.
After the pilot
Roll out and reuse
Workflow live
I rolled out the workflow and reused what worked across other use cases.
Approximate timings vary by use case.
- AI product ownership and consulting
- I framed the commercial opportunity and where AI could add value, balancing reusable enterprise capabilities, business affordability and what we could build.
- Human–AI workflow design
- I decided which steps belonged to rules, code or AI, and designed the review and escalation points where a specialist needed to step in.
- Product and content design
- I designed reviewable outputs that fit specialists’ existing tools, and a content design system controlling how AI could use and adapt templates.
- Engineering
- I built the lead workflow end to end: calculation engine, classification, workflow harness and workbook output.
- Change and capability
- I built and tested alongside specialists, used feedback to refine the workflow and supported people to use and question the output with confidence.
How I built with specialists
From real work to reusable patterns
I designed and built these workflows with specialists, iterating in the flow of their work. Their feedback shaped the patterns, checks and review experience.
Discuss the work together
Understand the work
Work with specialists to understand their goals, constraints and how the task happens in practice.
Build and test together
Build the workflow, test with real cases in the working environment and refine the checks and output.
Review and refine
Use specialist feedback to improve the patterns and prepare them for broader use.
This public case uses high-level descriptions and original conceptual illustrations to respect confidentiality.