01 / Prepare and do the work
Assist
Help people prepare drafts, fill in details, and complete repetitive tasks. They review and refine the results.
For example: turn meeting notes into a draft report, or use a document to help fill in a form.
Custom AI workflow development
AI built around
the way you work.
We get to know your people, daily habits, processes, and problems. Then we design and build AI workflows and specialised agents to help work get done with less friction.
What we build
Four ways AI can
help your people.
A workflow connects the steps, people, and information needed to get work done. We build AI into those steps through four pillars, choosing the right combination for your organisation.
02 / Check the work
Verify
Check work against your requirements and source information. Flag missing details, inconsistencies, and possible errors for a person to review.
For example: compare a report with its source data before it is shared.
03 / Understand the information
Analyze
Help people explore company information, spot patterns, and understand what needs attention so they can make informed decisions.
For example: see which projects are falling behind and explore what is holding them up.
04 / Find the guidance you need
Guide
Answer questions using your organisation’s documents, policies, and guidance. Help people find the right information and know what to do next.
For example: ask how to complete a process and get instructions with references to the source.
Your people set the rules, review results, and remain responsible for decisions.
One example / ADX Asia
What does this
look like in practice?
In ADX Flow, a payment request connects the person submitting it, the people reviewing it, and Finance—with the supporting information and AI help in the same system.
See how these four pillars support work across ADX Flow, with examples from payment requests and expense claims.
Follow a payment request through ADX Flow ↗How we work
Understand it. Build it.
Keep improving it.
Identify and define the problem.
Learn how people work today: their habits, responsibilities, tools, and processes. Find where time is lost, work is repeated, or errors occur. Agree on the problem and what improvement should look like.
Design the new workflow.
Work through a simpler way to get the job done. Decide where AI can help, what people will do, and where human review or judgment is needed.
Design the AI solution.
Plan the data, agent instructions and tools, runtime, model, and deployment. Make the tradeoffs around your organisation’s privacy policies, cost, speed, and accuracy needs.
Build and test it.
Build the workflow and specialised agent, connect the required systems, and test normal cases, errors, and exceptions using representative work.
Document it.
Record how the workflow works, what the agent can do, how people use and review it, and what to do when something needs attention.
Deploy and hand it over.
Put the solution into use, train the people working with it, and make ownership and support responsibilities clear.
Monitor, evaluate, and improve.
Review the agent’s effectiveness over time. Use feedback and observed results to refine its instructions, knowledge, and behaviour, and improve the workflow as the organisation’s needs change.
Designed for your organisation
The right choices
depend on your work.
Privacy
What information can AI use?
Your policies guide data access, where information is handled, who can see it, and what the agent is permitted to do.
Cost
What does useful work cost?
Consider model use, infrastructure, maintenance, and human review alongside the value of the completed work.
Speed
How quickly must it respond?
Choose a setup that suits the task, whether someone is waiting for an answer or a longer process runs in the background.
Accuracy
What does a correct result require?
Define checks, evaluations, and human review around the consequences of getting the work wrong.
Inside the solution
What we put
in place.
Each part supports the workflow. The choices follow your organisation’s requirements.
- Data
- The company information the agent needs, organised so it can find and use the right material with appropriate access.
- Harness
- The instructions, tools, permissions, and checks around the AI: what it can do, how it does it, and when it asks a person for help.
- Runtime
- The environment where the agent runs, connects to your systems, keeps track of work, and handles interruptions or errors.
- Model
- The AI model selected and evaluated for the task, balancing your privacy requirements, cost, speed, and accuracy.
- Deployment
- How the solution becomes available to people, with access controls, documentation, monitoring, and a clear handover.
Our philosophy
More help from AI.
Control stays with people.
We define what the agent may do, which results people must review, and who owns each decision. Effectiveness is evaluated against the work: time taken, total cost, accuracy, and the corrections people need to make.
Talk to Kihon
Let’s look at one workflow.
Tell us about a process that takes too long, involves too much repetitive work, or is difficult to get right. We’ll start by understanding it.