ntwrk/
Book a call
OPERATING INTELLIGENCE

AI working inside your operation.

Start with one workflow your team needs to improve. Test the idea in a 30-day proof, then decide whether it deserves a production build.

Book a call A 30-day proof on one decision.
See how we work
01 / THE GAP

The demo worked. The workflow still needs work.

A useful demo leaves practical questions unanswered. Those questions belong in the assessment, which starts with the process, the people responsible for it and the measure that would make a change worthwhile.

The demoIt produced a good answer.

On the example it was shown.

Who checks the result? What happens when the data is missing? Where does the output go? Who fixes the workflow when something changes?
The assessment answers these before anything is built. Then we test one part of the work.

Left unanswered, those questions are where an AI pilot stalls before production. They belong in the assessment, which starts with the process, the people responsible for it and the measure that would make a change worthwhile. Then we test one part of the work.

General-purpose AI tools may already be the right answer. A dedicated workflow earns its place when it handles a repeated task well enough to justify the effort of running it.

02 / THE FIRST 30 DAYS

Test one workflow before funding the build.

The first step is to map the work, identify the decisions worth improving and rank the candidates by value, feasibility and risk. Together, we choose one to test against agreed cases and measures.

01Current-state map
02Ranked use cases
03Working proof, with test results
04 · DecideDecision pack

The recommendation may be to build, buy, change the process or stop.

The proof runs in your environment or, where access requires it, an agreed controlled NTWRK workspace. The scope records the environment and the deliverables to be transferred. The proof’s dependencies and known limits are documented with it. The proof is not production. If the evidence supports a build, we scope integration, access, monitoring, human review and handover as the next engagement.

Book a call A 30-day proof on one decision.
03 / EXAMPLES

Look for work that repeats, with a result you can check.

Example 01

Prepare a case for review.

Gather the relevant records, identify missing information and prepare a draft for a person to assess.

  1. Gather records
  2. Flag missing information
  3. Prepare a draft
  4. A person assesses
Example 02

Triage incoming requests.

Apply agreed categories, suggest the next owner and send uncertain cases for review.

  1. Apply agreed categories
  2. Suggest the next owner
  3. Uncertain cases go to review
Example 03

Check a completed task.

Compare the work with an agreed checklist and flag exceptions for the person responsible.

  1. Compare with the checklist
  2. Flag exceptions
  3. The person responsible decides
Illustrative starting points for a 30-day proof. Suitability depends on the process, data, risk and people involved.
Sample

Inside a decision pack

A 30-day proof should leave you with enough information to decide what happens next. This example shows the question being tested, the cases used, the limits found and the reasons for the recommendation.

Illustrative example · synthetic information

View the sample decision pack

04 / FIT TEST

Four questions find the candidate.

Start with a repeated task whose outcome you can judge. Then ask:

  1. Is the task narrow and stable enough to describe a good result?Yes
  2. Can your team supply representative examples and separate cases for testing?Yes
  3. Is the task frequent or important enough to justify a dedicated workflow?Yes
  4. Can it run with appropriate privacy, security and human oversight?Yes

Four yeses make it a candidate for a controlled test.

Read the thinking behind the fit test

Test the practical options against the same measure
01 · Three optionsCurrent processExisting productProposed workflow
02 · 03Same representative cases, compared

Quality, cost, response time, failure handling, operating effort.

04 · RecommendationThe option that gives the business the best result.

It may not be the new workflow.

Compare the proposed workflow with the current process and any suitable existing product. Look at errors, cost, response time and the work required to operate it.
05 / RESPONSIBILITY

Make responsibility part of the build.

A production workflow needs more than a model that performs well in a test. It needs a named owner, access controls, operating instructions and a way to find and resolve failures.

Responsibility inside the workflowIllustrative · not a client system

  1. Inputs

    The information the workflow is allowed to use.

  2. Decision logic

    The rules or model that propose an action.

  3. Human review

    A person approves the actions the scope says require approval.

    Decision point

  4. Approved action

    The system takes only the actions the scope allows.

  5. Decision record

    What the scope says will be recorded.

With your team, we define the human review points. The handover includes the agreed workflow, configuration, test cases, documentation and third-party dependencies.

For decisions involving personal information, the record should help your privacy team understand what the system does and what information it uses.

Read the OAIC guidanceSee the handover checklist

New privacy-policy disclosures apply to covered Australian entities using personal information in certain computer-made or computer-assisted decisions that could significantly affect an individual’s rights or interests. The requirement concerns disclosure; it does not require a business to build or own the software.

START WITH THE WORK

Bring the workflow that should work better.

Tell us where it slows down, where judgement matters and what a better result would mean. The first conversation will establish whether it’s worth testing.

Book a call First engagement: a 30-day proof on one decision.
See how we work