Works inside the tools you already use.
We work with your existing inputs, systems and handoffs. Small workflow changes are agreed where they meaningfully improve the outcome.
An AI implementation lab
AI that works
the way you do.
We build and manage AI that handles agreed routine tasks in the tools your team already uses.
Work With
Our technology shortlist
Models, agents, data and cloud platforms.
Models
Agents & automation
Knowledge & access
Cloud
Technology references only. No partnership or endorsement is implied.
Your team’s working day
Your team’s day.
Filled with admin.
Less routine work.
More team capacity.
A team of agents.
Built around yours.
Give your team
room to think.
Copying. Checking. Chasing.
Routine admin fills your team’s day.
Give routine work to AI.
Give your people room to think.
Orders preparedDifferences flaggedFollow-ups drafted
Your team keeps the decisions, exceptions and customer relationships.
Scroll to make room
Continue to our thesis
Your team’s illustrative working day contains six routine tasks. Reading orders and updating records move to an orders agent. Matching records and flagging differences move to a reconciliation agent. Chasing information and preparing updates move to a follow-up agent. A flagged price exception returns to your team for judgment. Your team has more room for customer care, decisions and complex problems. This is an illustration, not measured time savings.
The opportunity
Software should adapt
to the business.
For years, businesses changed how they worked to fit their software. We believe AI is changing that equation.
As the cost of building and adapting software falls, custom execution becomes possible for more workflows. In some cases, fitting AI to your business may cost less than fitting your business to another product.
That is our thesis. We put it to the test through complete implementations and the real cost of keeping them running.
Your business sets the brief.
A secondee joins an existing team to do a defined job. We bring that idea to AI, with agreed tasks and ongoing implementation support.
Your rules, made operational.
Business context, permissions, execution steps and checks are tailored to the task. Your team keeps control of approvals and exceptions.
Whatever fits your systems.
Your workflow determines the tools we use, from software rules and APIs to AI models. We evaluate choices against your task.
Built for you. Reusable where it helps.
You benefit from reusable components and evaluation methods where they fit. Your data and configurations stay separate.
Our approach
Keep both ends.
Adapt the middle.
Your customers keep sending information in familiar ways. Your team keeps getting the results it needs. Secondee handles agreed tasks in between.
Your existing input
A customer emails
an order.
The same inbox. The same customer experience.
Customer orderPurchase order attached
Secondee takes it from here
The work
in between.
- Read the order and attachments
- Match customer and product records
- Check required details
- Prepare the order for review
Missing information? Flag it to the owner.
Your familiar outcome
An order, ready
for review.
Your operations team stays in its existing order system.
Order record preparedAwaiting human approval
Prove the work. Improve how it runs.
We’re developing a way to make proven AI workflows more efficient without weakening checks. First, improve how the work gets done. Then consider smaller models where they make sense. Test each change against agreed quality and total-cost criteria before rollout.
Reuse this method as needed, without repeating every step each time. Execution improvements can go straight from step 03 to the shared release checks in step 05; step 04 is optional.
-
Set the quality baseline
Define the task, acceptance checks and approval boundaries. Use a capable model to establish a working baseline, including errors, human review, completion time and the full cost of delivery.
-
Keep evidence of what happened
Record the steps taken, tool results, corrections and checked outcomes, not just the final answer. Use only authorised data, and keep evaluation cases separate from records used to optimise the workflow or train a model. Permission to record is not permission to train; customer data stays separate.
-
Improve execution first
Keep the execution model fixed while testing how work is allocated, retried, run in parallel or stopped. Replay recorded paths where the evidence supports it, then validate promising changes on fresh tasks. The aim is less unnecessary work without weaker checks.
Replay requires recorded steps and valid dependencies. Changes to models, reasoning settings, prompts, tools or state-dependent execution paths need fresh execution or isolated tests.
-
Specialise where it pays
Try rules and existing smaller models before training. Use a task-specific model only when the data is permitted, performance meets the task standard, and the expected benefit outweighs training, deployment and ongoing costs.
-
Verify each change before rollout
Test against unseen cases, then run alongside the current workflow without taking external actions. Adopt a change only when it clears agreed quality, critical-error and total-cost gates. Otherwise, keep the current version.
Apply the same release checks to execution and model changes. Keep permissions, approvals and escalation paths intact, with monitoring and a way to roll back.
Compare candidates using the same acceptance criteria and evaluation cases. Add fresh held-out cases between evaluation cycles. When comparing execution strategies, also hold model settings, evaluators and necessary environment conditions fixed.
Total cost includes model usage, implementation, optimisation, evaluation, training, deployment, operation, human review, rework and maintenance.
How we work
One workflow.
Then the next.
Start with a bounded task and a clear definition of done. Expand when the evidence supports it.
-
01
Diagnose
Map the workflow, choose a task and agree the owner, operating boundaries and baseline.
A clear scope -
02
Run a paid pilot
Build and evaluate against real work. Measure quality, review effort and the full cost of delivery.
Evidence to decide -
03
Operate & improve
Monitor results and test changes to how the work runs, as well as the model used. Assess task quality, human review and total delivery cost together. Keep releases controlled, with agreed approvals and rollback.
Ongoing operation
The standard we work to
Useful work.
Measured honestly.
We do not start with an automation percentage. We start with the value left after all the costs are counted.
- Net time saved
- After human review, corrections and rework.
- Real operating impact
- Less backlog, overtime or outsourcing, or more work your team can handle.
- Positive total return
- After implementation, model usage, operation, maintenance and errors.
Workflow library
Find a familiar
part of your day.
From checking invoices to chasing updates, explore where AI could ease the repetitive parts. Your team brings the context, makes the decisions and handles what needs a person.
15 workflows and 12 role profiles, informed by the Human Middleware Index (opens in a new tab), a research project by Victor Zhang, Secondee’s founder. These are illustrative opportunities, not Secondee customer case studies.
Loading the workflow library…
No matching workflows. Try a different word or view all teams.
Time calculator
Give your team
some time back.
Make a first estimate of the repetitive work AI could lighten. Start with a workflow or role, then use your own numbers. The result allows for review, corrections and ongoing oversight.
The calculator loads with the workflow library.
Illustrative first-year estimate
Back to assumptions0hours back
each working week
each working week
0hours over a year
A$0annual value of that time
Time for customers, careful decisions, a smaller backlog or a more manageable day. The monetary figure values that capacity; cash savings depend on spending actually avoided.
Your cost scenario
No cost scenario has been set. Add project costs or a share of time value that avoids actual spending to explore one.
- Annual spending avoided
- Annual running cost
- First-year net, including setup
- Simple payback
Only the share you identify as avoided spending funds this calculation.
Use this estimate to choose a small, measurable pilot with your team. How we work
Preview the email summary
Where the estimate comes from
Human Middleware Index is a research project by Victor Zhang, Secondee’s founder. Its research priors are not independent verification of Secondee’s delivery.
For each workflow: team hours × repetitive-work share × estimated AI completion × first-year rollout, less ongoing oversight. Role estimates add their task shares once and leave unallocated time untouched.
Annual hours use your working weeks. Time value uses annual employment cost divided by 1,800 productive hours. Low, central and high scenarios reflect different assumptions; they are not a statistical confidence interval.
The starting research describes configured products in businesses of roughly 50 to 300 people. Actual systems, task mix, volumes and quality requirements can change the result. A pilot measures the outcome on your own work, including review and rework.
The research informs opportunities and assumptions. It does not establish that Secondee has delivered these outcomes or connects to every system. Calculator inputs stay in your browser. Sharing the result summary is optional; it excludes raw assumptions.
Read the Human Middleware Index research (opens in a new tab) · Source licence
Start with Secondee
What’s waiting
between your systems?
Tell us what gets copied, checked, chased or re-entered.
One
workflow is a good place to start.