The method
Find → Measure → Build → Launch → Operate → Improve
The same six steps on every engagement. This page says what happens in each of them, and when — because the biggest reason companies delay this kind of work is not cost, it is not knowing what the next six weeks look like.
The six-step process
From one costly workflow to a measured result
Each step has a named output, an owner and a point where you can decide whether to continue.
- 01
Find
Day 1 to day 3
Map the workflow and identify the bottleneck.
- We sit with the people who actually perform the work, not just the people who describe it.
- We record the steps, the systems touched, and the decisions being made at each step.
- We come back with the two or three points in the workflow where the time and errors actually accumulate.
- 02
Measure
Week 1
Capture the existing baseline and select the KPI.
- Volume, minutes per task, people involved, error and rework rate, and where the queue backs up.
- We agree on one operational KPI and what an acceptable result looks like — in writing, before anything is built.
- If the baseline says the workflow is not worth automating, we tell you that instead of building.
- 03
Build
Week 2 to week 4
Implement the automation, integration or internal system.
- We choose the technology after the problem is understood: sometimes AI, often plain software and an integration.
- You see working software early and often, in your own data, not in a slide.
- Everything is version controlled, documented and deployed from a repeatable pipeline.
- 04
Launch
Launch week
Deploy it into the real workflow, with monitoring and safeguards.
- Human approval stays in the loop wherever a wrong call would be expensive.
- Logging, error alerts and a rollback path are in place before the first real transaction runs through it.
- We run alongside the existing process until the numbers say it is safe to switch over.
- 05
Operate
First 30 days
Handle failures, changes and maintenance.
- We watch it in production and fix what breaks, including the things that break because a vendor changed something.
- Exceptions get routed to a person with enough context to resolve them quickly.
- You get a KPI dashboard showing the before and after, not a status email.
- 06
Improve
Quarter 1 and beyond
Measure the KPI and keep optimizing against it.
- We review the KPI with you on a set cadence and bring the next improvement, ranked by value.
- Prompt, model, rule and threshold changes are tested against the baseline before they ship.
- When the workflow is done improving, we go find the next one.
Week by week
What a first engagement looks like
Roughly four weeks from kickoff to launch, then 30 days of operating support. Every one of these has a named output, not a status update.
Day 1
Kickoff. We meet the people who perform the workflow today.
Week 1
Workflow mapped, baseline captured, KPI and acceptance criteria signed off.
Week 2 to 3
Build, with working software in your data reviewed at least twice a week.
Week 4
Launch into the real workflow with monitoring, safeguards and a rollback path.
Days 30 to 60
We operate it, fix what breaks, and optimize against the KPI.
End of pilot
KPI report against the baseline. You decide what happens next.
Step 01 in practice
What “find it” actually produces
A workflow as it runs today, and the same workflow with the manual parts removed. Clients regularly tell us the map alone was worth the engagement, because it was the first time the whole process was visible in one place.
Illustrative example — order entry at a mid-size distributor
- 1
Customer email arrives
- 2
Employee reads it
- 3
Copies data into a spreadsheet
- 4
Checks stock and pricing
- 5
Updates the ERP
- 6
Emails the manager
- 7
Manager updates the report
Time per order
0min
Seven manual touches. Every one of them can be wrong.
Result
85% less manual processing
Same orders. Same people. The judgement stays with them and the typing does not.
Step 03 in practice
Right now, a person is the integration.
Most of the build is not clever. It is connecting systems that were never introduced to each other, reliably enough that nobody has to check.
Email
Orders, requests
CRM
Customers, deals
ERP
Inventory, orders
Accounting
Invoices, payments
Documents
PDFs, contracts
Internal DB
The spreadsheet
Right now, a person is that middle circle. They read the email, look it up in the ERP, type it into the CRM and tell accounting. We replace the copying, not the person.
Step 04 and 05 in practice
Every stage, including the ones nobody demos
What happens, what can go wrong, and what we do about it.
Click any stage. This is what a workflow we operate actually looks like.
What happens
The model reads the document and produces structured fields — line items, quantities, dates, references — each with a confidence score.
What can go wrong
The model reads a field wrong, or is confidently wrong.
What we do about it
A confidence threshold set from measured accuracy on your own data. Anything under it goes to a person. Every extraction is logged with its input, so a wrong one can be found and reviewed later.
Expectations
What this needs from you
What we need from you
- One person who owns the process and can make decisions about it.
- Access to the people who actually perform the work — two or three sessions of about an hour.
- Read access to the systems the workflow touches, when we get to that stage.
- A willingness to be told the workflow is not worth automating.
What we do not need
- A documented process. Mapping it is the first thing we do.
- A technical specification. Bring the business problem instead.
- A decision about AI versus automation versus software. That is our job.
- Your whole IT roadmap. One workflow is enough to start.
Don’t hire us for an AI transformation. Give us one workflow.
We document it, establish the baseline, build the system, launch it and operate it for 30 days. You see the actual result before deciding whether DevnTech should touch anything else.
If it does not meet the acceptance criteria we agree before development starts, there is no implementation fee.
Show us the process that wastes the most time.
We’ll tell you what should be automated, what AI should handle, what should stay plain software, and whether rebuilding it is worth what it would cost.
- For businesses with $5M to $30M in revenue
- No long-term commitment
- Start with one workflow
- Keep ownership of everything we build