Company · Approach
Our Approach: Understand, Decide, Act, Verify
Start with the business problem. Keep people in control. Make every decision traceable. Ship one workflow before the next.
The short version
Start with the business problem
Most automation projects that stall were scoped around a technology. Someone wanted to use AI, picked a process that seemed to suit it, and found out late that the rules were unsettled, the documents were messier than the samples, or the result had nowhere to go.
We work the other way round. The first question is which queue in your business waits longest on a person, and what that costs. The second is whether the rules for clearing it are written down. Only then do we look at which product, agent or integration fits. Sometimes the answer is that the process needs fixing before anything is automated, and we will say so.
Everything below follows from that: the four moves every workflow makes, how the first one gets picked, and the list of things we will not automate however the project is scoped.
The four moves
Understand, decide, act, verify
The order is the point. Nothing acts before a rule has decided, and nothing closes until the result has been checked.
-
01
Understand
Classify what arrived and read the fields that matter. Each value gets its own confidence score, so a smudged GSTIN does not drag down a clean invoice total.
Bank statement → 312 lines, each scored
-
02
Decide
Run your policy exactly as written. If the policy is vague, we stop and ask you to settle it, because an agent cannot follow a rule nobody has agreed.
Invoice over ₹5,00,000 → finance controller
-
03
Act
Do the work through the system itself: post the entry, raise the task, send the approval request, update the status.
Approved → posted to the ERP with the PDF attached
-
04
Verify
Check that the action actually landed. Keep what was read, which rule applied and what came back, and send any failure to a person with the reason.
Posting confirmed → audit record closed
Picking the first workflow
The boring process usually goes first
We score candidate processes on three things. The one that wins is rarely the one people find most exciting, and that is fine.
Choosing the first workflow · Illustrative example
Illustrative data. Contract review is the process people find most interesting, and it is usually the wrong place to start.
Volume
Enough documents a month to measure the difference.
Settled rules
An approval policy someone has already written down.
One system
A single place to post the result, at least at first.
How deep the work goes
Configure before extending, extend before building
The cheapest route that solves the problem wins. We check the products first and write new code last.
Three depths of work, checked in this order
DocPro and FinSuite set up with your fields, rules, approval bands and ERP mapping. The shortest route, and the first one we check.
A new document type, a custom agent, or a workflow that crosses three systems, assembled from parts the platform already has. Most custom work lands here.
A new product, a modernized application or a longer AI program. Engineered on the same platform, so review queues, the audit trail and integration come with it.
Most projects mix the three: a configured product, a couple of custom agents and one integration nobody has built before.
Engagement shapes
Four ways to work with us
Each one ends with something you can use, even if you stop there.
Engagement shapes, with indicative timelines
Discovery is useful on its own. Some teams take the scope and build it themselves, and that is a perfectly good outcome.
Where we stop
What we will not automate
These hold however a project is scoped. An automation you cannot bound is one you should not switch on.
Judgement
Calls nobody can write a rule for
If two experienced people would decide differently, the agent prepares the case and a person decides.
Money
Payments above your thresholds
Approval bands are yours. Above them, a named person signs off every time, and the agent cannot change that.
Unclear process
Work nobody can describe yet
We map it first. Automating a process before it is agreed only makes the disagreement faster.
Unread inputs
Documents below the confidence threshold
Low-confidence fields go to review with the image beside them. They are never quietly guessed.
Policy changes
Rules edited without an owner
Changes to thresholds and rules are versioned and need an approver, so a quiet edit cannot change outcomes.
Your systems
Access we have not been granted
Agents act with the permissions you give them, through the integration your IT team approved, and nothing more.
FAQ
Questions about how we work
Why start with one workflow instead of a programme?
Because a programme built on an untested workflow inherits its problems at scale. One process, run on real volume with the controls in place, tells you what the accuracy, the exception rate and the review effort really are. The second workflow is then planned from evidence.
How long does the first workflow take?
Typically 4 to 12 weeks, an indicative range that moves with the number of document types, the workflows involved, the integrations and the deployment model. Discovery gives you a figure for your own case.
What do you need from our team?
A process owner who can settle rules when they are unclear, about a month of real documents, access to the system the results post to, and an hour or two a week from someone who knows where the exceptions come from.
What happens when the agent is not sure?
It says so. Each field carries a confidence score, and anything below the threshold you set goes to a review queue with the source image beside it. The reviewer corrects it, and the correction is kept with the record.
Can we change the rules later?
Yes. Rules and thresholds live in configuration your team can edit. Each change is versioned with who made it and when, so you can see which version of the policy any past decision followed.
Keep reading
Read next
About PTAS AI
What the company builds and how the products fit together.
Read more →Security
The controls behind the approach: access, audit, retention, governance.
Read more →Human-in-the-Loop AI
Why review is a control, and how to set the thresholds.
Read more →Automation Readiness
A guide to judging whether a process is ready to automate.
Read more →Next step
Find the queue worth automating first
The assessment scores your processes the same way we would in discovery, and tells you plainly which one to start with and what should stay manual.