Automation opportunity discovery

What it solves

The people who know a process best rarely have time to write requirements. Automation opportunities get lost between calendars that never line up, hand-written documentation, and context that fades with every handoff. The business needs a specification and criteria; product needs to compare opportunities without inventing metrics; engineering needs complete tickets.

This service turns what your team knows how to do into a specification that is ready for a decision. It is not a round of interviews that ends in a generic report: every statement is cited and confirmed by the person who made it.

How we run discovery

Each person receives an invitation link and takes part from the browser, with nothing to install. Before starting, they agree on which topics will be covered and with which AI providers, and they can withdraw that consent at any time: the session stops and the content is deleted. No audio is stored.

The interview runs by voice or text, one question per turn. It separates the problem from the solution the person suggests, asks for concrete examples, never revisits what is already confirmed, and accepts interruptions. It can be paused and resumed with a summary of what is confirmed and what is pending.

Every fact is cited back to the moment in the conversation where it was said. The person confirms it with four options: yes, no, it depends, or correct. When two answers contradict each other, the contradiction is shown and the person resolves it, not the model.

With the facts confirmed, a functional draft is written and reviewed first by the business and then by engineering. The business sets scope, impact, and confidence; engineering estimates the effort. At each gate, a person decides whether to continue or stop.

What we deliver

A functional specification: problem, users, current and future workflow, business rules, acceptance criteria in given/when/then format, and explicit exclusions. Anything nobody could confirm is marked as pending, with the role that has to resolve it. It exports to Markdown and JSON so it can go into your management tool; it never creates tickets or a backlog on its own.

Discovery ends there, with a specification that is ready for a decision. If you decide to go ahead, the pilot is the next step, and the Newstools team runs it with you following the approach we describe in "How we work": discover, pilot, operate.

Who it is for

Operations leaders who suspect there is repetitive work but cannot quantify it; product owners who need to compare opportunities against the same criteria; engineering teams tired of receiving half-formed requests. It works the same in a newsroom, an administrative department, or a sales team.

What is behind it

The method relies on a tool of our own in which the model never decides permissions or writes anything by itself: consent, interview steps, and citations are controlled by deterministic code. Every requirement can be traced to its acceptance criterion and its test. Usage metrics contain no conversation content.

It has been validated in tests with synthetic data, with no pilots with people yet. In those tests, 34 requirements and 15 explicit exclusions were traced. For voice, we compared three configurations with 24 synthetic sessions each: the one we chose responds in 1.36 seconds at the median and in under 1.86 seconds for 95% of turns, below the 1.5 and 2 second thresholds we set. A security review found no critical or high findings.