Scoring every draft against a UK utility's own guidelines

Twistag built an AI communications assistant for a major UK water utility — a Chrome extension deployed across 6,000 employees that scores every draft against the utility's own guidelines in real time, saving the communications team around 18 hours a week and scoring more than 3,000 communications a month. The pilot recorded zero non-compliant language flags in regulatory bulletins. Twistag built it as a compliance system rather than a productivity tool: five scoring dimensions calibrated on hundreds of the utility's real communications, and a custom prompt layer encoding its regulatory language rules, updated without model retraining or redeployment.

The utility had six thousand employees, thousands of communications going out every month, and no reliable way to keep any of it consistent. Internal guidelines existed; nobody followed them under deadline pressure. Water utilities communicate under scrutiny in a way most industries do not. A service disruption notice that sounds defensive, a regulatory filing with ambiguous phrasing, a public announcement that reads like a different company wrote it from last month's — all of it erodes public trust slowly and visibly. Some of that inconsistency is more than a brand problem: the wrong sentence in a regulatory bulletin can be treated as an admission of liability. The client is not named publicly.

Twistag started with two weeks of discovery — working sessions with comms leads, IT and the people writing the documents daily — and came out knowing this was a compliance system, not a productivity tool. Those are different products with different scoring requirements, different prompt engineering and different definitions of success.

What Twistag built is a Chrome extension that runs inside any browser-based tool the team already uses: email clients, document editors, internal content systems. It captures text as people write, sends it to an analysis backend on Google Cloud Platform, and returns scoring and suggestions within seconds. The scoring engine was calibrated with the utility's communications team on a labelled training dataset built from hundreds of real communications, teaching an OpenAI-based pipeline the difference between a regulatory filing and a community update, and between acceptable plain language and phrasing that sounds like an admission of fault. Content is scored across five dimensions — clarity, empathy, actionability, tone and accuracy — and below threshold the system generates specific rewrite suggestions. The employee decides what to do with them. The AI does not publish anything.

The extension is the reason it was adopted. It runs inside the email client and document editor people already have open, stays invisible until content needs attention, and surfaces suggestions as an overlay rather than a separate destination. Twistag has seen across enough integrations to treat it as a hard constraint: if a tool makes someone leave the application they are in to get a suggestion, most will not bother, and the tool is quietly uninstalled.

Stack: OpenAI, Python, Google Cloud Platform, Next.js. Services: AI agents and applied AI, AI-native product engineering, prompt architecture and evaluation.

Six weeks into the pilot, the assistant was processing more than 3,000 communications a month. The communications team recovered around 18 hours a week, which went into proactive stakeholder outreach and the strategy work senior editors had been hired to do but could not reach because they were reviewing first drafts. The pilot recorded zero non-compliant language flags in regulatory bulletins — not significantly reduced, zero. For a utility under regulatory scrutiny, that is a different category of result from a productivity improvement. It is risk removed.

An improvement to clarity can't introduce compliance risk. The two are evaluated together.