BatchHatch

DESCRIPTION

BatchHatch is a browser-based billing correction tool built for Northwind Energy's call-centre agents.

Northwind has 519,000 customers and zero smart meters. Every unbilled period gets estimated by a 2012 algorithm that has never been updated, producing over 13,000 billing errors a month. When a customer calls disputing their bill, agents can see the problem but have no way to fix it. The correction goes back into the same overnight batch job that caused it. Median wait: 28 days. Meanwhile, 44% of customers call back within a week.

BatchHatch puts the billing engine in front of the agent, live on the call.

The agent reads the customer's physical dial meter, enters the digits, and a faithful port of Northwind's own COBOL rating module recalculates the bill in milliseconds. The corrected record is written and queued as an actual 80-column SYS-01 batch record, ready for the 2 AM mainframe run tonight, not in four weeks. Case closed before the customer hangs up.

WHAT IT DOES

Mechanical dial visualiser. An interactive replica of a physical meter with alternating clockwise and counter-clockwise dials, gear linkage animation, Web Audio click sounds, and a live bill preview as you scrub the needle. A Customer Perspective Assistant flags ambiguous dials and surfaces the exact words the agent should say to guide the customer through re-reading their meter.

Validation before it reaches the engine. Readings below the previous read, zero-consumption entries, and physically implausible usage are all blocked before the billing engine ever runs. The batch file has a structural 0.00% rejection rate.

Hold before it leaves. Agents can spot a suspicious estimated bill in the backlog and suspend it from the overnight run before it ever reaches the customer. No complaint ever opens.

Operational shockwave. After each fix, the page shows in plain numbers what that one correction just stopped: the $34 back-office correction, the 28-day queue, the callback cost, the escalation path, all cancelled on this call.

AI agent in advisory and action mode. Powered by Gemini via Google's OpenAI-compatible endpoint, with four tool calls wired into the billing engine. In advisory mode the model physically cannot touch the billing system because the tools are removed server-side, not just via prompt instruction. In action mode it can triage, rate, and close.

One-click after-call work. A pre-formatted CaseTrack note is generated the moment the bill is fixed. Copy to clipboard in one click. Turns 180 seconds of manual after-call writing into 5.

THE NUMBERS

70% adoption, first-contact resolution rate going from 41% to 80%, produces around $861,000 per year in avoided callbacks, manual corrections, and complaint deflection. Implementation cost: $65,000. Payback: 4 weeks. The smart meter alternative costs $77 million and has been deferred twice.

The previous AI pilot, AskNorthwind in 2025, reached an 82.6% escalation rate because it could not touch the billing system. BatchHatch can. That is the only difference that matters.