Water treatment service report summaries, drafted from your technicians' field test sheets for your account manager to approve.

A month of visits leaves a pile: field test sheets for each system, notes on what was fed or adjusted, an operator's log, sometimes a controller printout or a lab result. An account manager puts it in the order the customer expects, carries each number over, finds the range the program states for it, and writes down what ran out of line or never got tested.

For the test, AI reads copies of those sheets for months you already reported and drafts the customer's monthly service report: system and date, test readings by name as recorded, the control ranges the customer's program states, chemical feed notes, and every reading out of the stated range or missing. It doesn't decide what a reading means or what to feed next. Your reviewer sets each draft beside the report the customer received.

The test list changes by system. The University of Arizona's 2026 request for its central utility plants sets the minimum tests for each scheduled visit. Tower water runs through pH, conductivity, ORP, M-alkalinity, calcium hardness, silica, PTSA trace, phosphonate, azole, iron, copper, turbidity, free halogen and a biological count. Boiler feedwater brings sulfite, steam boiler water brings polymer and P- and OH-alkalinity, and condensate brings total amines. A blank on a sheet is a missing reading only if that test belonged to that system that day, so the draft reads each customer's own test list before it calls anything absent.

Two people can test the same water. That request has the provider bring its own reagents and equipment to verify plant results, and has plant staff run dip slides on open and closed systems each week. The biological count at a visit comes from a strip inoculated 48 hours earlier by university personnel and incubated at 35 degrees Celsius, so its date is not the visit's date. The draft gives the operator's number and your technician's number lines of their own, with who took each and when. It never averages them or picks a winner.

The range belongs to the program, and the program changes. Arizona's 2023 specification gave bidders a monitoring sheet with four blanks for every test: test for, method, control range and test frequency. That year its chilled water list included nitrite. The 2026 list drops it for silica, under a silica and azole inhibitor program. A range read from last year's sheet produces a false exception, so each draft names the program version it read, and a test with no stated range is shown as exactly that.

Some results run on a slower clock, or come from a machine. Arizona's corrosion coupons sit 90 days on tower, closed loop and steam condensate systems, in mild steel, copper and, where used, stainless or galvanized steel, with the data due within 30 days. Glycol concentration and deaerator dissolved oxygen are tested once a year. Its specification also has controllers send a weekly graph of ORP, pH, conductivity, trace, biocide, bleed and flow relays and temperature. A monthly report carries only some of that, so the draft marks which items fell due in the period. Where a controller printout and a hand test disagree, both appear and neither is called right.

What your account manager checks. Arizona's service representative has to tell the central plant supervisor, aloud and in writing, of any deviation from designated treatment levels, and quarterly reviews go through out-of-control exceptions by system. So your reviewer asks: did every out-of-range reading reach the list? Was a number altered on the way over? Were units kept as written? Did a recommendation appear that the technician never wrote? The draft offers none of its own, and nothing in it is an engineering judgment.

When your reviewer says yes, new field sheets feed the same draft reports for about 30 days, at a fixed price you see before signing.

Your account manager approves every report before a customer reads it. What to feed, what to change and what to tell the customer stay where they are today.

A next job would be decided and priced separately, with monthly support starting once one is live.