Skip to content
Vinod Jose

8 min· water· ai

One program, six records

Columbia, Tennessee's water supply program shows up under at least four names in the public record. A vendor can trust a profile of it only once those names join.

EPA's table of closed WIFIA loans has a row dated 24 September 2026: City of Columbia, Long-Term Water Supply Program, Tennessee, $272 million. The board that runs the program belongs to Columbia Power and Water Systems (CPWS), and its agendas call the program "LTWSP". An EPA notice from October 2019 invited Columbia Power and Water Systems to apply for a $106 million loan for its "Long Term Water Supply Program", with no hyphen. So one program has a city as borrower, a utility board as owner, three spellings and two loan amounts seven years apart.

I believe the bulk of the work in using AI on utility data is joining records like these. Each record you join makes the next question cheaper to answer.

What the book says

A new book makes a similar case for companies in general: The Enterprise Brain, by Ragy Thomas, Sravan Vadigepalli and Chandhu Nair (2026). I am working from the public material on the book: interviews and the book's site. In a YouTube interview, Ragy Thomas says about 80 percent of the work is scaffolding, the data, connectors, permissions and review around the model. The book's description says firms built a body and a nervous system, and now have an organization that "sees everything but cannot think."

I'll admit I raised an eyebrow at the 80 percent. I know of no one who has measured that split for water data. Still, the shape fits what I see in utility records. Utilities publish board minutes, capital plans, permits and loan records in the public domain. Each vendor team reads a few of them, and the records do not point to each other.

I agree with the authors here. For water, I believe the value lies in building the scaffolding and the harnesses specific to the utility domain.

Six records for one program

Here is the Columbia program across the records I checked.

  1. CPWS board agendas and minutes, 2025 to 2026. The owner is the CPWS Board of Public Utilities. The program is "LTWSP" ("LTWSP Design Contracts", "LTWSP Supplemental Geotechnical Services", "GMP Amendment 10 - LTWSP"). Pieces of it appear under their own names, such as the raw water pump station ("RWPS") and "LTWSP WTP Backwash Recycle".
  2. EPA's finding of no significant impact, 12 August 2025. It sits on the program's environmental assessment, NEPA record N2025256. The EA file is named "CPWS LTWSP". It calls the existing plant the "EWTP" and places the new intake at Duck River mile 100. It also cites the state withdrawal permit, ARAP NRS23.228, and gives two different dates for it in two sections.
  3. The Master Bond Resolution, 18 March 2026. The CPWS board approved it 5-0 and sent it to the City Council.
  4. The SRF resolution, 22 April 2026. The minutes describe it as State Revolving Fund assistance for the water plant expansion, one component of the program.
  5. The Tennessee Local Development Authority resolution, 24 June 2026. The minutes use only "TLDA", the state authority that approves State Revolving Fund loans and the lien they sit on.
  6. The WIFIA loan. EPA's table names the City of Columbia, closes the loan on 24 September and lists $272 million. EPA's release announcing it is dated 2 October.

Items 3 to 5 I know from the CPWS minutes. The resolutions themselves are not posted with the minutes, so I cannot say what names they use inside.

The program also has identifiers that appear in none of the documents above. EPA's drinking water data lists CPWS as water system TN0000128, serving 84,617 people (second quarter of 2026). EPA's permit data gives CPWS one discharge permit, TN0004375, for its water treatment plant, where the facility also appears as "COLUMBIA WATER SYSTEM WTP". The wastewater plant has a different owner. The City of Columbia holds permit TN0056103 for it, under names that include "COLUMBIA STP" and "Columbia Wastewater Treatment Plant Improvements". So one city has two operators. A vendor who sells wastewater treatment equipment in Columbia is looking for the City, and the water supply program sits with CPWS. Neither ID appears in the loan or board records.

Columbia is a common case. In AquaIntel's utility database, 729 of 7,698 utilities (9.5%) are tied to two or more source records, mostly under different names. Most of those have two records, 10 have ten or more, and one has 34. And 549 utilities have both a drinking water system ID and a discharge permit linked, which is the join Columbia needs.

2 records545
3 records103
4 records33
5 to 9 records38
10 or more records10
Utilities tied to more than one source record in AquaIntel's utility database: 729 of 7,698 active utilities (9.5%), and the largest has 34. Separately, 549 utilities have both a drinking water system ID (PWSID) and a discharge permit (NPDES) linked, the join Columbia needs.AquaIntel utility database, 7,698 active utilities, 15 Oct 2026.
THE RECORD · WHAT IT CALLS THE PROGRAMJOINED1 CPWS agendas, minutes2025 to 2026LTWSP · RWPS2 EPA FONSI, N202525612 Aug 2025CPWS LTWSP · EWTP3 Master Bond Resolution18 Mar 2026name inside: not opened4 SRF resolution22 Apr 2026“water plant expansion”5 TLDA resolution24 Jun 2026TLDA6 EPA closed WIFIA loans24 Sep 2026City of Columbia · Long-TermLong-Term WaterSupply Programowner: CPWSIDS IN NONE OF THE SIXPWSID TN0000128CPWS water systemNPDES TN0004375CPWS water plantNPDES TN0056103City wastewater plantdashed: the link is a reading2019 invitation: “Long Term”
Six records, several names, one program. The PWSID and the NPDES IDs appear in none of them. Records 3 to 5 are cited from the CPWS minutes; the resolutions themselves were not opened. Sources: CPWS Board of Public Utilities agendas and minutes; EPA NEPA record N2025256; EPA WIFIA closed-loans table; EPA SDWA data (2026 Q2) and ECHO/ICIS permit download (20 Sep 2026).

What a join has to survive

The borrower and the board differ. The loan is to the City of Columbia, while every decision on the program is in the CPWS record. The city's own letter of 9 June 2026 to the TLDA explains why. CPWS is the city's board of public utilities and operates the city's water system, and the WIFIA loan is repaid from the system's net revenues "pursuant to the terms of the City's master water bond resolution." That is why the bond resolution goes to the City Council and why the city is the borrower. The link started as a possible match, and one more public document confirmed it.

The amount moved. A match on dollars fails, because $106 million in 2019 became $272 million in 2026.

Invitation to apply, EPA release, Oct 2019$106M
Loan in EPA's closed-loans table, Sep 2026$272M
Total WIFIA project costs, EPA fact sheet$556M
The same program, two loan amounts seven years apart. The loan is 48.9 percent of the $556 million in project costs (derived), against a WIFIA ceiling of 49 percent of total project costs that the environmental assessment states.EPA release, 22 Oct 2019 (invitation to apply); EPA WIFIA closed-loans table (Columbia row, 9/24/2026); EPA WIFIA fact sheet, City of Columbia (total WIFIA project costs); EPA NEPA record N2025256.

The name changes with the writer. "Long Term", "Long-Term" and "LTWSP" are the same program. "Water plant expansion" and "RWPS" are parts of it.

One event carries two dates. The loan closed on 24 September and was announced on 2 October. One record even disagrees with itself, as the EA's two permit dates show.

None of these is hard for a person who knows Columbia. Each is a place where software files a record under the wrong utility, or files it under no utility at all.

Utilities / Riverbend Water Authority / Aliases and IDsACME · workspace
Riverbend Water Authority
Utility profile · 4 possible matches awaiting review
OverviewAliases and IDsDocumentsPrograms
Names seen in public records
Name as writtenWhere seenStatus
Riverbend Water AuthorityUtility registerConfirmed
RWABoard minutes · 12 Aug 2026Possible match
City of RiverbendLoan table · 24 Sep 2026Possible match
Riverbend Long Term Supply ProgramEnvironmental assessment · Aug 2025Possible match
RBWTPCouncil agenda · 6 Oct 2026Possible match
Identifiers
PWSIDNot found in the records readNPDES IDNot found in the records read
Review: City of Riverbend
Same utility, or a different body?
Seen inLoan table · 24 Sep 2026 as borrowerCandidateRiverbend Water AuthorityWhyThe program name in the loan row matches the board minutes.UnknownThe legal link between the city and the authority.
Confirm matchNot a match
An aliases and IDs panel for a fictional utility. Each possible match waits for a person to confirm it, and the panel says which IDs the records read so far do not carry. Mockup on invented data: not a customer, not a real utility.

Why each join makes the next question cheaper

The book's site and Thomas's interviews describe "horizontal decoupling": each use case adds knowledge to a shared layer that later use cases reuse. The book's site also names an "App Trap", where intelligence stays stuck inside separate apps. I know that concept by name only.

In water, the shared layer is the joined utility record, and the payoff shows in later questions. Once LTWSP, CPWS, the City of Columbia and the EA are one record, new items land on the right program without fresh research. On 19 August 2026 the CPWS board approved a $2,919,850 Cummins generator package for the raw water pump station, bought through the Sourcewell cooperative contract. A system that already knows the RWPS belongs to LTWSP files that purchase with the program. The 28 October agenda item, "GMP Amendment 10 - LTWSP", attaches to Brasfield & Gorrie's construction contract the moment it posts.

The questions change too. A funding question (how is this program financed?) and a procurement question (who is buying for it?) read the same joined record. The answer to the second one is already in it: Brasfield & Gorrie under its guaranteed maximum price, with Hazen and Sawyer and Stantec on construction-phase engineering.

A Forbes article by Sandy Carter (11 August 2026) describes Belcorp, a consumer goods firm, building one governed layer that each use case reuses. It reports about 30 percent better integration efficiency and up to 20 percent faster time to market. Those are vendor-reported figures from a company far from a water board, and I would not quote them to a utility.

What I don't know

I do not know what share of the effort the join takes on water data. The TLDA packet lists the state loan under the applicant "Columbia" (loan 2024-10571, $75 million), so the state record carries the city's name too. I have not seen the bond resolution itself. And I have checked one program. Other utilities will break in other places.

Try this on Monday

Take one utility in your territory. Open three public records: its entry in EPA's drinking water system data (name and PWSID), its discharge permit in ECHO, and its last six months of board minutes. Write down every name, acronym and ID it appears under. Then look at the name on your own account record and see which of the three it matches.

At AquaIntel this join is the first step on every utility we profile.

Sources

I write about US water utility buying signals every two weeks. .