Cost of service study review that traces every formula and rebuilds every allocator before the rate case is filed
Every allocator rebuilt from its source before filing, so intervenors find nothing to fix.
See one case, screen by screen ↓A class cost of service study is a workbook of dozens of tabs and hundreds of thousands of formulas. It takes the revenue requirement, functionalizes it into production, transmission, distribution and customer costs, classifies it as demand, energy or customer, and allocates it to customer classes with allocators built from load research, customer counts and meter cost studies. Each case's model usually starts from the last case's, carried forward as a starting point.
So a 4CP allocator still reads last year's peaks, a subtotal stops one row short of a new account, or a number is typed over a ledger link. Each one moves money between classes. Today a second analyst ties out a few totals by hand, the outside consultant spot-checks, and formula auditing is a trace-precedents arrow, one cell at a time. Allocators are copied, rarely rebuilt from source. The people who find the error are commission staff and intervenors, in discovery, and the fix becomes an errata filing.
Estimated split for a full review of a 400,000-formula electric study, in working days, done by hand and with the solution. A full review by hand includes rebuilding the allocators, which is often skipped today.
Six agents map the model, trace every formula and rebuild every allocator; the cost of service manager decides and signs.
Lumina Utilities files its 2027 Electric General Rate Case, Docket E-2026-0418, on Fri 6 Nov 2026: a $1,284.6M revenue requirement and a requested increase of $96.4M. The class cost of service model is v14, built from the approved 2023 case model. Here is what happens to it, screen by screen, in the working solution.
Theo uploads LU_Electric_CCOSS_TY2026_v14.xlsx: 38 tabs, 412,880 formulas, test year 2026. The study opens as a flow from cost accounts to functions, allocators and the five customer classes, each band as wide as its dollars. v14 adds 3 tabs and 18,240 changed formulas, and the last review was v11 on Fri 2 Oct, done by hand: 3 totals tied out, allocators not rebuilt.
“Not reconciled to the revenue requirement since v11 (2 Oct, by hand) · v14 adds 3 tabs and 18,240 formulas”
Theo runs the review. The model mapper maps 9 account groups to 5 functions, 7 allocators and 5 classes. The formula auditor and the allocator recalculator place their flags where each error happens: F1 on the 4CP band, F2 on line transformers, F3 on distribution O&M, F4 on street lighting, F5 on weighted customers. The run takes 4 min 06 s, and a red line opens at the top of the flow.
“The model allocates $1,283,249,500 — $1,350,500 short of the revenue requirement of $1,284,600,000”
The study must allocate exactly the revenue requirement in Hana Sorensen's Schedule C-2, which she locked on Fri 2 Oct. Production, transmission, customer service and A&G tie to the schedule and the 2026 ledger. Distribution does not: the model carries $370,149,500 against $371,500,000. Three cells explain all of it: row 214 left out of the transformer subtotal (−$1,418,600), account 593 typed over its ledger link (−$542,300) and account 373 counted twice (+$610,400).
Ties to the dollar or explains why not; it never changes the revenue requirement.
The 4CP allocator in ALLOC!D27:H27 links to the 2025 load research tab, carried forward from the 2023 case model. The recalculator rebuilds it from LR_2026, Owen Pryce's AMI sample of 1,840 metered premises: summer peaks of 3,402, 3,561, 3,594 and 3,395 MW, an average of 3,488 MW, with residential at 1,608.0 MW. The residential share is 46.10%, not the 46.80% the model uses.
It uses the study's own method and flags the source; method questions go to the manager.
Theo selects Correct in model. The bands re-flow: residential falls to $645.5M, small commercial gains $0.97M and large commercial $2.13M. The correction goes to v15, a draft from v14, and v14 is kept as built for the record.
“Corrected in v15 · $3.10M moved from residential to commercial”
FUNC!J215 reads =SUM(J210:J213). Row 214, account 368.5 for transformers in new subdivisions, was inserted below the range in v12, so $1,418,600 of distribution demand reaches no class. The cell's history shows it: right with four rows in v11, row 214 added in v12, still left out in v14. The fix is one click: extend the range to J214.
It proposes the fix and never changes a formula itself; every finding names its cell and its effect.
Theo corrects F2, F3 and F4 together: the range is extended, account 593 is linked back to the ledger's $18,962,300, and street lighting comes out of the distribution demand pool. The red line turns green and the class totals count to their new values. Four of the seven findings are decided.
“Ties to the revenue requirement · the model allocates $1,284,600,000, to the dollar of Schedule C-2”
With the fixes in, the class results update: revenue-to-cost ratio, parity index and the increase each class needs to reach cost. Residential's parity index is 0.964. A typical home using 850 kWh pays $128.40 a month now and $143.94 at cost, $0.54 less than v14 showed. Marcus moves classes toward parity in steps in rate design.
“A million moved between classes is a residential bill change I have to explain.” — Marcus Bell
Against the approved 2023 study, the system is up 9.6% and industrial is up 13.7%, 4.1 points above it. Small commercial and lighting also moved more than 2 points, and both are already explained. For industrial, the comparer ties each driver to the 2026 ledger: the Brandt Ridge unit retired in June 2025 (+$9.10M), two data-centre customers, Halcyon Data and Westgate Compute, joined in 2026 (+$6.40M), transmission 12CP growth (+$4.60M) and distribution and customer costs (+$2.67M).
“No cell is wrong; the move goes in the testimony.”
F5 is a method choice: the industrial meter weight is typed as 1.0, while the March 2026 meter cost study puts it at 14.6. The 2023 study used 1.0 by agreement with the Ratepayer Advocate, so it is Renata's call. She accepts the meter cost study weight, accepts F6 and F7, and signs v15 as the study's witness at 09:28. All five checks are green, and the study goes to Celeste Duarte for the 6 Nov filing.
Agents find and propose; a person corrects, and the manager signs.
The same study, seen by the five people who build, tie out, price, sign and file it — what their case looked like, and what it looks like now.
Six agents map the model, trace every formula and rebuild every allocator; the cost of service manager decides and signs.
Reads the workbook and maps every link from cost account to function, allocator and customer class, draws the allocation flow, and compares each new version with the last one: the tabs and formulas that changed.
Traces every formula to its precedents and flags subtotals that skip rows holding values, numbers typed where the row links elsewhere, references to last year's tabs or files, a formula that differs from its row, and any amount that reaches a class by two paths.
Rebuilds every allocator from its named source: coincident and non-coincident peaks from load research, energy from billing plus losses, customers and bills from billing, weights from the meter and service cost studies. Reports any share that differs by more than the tolerance and the dollars it moves by class.
Ties every function to the revenue requirement schedules and the general ledger, and explains any gap cell by cell.
Compares class results with the approved prior case and explains every class move beyond the threshold with its drivers, each matched to the ledger.
Drafts the review note and errata log with cell references and the dollars each correction moves, ready for discovery.
Decides and signs. The agents propose; a named person decides.
Ask about any cell, allocator, class or tie-out, or ask for the review note to be drafted. Every answer points to its source.
Why does the residential share fall?
The production demand allocator in v14 still reads the 2025 load research: 46.80% residential. Rebuilt from this summer's four peaks — 3,488 MW on average, residential 1,608.0 MW — the share is 46.10%. On a $442.9M pool, that moves $3.10M from residential to small commercial (+$0.97M) and large commercial (+$2.13M).
Does the study tie to the revenue requirement?
Not yet. The model allocates $1,283,249,500, $1,350,500 short of Schedule C-2. Three cells explain all of it: row 214 left out of the transformer subtotal (−$1,418,600), account 593 typed over its ledger link (−$542,300), and street lighting counted twice (+$610,400).
Which classes moved most against the 2023 case?
Against the approved 2023 study, the system is up 9.6%. Industrial is up 13.7%, 4.1 points above the system: the Brandt Ridge retirement (+$9.1M) and two data-centre customers (+$6.4M).
Draft the review note for Renata
Drafted on the sign-off page: seven items, each with its cell and its source, the dollars each correction moves, and the tie-out to Schedule C-2. Items still open are marked, and Renata decides them before she signs.
13 screens from the working solution, on its sample data. Pick one to see it large.
Thresholds, allocator sources, the prior case and the checks are settings, chosen by the cost of service team.
| Setting | Default | Choose from |
|---|---|---|
| Materiality per finding | $50,000 | Set in $10,000 steps, from $10,000 to $250,000 — below it a finding is Low |
| Year-over-year explanation | 2 points | 1 point · 2 points · 3 points · 5 points — class change against the system change |
| Allocator tolerance | 0.05 points | 0.01 points · 0.05 points · 0.10 points of share before a flag |
| Reconciliation | To the dollar | To the dollar · within $1,000 · within $10,000 of the revenue requirement |
| Load research | LR_2026 · Jun–Sep 2026 | LR_2026 · Jun–Sep 2026 · LR_2026 · 12 months |
| Customer counts | Test year 2026 | Test year 2026 · Year-end 2026 — billing system, average of 13 months |
| Meter and service costs | Meter cost study 2026 | Meter cost study 2026 · Meter cost study 2022 |
| Prior case | E-2023-0112 · approved v22 | E-2023-0112 · approved v22 · E-2020-0287 · approved v31 — compared on every run |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample utility data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Missouri PSC staff testimony — about $4.6M change in a class cost of service; Panko's spreadsheet field audits, summarised by Pryor — 0.38–2.5 % of formula cells in error, 80–90 % of audited spreadsheets with a significant error). Context: Kentucky PSC case 2016-00371, corrected studies filed after a spreadsheet error; NARUC Electric Utility Cost Allocation Manual (1992); AWWA M1 base-extra capacity method. Lumina Utilities, the dockets, the people, customers and records named on this page are fictional — characters and sample data in the working solution.
A check of a utility's class cost of service model before it is filed: that every formula is right, every allocator matches its source data, and the study allocates exactly the revenue requirement. Cost of Service Study Review does this with six agents that trace every formula, rebuild every allocator and tie the model to the revenue requirement. The cost of service manager decides each fix and signs.
The model mapper maps every link from cost account to customer class. The formula auditor traces every formula for ranges that stop short, values typed over links, stale references, inconsistent formulas and double counts. The allocator recalculator rebuilds each allocator from load research, billing and cost studies, and the reconciler ties each function to the revenue requirement schedules and the ledger.
Every formula. In the sample study the review traced all 412,880 formulas and rebuilt all 14 allocators in one run of 4 min 06 s, and put each of the seven findings on the flow with its cell, its source and the dollars it moves.
No. The allocator recalculator uses the method in your study and flags the source; it never picks a different method. Method choices such as meter weights, peaks and classification go to the cost of service manager as a judgement finding, like the industrial meter weight in the sample case.
Yes. The same review runs on water studies built on the base-extra capacity method, with a check on the max-day and max-hour factors from the demand study. The sample utility has its 2027 water cost of service in the queue beside the electric rate case.
Yes. The agents never edit the workbook, never change the revenue requirement and never file. Nothing changes until a person accepts it, corrections go to a new version while the reviewed one is kept, the manager signs as the study's witness and the director approves the filing.
Yes. The materiality per finding, the year-over-year threshold, the allocator tolerance, how closely the study must tie to the revenue requirement, the load research, customer counts, meter cost study and prior case, the checks and the signers are all settings.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your cost of service model, its source data, your review checklist and a prior case — and it goes live once every quality gate has passed. We will walk you through it on a copy of your own last filed study first.
Every data request answered on time, with one number for one thing.
1-min demoRegulatory Precedent SearchEvery filing position backed by how regulators actually ruled, cited and trended.
1-min demoRegulatory Commitment TrackingEvery directive owned and dated within a day of the order, and closed with its proof.
1-min demoWe'll run Cost of Service Study Review on a copy of your last filed model and its source data.