11 min· building· ai· water
One utility, all the way down
Screening ends with a shortlist of names. This is what happens to each name on it: one utility, its whole public record, assembled into a document a person can act on. It is the thing this company started with, and the workflow that depends most on being pointed at the right accounts.
Most of what we build is a way of narrowing: a signal out of three hundred pages of minutes, a name out of an org chart nobody published, a bid worth answering out of a feed of hundreds, twenty accounts out of sixty-six thousand. This one does the opposite. It takes a single utility and tries to understand the whole of it.
It is also where this company started, before there was a platform or a plan. In December 2025 I read everything Gettysburg, Pennsylvania (a town of about ten thousand people) had put into the public domain, and turned it into five opportunities I believed they would want solved in the coming years. It took a week. It was meant to feed a newsletter. The newsletter never launched, because the profile turned out to be the product.
The job, as it exists today
Somebody is about to have a conversation with a utility they do not know well, and they would like to not waste it.
What they should know before they go: what this utility owns and what state it is in, what they have committed to spend and when, what they have already tried and how it went, who decides, who they already buy from, and what is coming up for renewal. That is seven things, and none of them is secret.
What they usually know: the last conversation, whatever is in the CRM, and fifteen minutes of searching the night before.
The gap between those two lists is not a knowledge problem. Every item on the first list is somewhere in the public record. It is a reading problem, and the reading is the part no one has time for. That is the same sentence that opened the monitoring piece, because it is the same underlying fact showing up in a different shape.
Why it is harder than it sounds
The volume is per account. A single utility gone through properly is several hundred documents and often past a thousand, once you go back far enough to see a pattern rather than a moment. We gather about fifteen years of history where the record allows (some utilities publish that far back and some do not). The reason is that a plant replacement cycle is longer than most people's tenure, and a decision made in 2019 is frequently the reason something is happening now.
The hard part is synthesis. This surprises people who assume the hard bit is finding the documents. Almost nothing worth knowing sits in one place. The useful sentence in a profile is nearly always a relationship between three documents that were written years apart by people who were not talking to each other: a condition assessment, a budget line that appears two cycles later, a rate case that justifies itself by pointing at both. No search returns that. Somebody, or something, has to hold all three at once and notice.
Every vendor is asking a different question, and the generic profile is worth very little to any of them. This is the cleanest distinction I know between this workflow and the screening one, and I owe it to a customer who drew the line for me in the middle of a call. Size, treatment type and compliance violations are screening questions, answerable across the whole market at once. Whether a utility does a particular job in-house or subcontracts it, and to whom, and whether they seem happy about it, is a profile question. There is no field for it. You only get it by going all the way into one organization.
And utilities are not one shape. A municipal authority publishes as one body. A large investor-owned operator is a corporation holding hundreds of systems, and a profile of it that does not separate the corporate view from the asset-level view is a document that answers no question anyone has.
What the workflow does
Start with the customer's questions. The profile is written from the vendor's point of view, the same principle as the fit-check and screening, for the same reason: we are not in a position to know which five facts about a utility matter to your business, and you are.
Read the whole public record for that one account. In practice this runs against three primary sources. First, the utility's own output: board minutes and agenda packets going back years, capital improvement plans, budgets, engineering reports, rate studies. The compliance record, which is where the pressure usually originates. And the money: bonds, funding applications, what has been raised and what it was raised for.
Those three answer different halves of the same question. The utility's own documents tell you what they want to do. Compliance tells you what they will be made to do. The funding record tells you which of the two they can currently afford, which is very often the difference between an opportunity and a wish.
Reconcile the history against the last twelve months. An archive read on its own produces something backward-looking: accurate about the arc, out of date about the present. People move on, priorities get reordered, and a document from 2019 often describes a utility that has since changed its mind. So a profile carries a current pass over the recent record on top of the historical one. The history tells you what kind of utility this is. The last twelve months tell you whether any of it is still true.
Assemble rather than summarize. The output is a set of positions: here are the opportunities we think are forming and why, here is the asset and vendor picture, here is who decides, here is what is under contract and roughly when it comes up, here is what they have already tried. Each with the document it came from, so it can be checked.
Then hand it to a person. The profile is finished when a salesperson can read it once and walk into a meeting better informed than the person across the table expects them to be.
Account Intelligence / Lake Aldon Water District
What it looks like in practice
This is an illustrative example. It is not a customer, and not a real utility.
You sell a treatment system. Screening has put a mid-sized utility on your shortlist and told you why: they have a contaminant problem you address. That is the extent of what a screen can tell you, and on its own it is enough to justify a call and not much else.
The profile goes further into that one account and comes back with four things. The utility commissioned a study on that contaminant three years ago and accepted it. Their capital plan carries a line for a treatment upgrade in the next fiscal year, sized in a way that tells you roughly what they think it costs. They floated a bond eighteen months ago that named the plant. And the operations manager has told the board twice that the current approach is "holding", which in a public meeting is a word worth noticing.
None of that is hidden, but all of it took reading. Together it changes the conversation from do you have a problem with this contaminant (a question they will find slightly insulting, because obviously they do and they have said so in public four times) to you accepted a study in 2023, you have a line in next year's plan, and here is what we would do about the part you have not solved. The value of the workflow is that you arrive having done the reading.
The most useful version of this I have seen in real work was less flattering to the customer and more valuable to them: a profile that showed a manufacturer that the aftermarket on equipment they had installed themselves was being serviced by a competitor. A company cannot see that about itself from the inside, and it came out of the public record of the utility rather than anything the manufacturer held.
A document, from someone who keeps saying documents are not enough
The monitoring piece argued that an alert is not a deliverable, and every workflow since has said some version of the same thing: the work is not finished until it lands where somebody already works. Then this one produces a report.
That is a fair thing to push on, so here is the distinction I would defend.
An alert asks for attention it has not earned, repeatedly, from someone who did not ask for it that morning. Its failure mode is volume: the more accurate the system gets, the more it sends, and the faster it gets filtered into a folder nobody opens.
A profile is read once, by someone who has already decided this account is worth their week. That decision was made upstream, by the screen, a territory plan or a bid that came in. The profile does not have to compete for attention, because attention has already been committed. That is the whole reason this workflow sits after screening in the running order rather than instead of it.
A customer once described the sequence back to me better than I had put it myself: identify the market problem you want to lead with, screen for the utilities that have it, then pull the deep profile on the ones you are going to pursue, and use it to decide what to say. So that is three workflows, in that order, and the profile comes last.
One more note on this. The first profile a customer gets is rarely the best one they get. The questions that matter to a business are hard to put fully into words in a first conversation. They surface by looking at a draft and saying go deeper here, drop that, add this. It usually takes two or three before the shape is right, and I would rather say so than have somebody read version one and conclude that is the ceiling.
What it cannot do
It is a scoping decision before it is a volume one. A profile is expensive in a way an alert is not. It is the deepest thing we make and the slowest, and depth is what you spend on the accounts you have already decided matter. Spread evenly across a whole territory, it is an expensive way of producing documents nobody reads, which is the problem this company exists to fix.
So "how many should we do" is the second question. The first is which accounts, and that is what the screen is for: run it, then profile what survives it. Done in that order, the number usually looks different from the one people start with, and every profile in it is pointed at an account somebody has already decided is worth the depth.
And a profile is a snapshot of a moving thing. Boards meet monthly. A document assembled today is a little less true after the next meeting and meaningfully less true after six, because people move, projects slip and money gets reallocated. So profiles are refreshed against the utility's own meeting cadence rather than ours. That means the deepest workflow we build depends on the simplest one, the monitoring layer, to stay current. A profile that is not refreshed goes out of date in the details, which is where it is hardest to notice.
And a document cannot be asked a question. This is the limit that turned out to matter most, and it only becomes visible once the profile is good. A weak one gets skimmed and forgotten. A strong one gets read closely, and a reader who is reading closely has a second question before they reach the end of it. Is that plant on the same permit as the other one? Has this come up before the board since? Who signed the last contract, and when does it end? The document cannot answer. The reader either comes back to us and waits a day, or drops the question, and mostly they drop it. We built the profile as the deliverable and then watched customers treat it as the opening of a conversation.
It also stops where the public record stops. Make, model and serial number live in the utility's maintenance system and no amount of reading gets you there. That edge, and why we leave it alone deliberately rather than reaching around it, is the monitoring boundary and has not moved.
Related
- Why can't we just ask ChatGPT?: an agent over the same curated record, for the questions a profile leaves open.
- Fifty thousand is not a market: deciding which accounts deserve this depth.