Why We Built AI into AccuDose —
and Where We Deliberately Didn’t
Every industry is racing to slap AI onto their product right now, whether it makes the product better or not. We didn’t want to be that. Here’s what we actually built, why we built it, and where we drew the line.
For years, AccuDose has been about substance over polish. Reliable telemetry. Real-time data. At a fraction of the cost of traditional SCADA and other standalone cellular-based systems on the market. Tools built for people who run lift stations, pump stations, and other municipal and industrial systems.
That hasn’t changed. It’s not going to.
So when we started building AI features into the next version of the platform, the question wasn’t “how do we add AI?” It was: where does AI actually do work that operators and directors can’t reasonably do themselves — at scale, quickly, or as efficiently? Because every industry right now is racing to slap AI onto their product, whether it makes the product better or not. And we didn’t want to be that.
What follows is what we built, why we built it, and where we deliberately drew the line.
Where AI Actually Helps in This Industry
The question we kept coming back to was simple: what work in water and wastewater operations is currently happening manually — or could be much more efficient?
Not “what tasks can a chatbot pretend to do.” Real work. Tasks that consume operator hours, that require pattern recognition across large amounts of data, that involve writing prose from structured information. Tasks where the limiting factor is human attention, not human judgment.
When we mapped that out, three categories stood out as places where AI could legitimately replace manual effort without replacing operator judgment:
Pattern recognition across telemetry data over time. A single operator can monitor maybe ten lift stations effectively. Beyond that, slow-developing patterns get missed — chronic infiltration, gradual pump degradation, seasonal flow shifts.
Cross-site reasoning. When something happens at one station, it often relates to what’s happening elsewhere. Across an entire fleet, that kind of reasoning doesn’t happen manually.
Generating compliance documents from operational data. SSO incident reports. Discharge Monitoring Reports. I/I assessments. These are documents whose content is determined entirely by data the system already has — yet the writing of them consumes hours of operator time, every month, forever.
Those three categories aren’t theoretical. They’re real work happening right now in every utility, eating real labor hours, with real costs. AI is genuinely good at all three. So that’s where we put it.
What We Built
I/I Detection — Pattern Recognition at Fleet Scale
Inflow and infiltration is one of the most expensive problems in wastewater. Groundwater leaks into cracked sewer laterals. Stormwater surges into the collection system during rain events. Both inflate operating costs. The problem is that nobody is looking, because no human can.
To detect chronic infiltration at a single station, you have to compare baseline inflow rates against historical norms over weeks of dry-weather data. To detect acute storm inflow, you have to correlate flow spikes against precipitation events. To rank stations by I/I severity, you have to do this across the entire fleet, continuously. That’s a perfect AI application — not because it’s clever, but because it’s mechanical, attention-intensive, and impossible to do by hand at scale.
What we built analyzes every site in your fleet continuously. Every station gets a severity score. Every station gets ranked. Stormwater inflow and groundwater infiltration are tracked separately because they’re different problems with different fixes. When you sit down to plan next year’s rehabilitation budget, you’ll know exactly which stations are the worst contributors — not because a consultant studied them for three weeks, but because the system has been watching them every minute of every day.
Fleet Chat — Language Access to Your Own Data
Every AI chatbot that ships in software right now claims to “let you ask questions in plain English.” Most of them are search engines with a conversational interface — they look at your screen, find the field that matches your question, and read it back.
Fleet Chat is a different thing. It’s connected directly to your live telemetry — every pump state, every alarm, every wet well level, every runtime, every event across your fleet. And it doesn’t just retrieve data. It reasons about it.
If you ask “which pumps have been running the longest today,” a search-engine-style chatbot returns a list. Fleet Chat returns the list and notices when the top-running pump is in a station currently in HIGH ALARM with both pumps cycling — and tells you the extended runtime is probably explained by the alarm condition, not a pump problem. Every experienced operator does this kind of reasoning constantly. Fleet Chat doesn’t replace that judgment. It makes the first draft of that interpretation happen in seconds.
AI-Generated EPA Reports — Eliminating Manual Compliance Work
Every wastewater operator who has ever filed an SSO notification knows the routine. Incident happens. You pull the timestamps. You cross-reference wet well levels at the time of overflow. You estimate duration. You estimate volume. You format it the way your state agency requires. You re-check the numbers. You submit. For one SSO, it’s an afternoon of paperwork. For monthly DMRs, it’s a recurring chore that consumes a half-day every month — forever.
Compliance reports are paperwork whose content is fully determined by data the system already has. The data exists. The format requirements are known. The math is straightforward. The only reason this work takes hours is because a human has to assemble it. The same task that takes an operator three hours takes the system ten seconds.
What this gives you isn’t just time savings — it’s consistency. Every report is generated from the same data source, formatted the same way, with the same compliance status checks. If you have someone on your team spending 20% of their week on compliance paperwork, the math on this feature is obvious.
This isn’t AI making decisions for you. It’s AI doing the data work so you can make decisions faster.
Where We Didn’t Add AI
The new AccuDose also includes two features that don’t use AI at all — and that was a deliberate choice.
The Standard We’re Holding Ourselves To
We didn’t add AI to AccuDose because AI is what’s selling right now. We added AI where the work was already there to be done — pattern recognition across fleet data, language access to telemetry, generation of compliance documents — and where it could measurably reduce the manual effort operators and directors are already putting in.
Where AI didn’t help, we didn’t use it. Where it does help, we built it carefully.
That’s the standard. We think it should be the industry’s standard.
The Next Version of AccuDose Is Launching Soon
If you’d like to see the new platform in action — including I/I Detection, Fleet Chat, and AI-generated EPA reports — reach out to our team. We’ll walk you through it.