Arcadis pitches VOTA AI subscription to Hillsborough County for water-main failure prediction

Arcadis presented a commercial predictive-modeling solution, VOTA AI, to Hillsborough County Water Resources Department staff on a procurement presentation identified as PS25-00208, outlining how the vendor would deliver likelihood-of-failure modeling, consequence and risk analytics, condition-assessment planning and project-level capital planning.

The presentation explained why Arcadis recommended VOTA AI and described project staffing, data-preparation steps, training and licensing. Arcadis project manager Celine Hyer said the county would not need Arcadis to develop a custom model: “We don't have to develop it. It's purpose built for water mains.” She said the Arcadis team would provide standard operating procedures, training and data-cleanup support as part of the project.

The county committee chair, Elizabeth Walton, identified the meeting and led introductions. Greg Ostus, Arcadis's predictive-modeling lead, described VOTA AI's capabilities and user interface, saying, “VOTA will do your likelihood of failure. It also has consequence of failure. It does risk. It does project planning.” Ostus said VOTA operates as a software-as-a-service subscription used by utilities since 2017 and that the platform had been applied to more than a million miles of mains in other utilities.

Nut graf: The presentation focused on whether the county's existing pipe inventory and break-history data are sufficient to produce reliable LOF outputs and on procurement options: Arcadis proposed loading and validating Hillsborough County's data, running the VOTA AI model, delivering exportable results and training county staff on ongoing use; licensing would be an ongoing subscription.

Inverted-pyramid details: Arcadis reviewed prior projects to show outcomes and the work required to reach them. The firm said it had configured commercial models for SAWS (San Antonio Water System) over 7,500 miles and for Hartford MDC, reporting better prediction accuracy from VOTA AI compared with simple age-based or regression models; Arcadis said the VOTA model performed roughly 55% better than their earlier regression model in one withheld-breaks test. Arcadis also described a workflow: data profiling and GIS cleanup, model runs, validation with withheld-break testing, and iteration as new break data are collected.

Arcadis staff reviewed specific data issues Arcadis found in the county's sample inventory: active and abandoned pipes are both present in the dataset; many install years are recorded in decadal spikes (estimated by decade rather than year); and some material-installation year combinations appear anomalous. Arcadis told the committee these anomalies can be addressed through targeted cleanup and refinement techniques—examples they cited included mining as-built drawings, using subdivision/plat dates, hydrant inspection dates and staff interviews to refine install-year estimates—but Arcadis also said the VOTA AI model can be run with the data “as is” and that iterative testing and validation will quantify any limitations.

Arcadis demonstrated the VOTA UI capabilities described in the presentation: map-based LOF, COF and risk visualizations; saved queries; pipe-level attribute detail; export options (CSV, JSON, shapefiles); RUL (remaining useful life) projections; withheld-break validation plots (ROC/precision-style comparisons); and a project-planning module that proposes replacement/rehab projects based on user goals (for example, reducing a break rate from 19 to 15 breaks per 100 miles per year over five years) and constraints (minimum/maximum project lengths and monthly production capacity). The presenters also discussed condition-assessment planning for transmission mains that lack adequate break history: using desktop risk results to prioritize targeted inspection or field assessment.

On procurement and licensing, committee chair Elizabeth Walton asked about pricing and subscription structure. Arcadis explained the vendor licensing model is similar to other utility software: Arcadis would license the VOTA product for the project and include a license fee in the proposal; the county would also have the option to buy a direct subscription with VOTA after project delivery. Arcadis summarized the licensing approach as an annual subscription.

Arcadis answered questions about deliverables and continuity. The firm said all model outputs and data can be exported for the county's GIS and Power BI systems if the county chooses not to continue a subscription. Arcadis said it includes training, support and documentation and could provide one-on-one training or workshops; VOTA also provides online help and vendor support.

The presentation closed with the Arcadis team offering pilot and implementation options, condition-assessment services and an asset-management plan path. The vendor emphasized the need for good break-history procedures and staff workflow to improve future model performance, and it proposed concrete data-refinement methods such as using hydrant inspection dates and subdivision records when available.

Ending: The committee timed the session for 30 minutes of presentation plus 15 minutes of questions; Arcadis noted that further technical details and the written proposal were available to committee members. The procurement presentation concluded and the meeting recording will be posted by county staff.