Most AI projects stall between the demo and the real systems
A chatbot that answers questions from a PDF is easy to show. Getting AI to read your job cost data, draft a submittal log from Procore, or pull a serial number from your PIM without breaking anything is the actual work. Our process front-loads the questions that decide whether a project works: which systems, which data, who owns the output, and what the AI is not allowed to touch.
Each phase ends with something you can hold: a report, working software, or a monthly operating record. If the next phase doesn't make sense, you stop there and keep what you paid for.
Discover: find the work worth automating, and price it
Discovery is a fixed-fee engagement, typically about 30 days. We interview the people who do the work, map the systems involved, and look at where time and money go.
- What happens: stakeholder interviews, workflow walkthroughs, a systems and data inventory, access requirements, and a first pass at risk and governance questions.
- What you get: a Findings & Opportunity Report that ranks use cases by dollar impact, with effort, dependencies and a recommended first build.
- How long: typically about 30 days, depending on how quickly we can get time with your team.
- Who's involved from your side: an executive sponsor, one or two people who run the workflows day to day, and whoever administers the systems in scope (IT or an outside provider).
Decision point: build, adjust, or stop
You review the report with us. You can move the top-ranked use case into a build, reorder the list, or stop. A portion of the Discovery fee can be credited toward a build signed within a set window, so Discovery isn't a sunk cost if you go forward.
Build: fixed-fee phases, defined deliverables
We build in fixed-fee phases or a monthly, fixed-fee over a given time period. Each phase has a written scope, a list of deliverables, and a list of what we need from you to hit the schedule. Work is connected to your real systems from the start, with access limited to what the scope requires. Anything outside the scope goes through a written change order before we do it.
- What happens: integration and data work, the application or agent itself, user testing with your team, and deployment into your environment.
- What you get: working software connected to your systems, admin and user documentation, and a handoff session with the people who'll use it.
- How long: set per phase in the statement of work. Phases are sized so you see working software early, not at the end.
- Who's involved from your side: a product owner who can make calls on scope, the end users who'll test it, and your system administrators for access and security review.
Decision point: accept, extend, or hand off
At the end of each build phase you review the deliverables against the written scope. You can accept and move into Operate, approve the next build phase, or take the system in-house with the documentation we've handed over.
Operate: someone accountable after launch
AI systems drift. Your data changes, your processes change, the models change. Operate is a monthly managed-services retainer that keeps what we built working and useful.
- What happens: monitoring, prompt and model tuning, fixes when connected systems change, user support, and new use cases added from the Discovery list.
- What you get: a running system, a regular review of usage and issues, and a backlog of next improvements ranked the same way as in Discovery.
- How long: ongoing, billed monthly, under the terms of your agreement.
- Who's involved from your side: a named internal owner, plus users who report issues and ideas.
What we need from you
Projects slip when these are missing. We'll list them in every statement of work so there's no surprise.
- An executive sponsor who can make decisions and clear roadblocks.
- Time with the people who do the work, not only their managers.
- A named contact for each system in scope, with authority to grant or deny access.
- Sample documents and data early: real bids, real inspection reports, real declarations, redacted if needed.
- Prompt review of deliverables at each decision point.
- A clear owner for the system after launch.
What we won't do
- Run pilots that never touch your real systems. If it can't connect to where the work happens, it isn't a pilot.
- Connect to any system before we've agreed in writing what the AI can and cannot access.
- Start work outside the agreed scope without a written change order.
- Promise outcomes we haven't measured. Projections in our reports are labeled as projections.
- Hide the prompts, configuration or documentation for systems you've paid for.
- Recommend AI where a report, a form or a spreadsheet would do the job.