Turning Scattered Field Knowledge Into Clearer Decisions

Disclosure. Clear Unit is a business affiliated with HelixAI. We are using this project to develop and evaluate HelixAI's approach. This is an ongoing initiative, not a finished product, and we describe only what exists today.

Property maintenance and renovation look like physical work, and they are. But most of the cost, delay, and frustration in a job starts as an information problem. Before anyone swings a hammer, someone has to decide what the job includes, what it will take, and what it should cost. That decision is only as good as the information behind it.

Where the information actually lives

On a typical turnover or renovation job, the facts that drive a decision are spread across many places:

• Requirements. What the owner or property manager asked for, what the unit's condition requires, and what local codes or building rules may require.
• Materials. What is needed, what it costs today, where it was bought last time, and what substitutions are acceptable.
• Labor. Who is available, how long similar work has taken before, and what the crew learned on the last job.
• Scope. What is in the job, what is explicitly out, and what changed after the first walkthrough.
• Experience. Knowledge that sits with the people who do the work: the quirks of a particular building, the supplier who actually has stock, the step that is always forgotten.

None of this is unusual, and none of it is unmanageable on a small job. The difficulty is that the information is scattered across messages, photos, receipts, memory, and spreadsheets, so every estimate begins by reassembling it.

Why this makes estimating hard

When information is scattered, three things tend to happen. Estimates depend on whoever happens to remember the relevant detail. Similar jobs get priced differently because the earlier reasoning is hard to find. And changes mid-job are discussed informally, which makes it harder to say later what was agreed.

The aim of this project is not to automate judgment. An experienced estimator's judgment is the valuable part. The aim is to make the information behind that judgment easier to find, easier to keep consistent, and easier to review.

What exists today

We want to be precise here. The following is built and deployed on Clear Unit's server:

• A project portal for Clear Unit jobs, with role-based access for the owner, crew, project managers, and property owners. Internal cost and margin information is filtered on the server, so it is never sent to roles that should not see it.
• Project records that hold notes, punch lists, and change-order requests that project managers and owners can approve or decline.
• Conversational logging. A small connector lets an AI assistant record mileage, material purchases, and hours against a specific project from a plain-language message, so the details are captured when they happen instead of reconstructed later.
• Assisted drafting. Through the same connector, the assistant can help draft bids, build shopping lists, add punch-list items, and attach receipts. It drafts only: sending, approving, invoicing, and completing work stay with people in the portal.

How the pieces connect today: a person talks to an AI assistant, a small local connector passes the request to the portal over an authenticated HTTPS connection, and the portal enforces who may see what.

What we are still working out

This is the part that matters for the larger question, and it is not finished. We are exploring how to connect the other inputs to estimating, such as requirements, material pricing, and labor history, so that a new estimate can start from what the business already knows rather than from a blank page.

We have not yet established how much of that can be done reliably, what accuracy is realistic, or how much review it needs. We have not measured time savings or cost effects, and we are not claiming any. What we have so far is a working foundation for capturing information consistently, and a clearer view of the questions worth testing next.

What we are learning

• Capture beats cleanup. Information recorded at the moment of work is more useful than information reconstructed afterward.
• Roles matter. The same job looks different to the owner, the crew, and the property manager. Showing each of them only what they need is part of the design, not an afterthought.
• People stay in the decision. Assistance is most useful when it prepares and organizes, and leaves approval to someone accountable.

Why we are sharing this

This is the way we approach every client problem: begin with the specific work, understand how information moves through it, and build only what the situation calls for. Clear Unit gives us a real operating business in which to test that approach, and we will update this case study as we learn more.

If your business has a similar problem, where the knowledge exists but cannot be found when it is needed, we would be glad to hear about it.