Skip to content
be-human.ai
← Back to the journal

AI by industry

AI Agents for Excavation Contractors: Estimates, Locates and Dig Logs

How an excavation contractor can use an AI agent on locate tickets, estimates and dig logs, and how to choose between an agent, a site overhaul or a rebuilt estimating flow.

Be Human typographic cover with the title: AI Agents for Excavation Contractors: Estimates, Locates and Dig Logs
Conceptual illustration created with AI · Be Human

An AI agent for an excavation contractor works best when it runs on the company's own records: locate tickets, utility markings, dig logs and the estimates built from them. Used this way, the agent can draft estimates with the right exclusions, check whether a locate is still valid before a crew mobilizes, and turn every finished dig into searchable history. Whether to start with an agent, a site overhaul or a rebuilt estimating flow depends on how complete and consistent those records are today.

Key takeaways

  • Locate tickets and dig logs are the most valuable memory an excavation company already produces, but they are usually scattered across portals, phones, field notes and inboxes.
  • An agent should start with one narrow job, such as estimate drafting from confirmed site facts or locate status checks, with a named person approving anything that commits money or crew time.
  • If dig logs are missing depth, utility owner or exact location for most jobs, fix the capture process before adding any automation.
  • An estimating flow that cannot show which assumptions came from which locate or site visit will repeat the same surprises, whatever tool sits on top of it.
  • Success should be measured by whether the next estimate on a similar site draws on earlier records and by how often a crew finds a utility the paperwork did not mention.

Why locate tickets and dig logs become company memory

An excavation company makes decisions based on information that expires or gets lost. A locate ticket has a validity window, a set of marked utilities, a response from each owner and often a note from the field about what was actually visible on the ground. A dig log records where the crew opened the ground, what they found, how deep they went, whether a line was damaged, and who signed off. Individually, each record is small. Over several years, the set of records is the closest thing the company has to a map of its own experience.

That is why the question is not only how to answer a phone or send a quote faster. The useful question is whether the company can recall what it learned on a street, a subdivision or a commercial lot the last time it worked there. When the answer lives in one estimator's memory, the company loses it whenever that person changes roles. When it lives in folders, nobody searches them before pricing a similar job.

An AI agent can make these records usable, but only if they are structured enough to read. Free text such as "gas line near curb, went around" is valuable to a person and hard for software to trust. A short set of consistent fields, such as ticket reference, site address, marked utility types, excavation method, depth found and a photo reference, is what lets an agent compare one job with another. Each new dig then adds a row that makes the next estimate a little better informed.

Where an agent helps, and where it should stay out

The strongest first use is estimate preparation from facts that already exist. The agent can gather the locate status, the site visit notes, the soil or surface conditions recorded in earlier dig logs, the equipment usually needed for similar work and the exclusions the company normally states. It drafts an estimate that lists each assumption and where it came from. A human estimator reviews the draft, adds judgment about access, schedule and risk, and signs it.

The second useful job is locate and ticket awareness. The agent tracks which tickets are open, which are approaching expiry, which have responses still pending and which sites have changed scope since the ticket was filed. It does not decide whether digging is safe. It surfaces the status and routes the question to the person responsible for the job.

The boundaries matter as much as the capabilities. An agent should not confirm that a site is clear, approve a dig without a valid locate, change a price already sent to a client, or write a dig log entry as though a crew had verified something it had not. Those actions belong to people. Authorization rules should make this explicit: the agent may read tickets, read logs, create drafts and post reminders, but it may not send commitments or edit confirmed records. The OWASP guidance on authorization describes the general principle of granting only the access a function needs, and it applies directly here. Good permission design also helps with prompt injection, since an agent that reads external text from vendor emails or utility responses should never be able to act on instructions embedded in that text.

Choosing between an agent, a site overhaul or a rebuilt estimating flow

Most excavation owners face one of three situations, and each calls for a different first move. The decision is less about technology than about the state of the records.

  • Choose an agent first when the records exist and are reasonably consistent, but nobody has time to search them. The task is retrieval, drafting and reminders.
  • Choose a site overhaul first when crews capture field information on paper, in messages or in photos with no common reference. The task is getting the capture right: a short field form, a ticket number on every dig log, and a rule that no site closes without a log.
  • Choose a rebuilt estimating flow first when estimates are built from memory and spreadsheet habits, so that two estimators price the same trench differently. The task is redefining what an estimate must state: scope, locate status, assumed conditions, exclusions, and the source of each assumption.

Often the answer is a sequence rather than a single choice. Capture comes first, then estimating, then the agent. Starting with the agent on top of inconsistent records tends to produce confident drafts that nobody trusts, which is worse than no draft at all.

A practical sequence for the first implementation

A first engagement should be narrow enough to observe and correct. The following sequence keeps the risk low and makes the result easy to judge.

  • Pick one service line, such as utility trenching for residential subdivisions, and collect the last set of locate tickets and dig logs for that line.
  • Identify the fields that every record should have and mark which ones are missing in most records. Those gaps define the capture work.
  • Write the estimate template with explicit assumption lines, including the locate status and the source of each condition.
  • Define the agent's allowed actions in writing: read tickets, read logs, draft estimates, post reminders to the named estimator. List the actions it may not take.
  • Set one success criterion that someone outside the agent can check, such as whether the estimator can trace every assumption in a draft to a record.
  • Run the draft process alongside existing estimates for a set period, compare the differences, and record each correction as a rule or a template change.
  • Only after the drafts match reviewed estimates reasonably well, extend the agent to ticket expiry tracking and to searching earlier logs for similar sites.

For a broader view of how teams structure agent work and evaluation, the Anthropic guide on building effective agents is a useful reference for keeping the scope of each agent small and its steps visible.

An agent is only as trustworthy as the record it reads, and the record is only as useful as the field that holds the fact.

Illustrative scenario: a utility trenching estimate

Illustrative scenario: a contractor is asked to price a fiber trench along a street that was partly excavated for a water main six years earlier. The estimator has a new locate ticket, two old dig logs from the same block and a photo of a manhole cover. In the current process, the estimator remembers the previous job, calls the foreman and builds the number from that conversation.

In the structured flow, the agent pulls the new ticket and notes that two utility owners have responded and one is still pending. It finds the earlier logs, which record a shallow gas service crossing near one property and a rock layer at a specific depth. It drafts an estimate that lists the trench length, the assumed depth, the rock allowance as a conditional line, the pending locate as a blocking item and the source for each condition. The estimator adjusts the allowance based on the foreman's view of the current pavement, approves the draft and sends it. The client sees a clear list of conditions, and the new dig log will record whether the rock assumption held. Nothing in this flow requires the agent to decide whether digging is safe. It only makes the company's own experience visible at the moment a price is set.

How Be Human handles this in a managed setting

Be Human designs and operates the agent layer around the company's own records, so the agent keeps a defined identity, a memory of the company's standard exclusions and templates, and access limited to the tickets, logs and estimate drafts it needs. The first engagement defines one useful role, the memory and tools that role requires, the approval boundaries and an observable success criterion, which for excavation work is usually traceability of estimate assumptions to records. Managed operations can then cover monitoring of the draft quality, recurring reviews of the rules, maintenance of any connections to ticket systems or accounting tools, periodic permission reviews and controlled changes when the business adds a service line. The work is delivered remotely, in English or Portuguese, and the company keeps control of what the agent may change.

Measuring whether the memory is compounding

The test of this approach is not whether the agent writes quickly. It is whether the company's records get more useful over time. Three checks are practical. First, when an estimator prices a site similar to one already dug, can they find the relevant prior log within a few minutes, and does the draft cite it? Second, are unexpected utilities or ground conditions recorded in the same structured fields each time, so that patterns become visible by street, soil type or line owner? Third, does the number of estimates that need a major rewrite after the field visit go down as the template improves?

These checks do not require a dashboard. A monthly review of a sample of drafts and dig logs, done by the estimator and the foreman together, is usually enough to show whether the memory is growing in the right direction. If the answer is unclear, the capture process is the place to look first. If the company is unsure which first task is worth handing to an agent, the free Be Human AI diagnostic can help identify it.

Frequently asked questions

Can an AI agent read locate tickets directly from the one-call or locate system? It depends on whether the system offers an export, an API or an email notification that the company is allowed to use. Where no direct access exists, the company can forward or upload ticket records into a controlled folder that the agent reads. Confirm local rules on who may access ticket data before connecting anything.

Is it safe to let an agent confirm that a site is clear to dig? No. Confirming a site is clear depends on valid locate responses, field verification and the judgment of the person in charge on site. The agent can show which tickets are open or expiring and which owners have not responded, but the decision to dig stays with a qualified person.

What should a dig log contain if the company wants it to be useful to an agent? At minimum, it should include the ticket reference, the site address or coordinates, the utilities marked and confirmed, the excavation method, the depth at which conditions were found, any damage or near miss, and a photo reference. Consistent short fields are more valuable than long narrative notes, although a short narrative can still be kept for context.

How does a company choose between buying an agent, overhauling its field capture or rebuilding its estimating flow? Look at the records first. If they are consistent and underused, start with an agent for retrieval and drafting. If field information is scattered, fix capture before automation. If estimates depend on individual memory and vary widely between estimators, redesign the estimate template so every assumption has a stated source, then add the agent where it saves the most review time.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →