AI for Well Drilling Companies: Permits, Driller Logs and Pump Service
Which driller logs, permit records and pump service history a well drilling company should digitize first, so AI agents answer from verified data.

An AI agent for a well drilling company is most useful when it can answer service questions from the company's own records: how deep the casing goes, where the pump was set, whether the last permit inspection was signed off. Those answers depend on driller logs, permit files and pump service history that are complete, dated and linked to the right well. Digitize those three record sets first, in that order, and give the agent read access to verified entries before it is allowed to draft or change anything.
Key takeaways
- Start with driller logs, permit files and pump service history, because most service questions depend on one of the three.
- Store records as dated entries linked to a stable well identifier, not as scanned folders that can only be searched by customer name.
- Give the agent read access to verified records first, and keep every write action behind a named human approval.
- Mark each record with who verified it and when, so the agent can separate confirmed facts from older notes and guesses.
- Choose one recurring service question the office answers every week and use it as the success criterion for the first agent.
Why well records decide what an agent can answer
A well is buried, and the pump that matters sits at the bottom of a hole that nobody in the office can see. When a homeowner calls about low pressure or a dry tap, the coordinator has to reconstruct the history from memory, a clipboard in a truck, or a technician's text message. An agent placed on top of that situation will either guess or hand every question back to a person, which is the outcome the owner was trying to avoid.
The practical fix is not a smarter model. It is a record base in which each well has one home, every document and visit points to it, and every fact carries a date and a source. Retrieval-based designs, described in the Google Cloud reference architecture for retrieval-augmented generation, pull the relevant records at question time. They only work well when those records are well identified, current and trustworthy. The agent can be no more reliable than the file it reads.
Driller logs: capture what a later service call will need
The driller's log is the most valuable document a well company produces, and it is often the hardest to use because it was written for the regulator and the crew, not for the person who opens it years later. Transcribe the fields that matter for future service into structured entries, and keep the scanned original attached.
- The property address and a stable well identifier that links to the permit, the completion report and every later visit
- The drilling date, the crew, and the driller's license reference as recorded on the job
- Depth intervals with the material described at each interval, because repair and replacement planning depends on it
- Casing material, diameter, and the depths where casing and grout were placed
- The screen or open interval and the depths where water was encountered
- The static water level and the yield or pump test result, with the test method noted
- The final completed depth and any deviation from the planned design
When a field is unreadable or missing, record it as unknown rather than leaving it blank. An agent can tell a coordinator that the casing depth is unknown, which is useful. It cannot be useful if it silently treats an empty field as zero.
Permit records: the paper trail that shows the job was done properly
Permit files tell you what was approved, what was inspected and what was signed off. For a well company they usually include the construction permit application, location and setback notes, inspection dates and results, the final sign-off, abandonment reports for older wells on the same property, and correspondence with the local authority. Requirements differ by state, province and municipality, so the exact list should come from your own authority's rules and your own counsel, not from a generic template.
The critical step is linking each permit to the well identifier used in the driller log. A permit filed under the property owner's name, with no well number, cannot answer the question a future technician will ask, which is whether the work on that well was inspected. Keep a short status field on each record, such as open, inspected, signed off or pending correspondence, and update it when the authority responds.
A permit that cannot be found from the well record is, for an agent, the same as a permit that does not exist.
Pump service history: the record homeowners call about
Pump service is where the records earn their keep. Most calls are about pressure, cycling, sediment, taste, or a system that stopped, and each of those questions depends on knowing what was installed and what was done since. Keep one history per well with the following entries:
- Installation date, pump make and model, and the setting depth as installed
- Control box, pressure switch settings and the type and precharge check of the pressure tank
- Each service visit, with the date, the technician, the findings and the parts used
- Water quality tests and any disinfection event, with results and the lab or method used
- Any change to the system, including a new pump, a new tank or a shift in the water table noted during a visit
Setting depth matters more than many owners expect. A replacement pump that is set without reference to the last recorded depth, the static level and the screen interval can run dry or fail early. An agent that can show the last setting depth beside the most recent static level gives the coordinator a grounded starting point before anyone climbs into a truck.
Attach every pump visit to the well, not to the customer, so the history survives a change of owner.
A digitization order a small team can finish
The most common mistake is starting with the oldest filing cabinet. The better order is to start with the wells the company services most often, then extend the record base outward as the team gains confidence in the structure. A workable sequence looks like this:
- Choose the active wells first, meaning the properties with a service call or a permit issue in the past year
- Create one record per well with a stable identifier and a short status field
- Transcribe the driller log fields listed above, marking anything unreadable as unknown
- Attach the permit and completion documents, and note who verified each one and when
- Log every pump visit against the well record from that point on
- Set edit rights so that only named staff can change a verified field, and review those rights regularly
This is slow in the first weeks and then faster, because each new visit adds to a record that is already structured. Old wells can be added as they come up for service, which avoids a large archive project that stalls before anyone uses it.
How an agent answers from verified records
Illustrative scenario: a homeowner on a rural property reports that the water pressure has dropped over a few days and asks whether the pump can be raised. The agent reads the well record. It finds the setting depth from the last pump installation, the casing depth from the completion report and a static level from the most recent visit. It also notices that the static level is marked as older than the latest service note and cannot be confirmed. The agent drafts a reply for the coordinator that states the known depths, names the static level as unverified, lists the records it used, and proposes a site visit. It does not promise a repair cost, and it does not approve the work.
The design principle behind this is simple. The agent answers with citations to the record, it marks uncertainty explicitly, and a person decides what happens next. Record text also needs care. A technician's note or an imported document can contain text that looks like an instruction, and the agent should treat anything in a record as data to read, not as a command to follow, a risk the OWASP guidance on prompt injection describes in detail.
How Be Human approaches the records work
Be Human does not begin with a model choice. It begins by defining one useful role, such as a service coordinator's assistant for pump calls, then identifies the memory and tools that role needs: the well records, the permit status, the visit history, and the scheduling tool the team already uses. Approval boundaries are written down before any action is enabled, so that drafting a reply is distinct from sending it, and changing a verified field is distinct from proposing a change. The success criterion is observable, for example whether the coordinator can answer a recurring pump question from the record without calling the technician. Managed operations then cover monitoring, recurring evaluations, permission review, and incident response when a record is wrong or a permission has drifted. The same records discipline appears in other service trades, as in the elevator maintenance callback and inspection example and the plumbing dispatch and callback model, where the quality of the history decides the quality of the answer.
If you want to find the first task worth handing to an agent in your own operation, take the free Be Human AI diagnostic.
Frequently asked questions
Do we need to digitize every old driller log before an agent can help? No. Start with the wells that have recent service calls, permits or pump work, because those are the records the team uses every week. Older wells can be added when a call or an inspection brings them back into view. An agent that is clear about which records are missing is more useful than one that waits for a complete archive.
Can an agent answer pump questions from a technician's text messages? It can read them, but text messages should be treated as unverified notes until a person confirms the facts and transfers them into the structured history. The agent can show that a note exists and when it was written, which helps the coordinator decide whether to check it. Confirmed facts should live in the well record with a verifier and a date.
Who should approve an agent's answer about a permit or a repair? A named person in the office or a licensed professional should approve any answer that commits the company to work, a cost, or a compliance position. The agent can draft the reply and cite the records behind it, but the approval step belongs to a person with authority over that decision. Permission reviews should confirm that the boundary still holds as the agent's tools change.
What is the first sign that the records are ready for an agent? The first sign is that a coordinator can find the well, the latest driller log fields, the permit status and the last pump visit in one place without asking anyone. The second sign is that unknown fields are visible as unknown. When both are true for the wells the team services most often, the first narrow task is ready to be tested against a clear success criterion.
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.


