Lodemark | On Artificial Intelligence
What we built.
A single reconciled record of an entire mineral estate, and on top of it, a way to use AI we can trust. That foundation is the difference.
One reconciled record, built and owned since 2011.
We built the record first. One place holding everything an owner has, inherited or purchased: the mineral, leasehold, and working interests; the units, properties, and wells; the divisions of interest; the leases, land documents, and contracts; the authorizations for expenditure. Beside all of it, what the operator reported producing and selling, set against what the state's filings say, and every dollar of revenue earned and expenses paid.
Most of what an owner is missing sits in six different places, and nobody has ever put them together. We did, and we keep it current.
Because we own the foundation rather than rent it, and because we can trust what it tells us, we can do things that a firm running someone else's software cannot.
What owning the foundation lets us do.
We use AI for the mechanical part: the assembly and transformation underneath the work, in hours rather than days. Every report then goes out in the format, on the cadence, and at the depth a particular stakeholder will actually read, rather than to one template. Where an owner needs the record to speak to software they already run, we build that connection, so the numbers tie out on both sides.
The foundation carries the load. The people carry the judgment.
Why we do not trust it blind.
AI is very good at producing language, and it will not reliably tell you whether it is right. It can hand back something fluent, confident and wrong. In this work the output is money owed to a client, or a forecast a board will budget from. An error nobody catches becomes a quiet loss that compounds year over year. So, we built the discipline that lets us use AI without taking on that risk.
The rules we do not bend
Every number ties to a line item.
AI never owns a number that has to reconcile. Where a figure has to tie to the penny, it ties to the penny, checked against the source record before the work is relied on. No tolerance band wide enough to swallow a real difference.
What produces the work cannot also vouch for it. When we use AI to assemble a working draft, a workbook or a data load, a person pulls the raw figures from the client's own statements. AI arranges them; it never supplies them. Otherwise a script the AI wrote ends up as the authority on numbers it produced itself.
When we use AI to check AI, we use a separate system in fresh context, because AI reviewing its own work is the weakest possible check. But the checks we lean on hardest are not AI at all. They are deterministic: a rule that returns the same answer every time, pass or fail, no judgment and no probability, code rather than AI. A check earns its place by being tested, verified to behave the same way every time, and locked. Once locked, it runs on its own. Until then the work stays a draft we check by hand against the record.
Judgment stays with people. The repetitive work and the cross-referencing are what the machine is for. What a finding means, and what to do about it, is ours.
None of this is new for us.
The firm has been management consultants and engineers to mineral owners since 1974, adopting every useful tool that came along and checking what it said against the record. AI is the newest tool. We put it to work the same way.
We do not bet on one vendor.
We do not bet the practice on any single AI vendor. The discipline sits above the tools, so they can change underneath it without the standard moving. As they improve, more of the work fits a check we can lock, so we do more with AI each year without loosening what has to tie out. The leaders today may not be the leaders in two years. The foundation does not move when the tools do.
We do not claim the work is ever perfect. No process removes the possibility of error. The discipline exists to catch errors against the record before they reach you.
The point
This is what we actually built: not a clever use of AI, but a foundation solid enough to build clever things on, and to keep building on. It is why we can serve institutions that would ordinarily need several times the people, and why we can keep extending what we do without the ground shifting underneath us.
The hard thing is not making AI do more. It is building something you can trust enough to keep building on.