[ 2026-07-26 ]
Beyond dashboards: making geological data conversational
What happens when geological data becomes accessible enough to question?
Raw geological data is not particularly scarce.
Its quality varies enormously and it is often stored in frustrating places, but most large exploration and mining organisations already collect substantial quantities of information.
The range of quantitative data associated with a drilling programme can be enormous: assays, geological logs, downhole geophysics, surveys, drilling activity, machine telemetry, photographs and operational notes.
In some cases, new data is being generated every few seconds.
Despite this, only a fraction of that information is likely to be used regularly in day-to-day decision-making.
The problem is therefore not necessarily the availability of data.
It is the ability to find, combine and interpret the relevant evidence while a decision still matters.
The limits of reporting
The traditional approach to making operational data accessible is through reporting platforms such as Power BI and Tableau.
These tools are extremely useful for formal reporting, agreed metrics and questions that need to be answered repeatedly.
But reports are normally built around questions that have already been defined.
Someone wants monthly drilled metres, so a report is created.
Someone else wants the same number by contractor, project or drilling method, so additional views are added.
Other users want downtime, expenditure, hole-completion information or performance against plan, and the reporting environment gradually expands.
This creates two problems.
The first is maintenance.
Geological and operational datasets change frequently. Fields are renamed, classifications evolve, relationships change and new sources are introduced.
Reports need to evolve with them.
The second problem is more fundamental.
A dashboard can only answer questions that somebody anticipated when it was designed.
That works well for monitoring known metrics.
It works less well when someone notices something unusual and wants to investigate it immediately.
Finding a column is not understanding it
Natural-language data tools are beginning to make it easier to locate datasets, generate queries and retrieve information.
That is useful.
But finding a field is not the same as understanding what it represents.
A system might locate a column containing iron assay values and correctly identify what those values measure.
It does not automatically understand why a particular assay result matters to the design of a drilling programme.
For a metallurgical diamond hole, a change in grade might affect decisions about sample representativity, geological boundaries, variability or the quantity of material required for testwork.
In an RC programme, the same geological variable might instead influence infill priorities, target boundaries or decisions about additional drilling.
The number is the same.
Its significance depends on the geological setting, the objective of the programme and the decision being made.
A technically correct query can therefore still produce a geologically meaningless answer.
This is why I think the more interesting problem is not simply data access.
It is interpretation.
Making geological data conversational
Imagine that a geologist could ask questions directly of a collection of governed geological and operational datasets.
Instead of first identifying the correct report, application or database table, the user begins with the geological or operational question.
For example:
Which holes were completed this week?
Where are drilling conditions repeatedly deteriorating?
What were the main causes of downtime during a particular programme?
Did a change in drilling performance coincide with a geological boundary?
Do observations recorded by field crews correspond with patterns visible elsewhere in the available data?
Some questions require little more than a simple aggregation.
Others require spatial analysis, chronology, text interpretation or the combination of several different forms of evidence.
That distinction matters.
The interesting opportunity is not simply replacing SQL with natural language.
It is making it practical to investigate relationships that would otherwise take too long to explore.
Once access becomes easier, users may stop asking only for predefined KPIs.
They can begin asking about chronology, relationships, causes and uncertainty.
The semantic layer matters more than the interface
Natural language can make data easier to access, but it does not remove the underlying complexity.
If anything, it makes data governance more important.
A useful system needs to understand how datasets relate to one another, which fields should be used and what common terminology actually means.
Consider something as simple as drilled metres.
Depending on the organisation and the question being asked, this could refer to planned metres, reported metres, validated metres or a figure derived from another operational system.
Those distinctions may be obvious to an experienced employee.
They are not necessarily obvious from the database schema.
The same problem appears throughout geological data.
A blank value might mean that information has not yet been entered, that an activity did not occur or that a particular field is not relevant.
A hole may have several identifiers across different systems.
A status field may describe the current state of a hole without preserving the history of how that state changed.
An operational note may contain useful evidence but be incomplete or highly subjective.
A trustworthy conversational system therefore needs to recognise ambiguity rather than simply producing an answer.
Access without interpretation is simply a faster route to the wrong answer.
This also creates an interesting secondary benefit.
Every question that cannot be answered exposes something about the organisation's data.
It may reveal a missing dataset, an unclear definition, a broken relationship between systems or important knowledge that currently exists only in the head of an experienced employee.
Conversational access can therefore become a form of audit.
It reveals not just what information an organisation possesses, but how well that information is actually understood.
How could this change field geology?
Field geologists collect large amounts of information but often have limited time to investigate the wider context around an observation.
A geologist may notice an unexpected change in penetration rate, an unusual contact, a deviation in hole trajectory or a recurring drilling problem.
Investigating that observation properly might require retrieving nearby logs, surveys, assays, photographs, drilling records and previous field notes.
Those records may exist.
The difficulty is finding and reconciling them quickly enough to influence the immediate interpretation.
Accessible and well-interpreted data could shorten the distance between an observation and a testable question.
A geologist might ask whether the same drilling signature appeared in nearby holes.
They might investigate whether it corresponds with a mapped or logged geological boundary.
They might compare similar intervals with assay results or search previous field observations for the same behaviour.
This does not replace geological judgement.
A data system did not stand at the rig.
It did not inspect the chips or core, speak to the driller or experience the combination of small observations that made something feel unusual.
What it can do is make more of the surrounding evidence available while that judgement is being applied.
field observation → question → supporting evidence → revised interpretation
If less time is spent locating information and reconciling systems, more time can be spent testing hypotheses.
The role of the geologist may therefore move further towards interpretation rather than information retrieval.
Their ability to ask a meaningful question becomes more important, not less.
Machine data as geological evidence
There is also a broader question about what should count as geological data.
Machine and operational datasets are normally collected for operational purposes.
But some of those signals may indirectly contain information about the geology being encountered.
Changes in penetration rate, drilling behaviour, vibration, energy use or other operating characteristics may correspond with changes in ground conditions.
That does not mean that an operational signal should automatically be interpreted geologically.
There are many competing explanations.
Equipment condition, operator behaviour, drilling parameters and maintenance issues can all create similar patterns.
But the availability of these datasets creates the opportunity to test whether relationships exist.
Machine telemetry may therefore become another source of evidence that can be compared with geological observations rather than viewed solely as an operational dataset.
From answering questions to noticing patterns
The next step is systems that do not always wait for somebody to ask a question.
Incoming data could potentially be compared with historical behaviour to identify unusual conditions.
Changes in drilling performance could be compared with previous holes.
New assay results could be checked against geological expectations.
Repeated operational behaviour could be assessed against earlier events.
The useful output would not simply be an alert.
It would explain what changed, where it occurred and which evidence makes the observation interesting.
There is an obvious danger.
Poorly designed systems could generate enormous numbers of weak correlations, unnecessary notifications or apparently confident conclusions based on incomplete information.
More data does not automatically create better decisions.
The objective should therefore not be to tell users everything that can be found.
It should be to identify the relatively small number of observations that deserve human attention.
Better questions, not just faster answers
It is tempting to think of conversational data access primarily as a more convenient reporting interface.
I suspect its more important effect may be the questions it allows people to ask.
What relationships have never been investigated because the relevant evidence sits in separate systems?
Could operational data contain useful geological signals?
How much field knowledge could be retained if observations were easier to connect with later evidence?
Can uncertainty be presented alongside an answer rather than hidden behind it?
Will easier access improve geological decisions, or simply increase confidence in imperfect information?
Static reports will remain useful.
There will always be recurring metrics that need clear definitions, stable calculations and consistent presentation.
But exploratory geological work is different.
It involves following observations, testing explanations and combining incomplete pieces of evidence.
The interesting question is therefore not whether AI can query geological databases.
It is what geologists might notice when the evidence surrounding their work becomes accessible enough to question.
