- An investigation is a decision tool, not a data-collection exercise. If you cannot name the decision each borehole supports, you are buying data you will not use.
- Between 80% and 85% of building failures and damages are related to unforeseen ground conditions, yet ground investigation typically costs 0.05–2% of project value.
- CPT is roughly three to five times more repeatable than SPT — coefficients of variation of 5–10% for cone resistance, against a standard deviation of about 3 N₆₀ for SPT.
- The three-phase ladder — desk study, preliminary, detailed — exists to reduce uncertainty progressively, not to satisfy a procurement checklist.
- AI is entering site characterisation, but data scarcity and interpretability remain unsolved. It accelerates interpretation; it does not replace the ground model.
Most ground investigations fail before the first rig arrives — at the moment someone writes a scope that says “six boreholes to 20 m” without stating what those boreholes are supposed to prove. This guide reorganises the subject around the four questions a ground model must answer, the three-phase ladder that answers them, and the cost arithmetic that explains why the cheapest investigation is almost always the most expensive decision on the project.
On this page
- The four questions a ground model must answer
- The three-phase uncertainty ladder
- Phase 1 — Desk study: the cheapest data you will ever buy
- Phase 2 — Preliminary investigation: confirming the model
- Phase 3 — Detailed investigation: designing against the model
- SPT versus CPT: the repeatability argument
- The reporting gap: where investigations fail silently
- The cost–failure paradox
- AI and the digital frontier
- A scope-writing checklist
- Frequently asked questions
1. The Four Questions a Ground Model Must Answer
Every geotechnical investigation, from a single-house extension to a metro line, exists to answer four questions. The techniques change with scale and ground condition, but the questions do not.
1.1 Why “What Is There?” Is Not Enough
A borehole log tells you the sequence of strata at one point. It does not tell you how the strata behave under load, how the groundwater responds to excavation, or which failure mode is credible. Those are the questions that govern design, and they are the questions that most investigation scopes fail to state.
The distinction matters because it changes what you buy. If the question is “what is there?”, boreholes and logging are sufficient. If the question is “how strong is it?”, you need in-situ testing and laboratory work. If the question is “what can go wrong?”, you need a ground model — an interpreted hypothesis that can be tested, updated, and designed against. Most procurement documents buy the first and assume the fourth.
The Scope Test
Read your investigation scope and ask: which of the four questions does each line item answer? If a line item answers none of them, it is either a compliance requirement or a waste of money. If an entire question has no line item, the investigation is incomplete regardless of how many boreholes it contains.
2. The Three-Phase Uncertainty Ladder
Ground investigation is not a single event. It is a sequence of phases, each designed to reduce uncertainty to the level the next decision requires. The phases are not bureaucratic stages; they are decision gates.
| Phase | Decision it supports | Typical techniques | Uncertainty reduced |
|---|---|---|---|
| Desk study | Is the site viable? What are the risks? | Geological maps, historical records, aerial imagery, previous reports, site walkover | Whether to proceed, and what to investigate |
| Preliminary investigation | What is the general ground model? | Widely spaced boreholes, trial pits, geophysics | Stratigraphy and first-order properties |
| Detailed investigation | Can the design be finalised? | Targeted boreholes, CPT, laboratory testing, monitoring | Design parameters, groundwater regime, failure modes |
Phases should overlap and iterate. A desk study that raises a new question should trigger a targeted addition to the preliminary phase, not wait for the detailed phase.
Why Phasing Is Not a Luxury
A desk study that costs a few thousand pounds can reveal a former landfill, a mining cavity, or a geological fault that would make a detailed investigation at that location pointless. Spending the detailed investigation budget before the desk study is complete is the most common form of value destruction in ground engineering — and it happens on projects of every size.
3. Phase 1 — Desk Study: The Cheapest Data You Will Ever Buy
The desk study is the highest-leverage phase of the investigation, and the most frequently abbreviated. It is also the cheapest: a Phase 1 desk study in the UK typically costs £500–£3,000 depending on the site and the depth of the historical review.
A proper desk study assembles four categories of information:
- Geological — published maps, memoirs, borehole archives, and regional geological history. The geological history matters as much as the map: a site underlain by glacial till behaves differently from one underlain by weathered mudstone, even if both appear as “clay” on a log.
- Historical — old maps, aerial photography, planning records, and industrial history. Former uses are the single most reliable predictor of contamination and made ground.
- Topographic and hydrological — slope angle, drainage patterns, flood records, and groundwater abstractions.
- Regulatory and archival — environmental permits, pollution incidents, mining records, and previous geotechnical reports for the site or its neighbours.
What the Desk Study Actually Delivers
The output is not a file of documents. It is a preliminary ground model — a written hypothesis of the strata, groundwater, and geohazards expected at the site, with the specific uncertainties flagged. That hypothesis is what the preliminary investigation is designed to test. A desk study without a hypothesis is a literature review; a hypothesis without a desk study is a guess.
BS 5930:2015+A1:2020 and Eurocode 7 both require the desk study to be carried out at an early stage and to inform the scope of the intrusive investigation. The standard does not prescribe its depth; that is an engineering judgement based on the consequence of being wrong.
4. Phase 2 — Preliminary Investigation: Confirming the Model
The preliminary investigation exists to test the desk-study hypothesis, not to finalise the design. Its characteristic features are wide spacing, limited depth, and broad coverage — enough to confirm the general sequence of strata and identify where the model is wrong.
The key design decisions at this phase are not “how many boreholes?” but “where should they be, and what should they prove?” A preliminary investigation that places boreholes on a regular grid without reference to the desk-study uncertainties is spending money to confirm what is already suspected.
| Technique | Best for | Limitation at preliminary stage |
|---|---|---|
| Cable percussive boreholes | General stratigraphy, sampling in soils | Slow; limited depth; disturbed samples in granular soils |
| Rotary boreholes | Rock, deep investigation, high-quality sampling | Cost; requires access and water supply |
| Trial pits | Shallow stratigraphy, made ground, services | Depth limited by excavation safety; not suitable below water table |
| Geophysics | Continuity between boreholes, buried features, depth to rock | Indirect; requires borehole calibration; ambiguous in layered ground |
The preliminary phase should always include enough in-situ or laboratory testing to assign first-order parameters. A log without a single strength measurement is a stratigraphic sketch, not an investigation.
The Cost of Confirming the Wrong Thing
A preliminary investigation that places all boreholes on a grid can confirm the general stratigraphy while missing the lens of soft clay, the buried channel, or the fault that controls the design. The uncertainty has not been reduced; it has been averaged out. The correct question is not “have we covered the site?” but “have we tested the hypothesis where it is most likely to fail?”
5. Phase 3 — Detailed Investigation: Designing Against the Model
The detailed investigation produces the parameters the design will actually use. It is targeted, not uniform: it concentrates effort where the preliminary phase showed the ground model is uncertain or where the consequence of being wrong is highest.
At this phase, the investigation must answer the fourth question explicitly — what can go wrong? That means identifying the credible failure modes and testing for the parameters that govern them:
- Bearing capacity and settlement — strength and stiffness profiles, consolidation characteristics.
- Slope stability — shear strength, pore pressure regime, and the location of any weak layer.
- Retaining structures — earth pressure parameters, groundwater response to excavation.
- Pile design — stratigraphy at depth, rockhead, and the presence of obstructions.
- Seismic or dynamic loading — shear wave velocity, liquefaction susceptibility.
The Forensic Evidence
A forensic geotechnical investigation of a multi-storey car park failure in a hilly region found that the investigation had not reached the required depth, the safe bearing capacity had been improperly estimated, and the site had been developed without adequate slope protection or drainage. The structure showed distress during construction and the retaining wall failed after heavy rainfall. Every one of those findings traces back to a question the investigation did not answer.
The detailed investigation is also where monitoring should be installed, not just readings taken. A piezometer read once during the investigation gives a snapshot; a piezometer monitored through a seasonal cycle gives a regime. The difference matters for any design that depends on pore pressure.
6. SPT versus CPT: The Repeatability Argument
The choice between the Standard Penetration Test and the Cone Penetration Test is often presented as a matter of tradition or availability. It is better understood as a choice about measurement error.
| Characteristic | SPT | CPT |
|---|---|---|
| Output | N value (blows per 300 mm), plus a disturbed sample | Continuous cone resistance, sleeve friction, pore pressure |
| Repeatability | Standard deviation of about 3 N₆₀; four to five times less repeatable than CPT | Coefficient of variation of 5–10% for cone resistance |
| Operator influence | High — hammer energy, rod length, borehole diameter all require correction | Low — the cone is pushed at a controlled rate |
| Sampling | Yes — disturbed sample for classification and testing | No — no sample recovered |
| Experience database | Very large; a century of correlations | Large and growing; strong in sands and soft clays |
| Best use | Where a sample is needed; where CPT cannot penetrate | Where continuous stratigraphy and repeatable parameters are needed |
Repeatability values from comparative studies. The practical implication is that a CPT-derived profile is smoother and more reproducible, while an SPT profile carries more scatter that must be interpreted rather than averaged.
The Counter-Intuitive Finding
Because SPT N values are so variable, the most reliable way to obtain an N value in some ground conditions is to perform a CPT and convert it to an equivalent N₆₀ using a validated correlation. The derived value can be more reliable than the directly measured one. That is not an argument for abandoning SPT — it is an argument for knowing which measurement carries the error.
7. The Reporting Gap: Where Investigations Fail Silently
Incomplete and incorrect reporting of subsurface information is one of the major causes of geotechnical risk — and it is distinct from inadequate investigation. A project can drill enough boreholes and still fail because the report does not communicate what the ground model is, what it assumes, and what remains uncertain.
The common failure modes in reporting are consistent across case literature:
- Factual logs without interpretation — a stack of borehole logs is data, not a ground model. The designer is left to construct the model, often without the geological context the investigator held.
- Unstated assumptions — the report presents parameters without recording the correlations, judgements, or analogies used to derive them.
- No statement of residual uncertainty — the report does not say what was not investigated, what was assumed, or where the model is weakest.
- Miscommunication between investigator and designer — the person who saw the ground is not the person who designs against it, and the handover loses the tacit knowledge.
The Reporting Test
Give the report to an engineer who has never visited the site and ask them to state the ground model in one paragraph, including its three greatest uncertainties. If they cannot, the report has failed — regardless of how many pages it contains or how many tests it records.
8. The Cost–Failure Paradox
Ground investigation is one of the cheapest line items on a construction project and one of the most consequential. The arithmetic is stark.
| Metric | Value | Source context |
|---|---|---|
| Ground investigation as % of project cost | 0.05–0.22% for buildings; 0.20–1.55% for roads | Published cost reviews |
| Ground investigation as % of foundation cost | 0.5–2.0% | Typical building projects |
| BRE Digest 322 minimum recommendation | 0.2% of total project cost | Low-rise buildings |
| Case-study finding for adequate investigation | 0.42% of project cost | Low-rise building study |
| Recommended upper bound for high-risk projects | Up to 2% of project cost | Value engineering case studies |
| Building failures related to unforeseen ground conditions | 80–85% | European statistics |
| UK highway contracts: outturn above tender | 28% | 1992 government report |
| Share of that increase attributable to earthworks and unforeseen ground | 44% | Analysis of 17 contracts |
Percentages are not directly comparable across sources; the pattern — a small investigation cost against a large failure cost — is the transferable finding.
The Arithmetic That Ends the Argument
A soil test that costs 0.5–1% of project cost against a foundation failure that costs 10–100% of project cost is not an expense. It is the cheapest insurance on the project. The industry still debates it because the cost of the investigation is visible today and the cost of the failure is contingent — and humans discount contingent losses. The engineering answer is to price the risk, not to discount it.
The paradox is not that clients are irrational. It is that the cost of the investigation is certain and immediate, while the cost of the failure is uncertain and deferred. Procurement systems reward the visible saving. The forensic literature records the invisible cost.
9. AI and the Digital Frontier
AI is entering geotechnical site characterisation through four routes, and each has a different maturity level.
| Application | What it does | Maturity | Principal limitation |
|---|---|---|---|
| Intelligent site investigation | Optimises borehole location and depth from existing data and uncertainty maps | Research to early commercial | Requires a prior ground model to optimise against |
| CPT/SPT interpretation | Predicts soil type and parameters from cone or blow-count data | Commercial in places | Correlations are regional; transferability is weak |
| Predictive modelling of soil behaviour | Surrogate models for constitutive response | Research | Data scarcity; interpretability |
| Digital site characterisation | Integrates geological, geophysical, and geotechnical data into a 3D model | Emerging | Data standards and interoperability |
AI reviews consistently flag data scarcity and interpretability as the two unresolved barriers. The technology accelerates interpretation; it does not create ground truth.
The Rule for AI in Ground Investigation
AI can tell you where to look and what the data probably means. It cannot tell you what is actually under the ground. The output of any AI-assisted interpretation must be tested against physical investigation, and the engineer remains responsible for the ground model. A model that cannot explain its prediction is a hypothesis generator, not a design basis.
10. A Scope-Writing Checklist
- State the decision the investigation supports. “Foundation design for a 12-storey building” is a decision. “Site investigation” is not.
- State the four questions explicitly. What is there, how strong is it, where is the water, what can go wrong. Each must have a line item or a stated reason for omission.
- Do the desk study first, and write the hypothesis. The preliminary investigation should test a stated model, not explore a blank page.
- Place boreholes against uncertainty, not on a grid. Target the locations where the desk-study model is most likely to be wrong.
- Specify the testing, not just the drilling. A borehole without in-situ or laboratory testing answers only the first question.
- Install monitoring, not just readings. A piezometer read once is a snapshot; a monitored piezometer is a regime.
- Require an interpretive report, not just logs. The report must state the ground model, its assumptions, and its residual uncertainties.
- Budget against consequence, not against precedent. A high-consequence site justifies a higher investigation percentage than a low-rise building on benign ground.
- Plan for iteration. If the investigation disproves the model, the correct response is to re-interpret and investigate further, not to proceed on the original assumption.
11. Frequently Asked Questions
How much should a geotechnical site investigation cost?
Typical ranges are 0.05–0.22% of total project cost for buildings and 0.20–1.55% for road projects, or roughly 0.5–2% of foundation cost. BRE Digest 322 recommends a minimum of 0.2%, and a case study found that adequate investigation for low-rise buildings required 0.42%. High-risk sites justify up to about 2%.
Why do so many projects skip or shorten the desk study?
Because its value is invisible until it prevents a problem. A desk study costs £500–£3,000 and produces a written hypothesis rather than a tangible asset. Projects that skip it often spend the same money later on a detailed investigation in the wrong location, or on remedial work after construction reveals a hazard the desk study would have flagged.
Is CPT better than SPT?
For repeatability and continuous profiling, yes — CPT has a coefficient of variation of 5–10% for cone resistance, compared with a standard deviation of about 3 N₆₀ for SPT, making it roughly three to five times more repeatable. But CPT recovers no sample, and SPT has a century of empirical correlations. Most high-quality investigations use both, with CPT for continuity and SPT for sampling and correlation.
What is the most common cause of geotechnical failure?
Unforeseen ground conditions, which account for an estimated 80–85% of building failures and damages in European statistics. The underlying causes are usually inadequate investigation depth, misinterpretation of data, incomplete reporting, or a ground model that was never explicitly stated and therefore never tested.
How deep should boreholes go?
Deep enough to answer the design question, which usually means penetrating all strata that could influence the behaviour of the structure — not merely reaching a convenient refusal. A forensic case study of a car park failure found that the investigation had not reached the required depth, and the safe bearing capacity was consequently misestimated. Depth is a design decision, not a drilling convenience.
Can AI replace a ground investigation?
No. AI can optimise borehole locations, interpret CPT and SPT data, and integrate multiple datasets into a 3D ground model, but it cannot create ground truth. The two persistent barriers are data scarcity and interpretability, and the engineer remains responsible for the model. AI accelerates interpretation; it does not substitute for physical investigation.
What should a geotechnical report contain?
At minimum: a stated ground model, the factual data that supports it, the assumptions and correlations used, the design parameters with their basis, the groundwater regime, the credible failure modes, and a statement of residual uncertainty. A report that contains only logs and test results has not done its job — the interpretation is the deliverable.
Why does phasing matter if the budget is fixed?
Because phasing spends the next increment only when the previous phase has shown it is worth spending. A desk study that reveals a former landfill can redirect the entire intrusive budget to a different part of the site. Phasing does not increase the total cost; it reduces the cost of investigating the wrong thing.
Related reading on BuildersMeasures: Slope Stability Monitoring: Closing the Loop Between Prediction and Failure and Landslide Mitigation: Drainage First, Structures Second.