WillCo Digital Drilling Intelligence - Edition #17: Decision Trust in Drilling Automation. Sep 14 - Sep 20, 2026
🛢️ Top Story
⏱️ Corva Reports Up to 1.2 Rig Days Saved per Bit-Trip Decision in Oman
Corva-reported case study. WillCo reads it as decision trust in practice: a predictive bit-wear model applied to the oldest question on the rig floor - keep drilling, or trip for a new bit?
A prediction is only worth the decision someone makes with it (WillCo analysis). Corva, a real-time drilling data and applications company, published a case study describing a predictive bit-wear workflow deployed with a major operator drilling in the abrasive formations of Oman. The workflow estimates bit condition while drilling and supports the trip-or-continue decision at each bit-trip point, so the operator can avoid pulling a bit that still has useful life and avoid drilling on one that is spent. Corva reports the workflow saved up to 1.2 rig days per bit-trip decision by reducing unnecessary trips.
The engineering interest here is not the model in isolation but where it sits in the workflow (WillCo analysis). Bit-wear estimation is a long-standing problem; what this case describes is a prediction delivered at the exact moment a high-cost, hard-to-reverse decision is made, and, on WillCo's reading, an operator willing to act on the recommendation at the bit-trip point. That is the decision-trust frontier: the step from a dashboard number to an action taken on it.
Multiply 1.2 rig days by your own verified spread day-rate. That is the number you personally avoid on every trip you do not take too early or too late (WillCo analysis; run your own figures, since the source discloses no dollar value and does not name the rig environment).
This is exactly the kind of prediction to pressure-test before it reaches a driller, using the five questions in Recommendation 1.
Confidence: vendor case, one field - directional, but the decision-trust pattern is the point. (Sourcing detail in References.)
📰 Key Developments This Week
🤝 YPF Doubles Down on Real-Time Well Construction With Corva
Renewed Corva collaboration across Vaca Muerta, framed toward connected, predictive, and autonomous operations.
If you run a basin-scale program, the YPF-Corva renewal is a live example of the buy-versus-build call you own. (WillCo analysis) It helps to picture the two ends of that spectrum, with plenty of hybrid arrangements in between:
Standing capability that compounds: a real-time platform institutionalized across every well in the program, its value building with each one.
Point tool that ships once: a single workflow license that delivers value on the wells it touches and stops there.
YPF and Corva renewed their real-time well-construction collaboration and framed the next phase as increasingly connected, predictive, and autonomous operations across Argentina's Vaca Muerta. Corva frames the shared vision as continuing to establish YPF's Real-Time Integration Center (RTIC) as a "digital operating system" for well construction - an institutional commitment, not a one-off license. The cost signal to watch: whether spend is structured as a recurring platform investment or a single purchase (WillCo analysis).
The renewal is described in directional terms: a shared real-time data foundation and applications supporting well-construction decisions at scale. No well counts, autonomy levels, performance percentages, or contract values were disclosed (state as undisclosed; do not estimate). Sources: Corva blog, Sep 3, 2026, corroborated by YPF corporate news (novedades.ypf.com) within the same news window - operator and vendor communications, directional and not independently benchmarked.
The link to the top story is deliberate (WillCo analysis): the Oman bit-wear case shows a single trusted prediction saving rig days on one decision; the YPF renewal is what it looks like when an operator tries to institutionalize that same predictive-decision layer across a whole basin program. Scaling decision trust is an organizational commitment, not a software purchase.
⚡ Hours to Under Five Minutes on an Offshore BOP Troubleshoot (Vendor-Reported)
Aquila Engineering's Iris, a domain-grounded multi-agent AI, compressed one offshore blowout-preventer troubleshooting task.
Hours to under five minutes: that is how far an offshore crew compressed a blowout preventer troubleshooting task using Aquila Engineering's Iris, a multi-agent AI that answers from a curated engineering knowledge base. A blowout preventer problem is well-control-critical rig time, so that compression lands where minutes are most expensive to lose.
Iris consolidates domain knowledge - bulletins, technical reports, and lessons learned - so a frontline engineer can query it directly instead of hunting through documents. Aquila, led by President Jose Meraz, described the platform in trade coverage of the IADC Advanced Rig Technology community; as Meraz put it, "AI can compress hours of offshore troubleshooting into minutes by giving frontline engineers instant access to the domain-specific knowledge they need." Source: Drilling Contractor, Sep 8, 2026 (a vendor-supplied single-case example, not an independently measured average).
(WillCo analysis) This is the troubleshooting face of decision trust, and the buyer takeaway is blunt: the differentiator is not the language model but the curation and accountability of the domain knowledge behind it, which is the question to put to any vendor in this category. A blowout preventer event is exactly where an engineer must trust the answer before acting, and exactly where the failure mode of a generic AI - a confident but wrong response - is unacceptable. Operationally, compressing a BOP troubleshoot from hours to minutes can reduce diagnostic delay and potentially reduce the non-productive time associated with BOP troubleshooting, at the point where downtime is most dangerous. Do this today: ask your current assistant vendor two questions - what corpus is the assistant grounded in, and how is a wrong answer caught before an engineer acts on it?
📊 Data-Driven Trends
First: the frontier is moving from the interface to the decision (WillCo analysis). Edition #16 documented interoperability - getting rig systems to exchange commands. This window's items all sit one layer up, at the point of decision. Connectivity was the last bottleneck; trust in the decision is the next one.
Second: this is a vendor-and-operator-reported window, so the discipline is validation (WillCo analysis). Several of this edition's strongest signals are vendor- or operator-reported and lack independent performance benchmarks, and the operator is unnamed in the strongest one. That is not a reason to ignore them - it is the reason to standardize the questions you ask before adopting: model inputs and error bounds, training-data provenance, independent validation, and clear accountability for a wrong recommendation.
Third: human oversight is being reframed as engineering, not sentiment (WillCo analysis). The IADC panel's argument - detailed in the WillCo Perspective below - is that keeping a person in the loop is a design requirement driven by data quality and multi-loop coordination, not a temporary comfort measure on the way to full autonomy.
🧭 WillCo Perspective: Decision Trust Is the Next Bottleneck
Automation is not autonomy, and human oversight remains part of the architecture. An IADC Advanced Rig Technology panel made the case that closing the loop in drilling still requires human expertise, and framed that requirement in engineering terms rather than as caution. Panelists included Kowalchuk of Taurex Drill Bits, Carroll of SLB Well Construction, and John de Wardt of de Wardt and Company. Their argument: a modern well is not one control loop but several running at once - trajectory, drilling-dysfunction mitigation, equivalent circulating density, and managed pressure - and coordinating those loops under uncertainty is where human judgment remains a design requirement. Source: Drilling Contractor, Sep 9, 2026.
The data-quality point is the one to internalize. De Wardt cited a 2014 study ("Chesapeake: The Role of Data in Drilling"), presented for an SPE DSATS workshop, that found approximately 24 percent of the drilling data examined across six rigs was inaccurate (a 2014 six-rig study figure, not a current industry-wide benchmark). (WillCo analysis) That study figure reframes the entire decision-trust discussion: a predictive bit-wear call, an autonomous ECD adjustment, or an AI troubleshooting answer is only as trustworthy as the data feeding it. This is the garbage-in, garbage-out problem in a higher-stakes form: WillCo’s Reaming Is Not Washing showed how a single mislabeled hole-enlargement record quietly poisons every downstream KPI and model, and an autonomous decision only raises the cost of that same bad input. If a meaningful fraction of that data is wrong, the correct engineering response is not to remove the human but to design the human’s oversight role deliberately. The panel’s position is that oversight is where accountability for that residual uncertainty lives.
WillCo Perspective: The practical test is narrower than trust in the abstract: at 3 a.m., does the engineer have enough confidence in the model to act on its recommendation? In most cases today, not without doubt, and for a concrete engineering reason. Many predictive models are purely data-driven: they fit patterns in the data without representing the underlying physics or the limits of the engineering calculation, so their outputs carry uncertainty the number on the screen never shows. Add machine-learning models trained on one set of wells and extrapolated to a different environment without proper re-training, and that uncertainty compounds. WillCo has argued this before: Why Pure Machine Learning Is Not Enough for Drilling Optimization makes the case that data-driven models need physics and engineering constraints to be trustworthy on a real rig. So the doubt an engineer feels at the bit-trip point is justified engineering judgment, not timidity: the recommendation deserves scrutiny precisely when the model cannot show its own blind spots.
This connects directly to the top story (WillCo analysis). The reason the Oman bit-wear case matters is that it indicates a recommendation influenced a real trip-or-continue decision. The reason the IADC panel matters is that it specifies the conditions under which that kind of trust is earned: validated data, understood error bounds, and a named human accountable for the call. The two are the same argument seen from opposite ends - one shows trust paying off, the other shows what has to be true for it to be safe.
🛠️ Practical Recommendations
1️⃣ Recommendation 1 (Do this quarter - one page) - Write the Questions Before You Trust the Model
Standardize a short due-diligence checklist for any predictive or AI recommendation that will influence an operational decision. Call it The 5-Question Trust Test - the reusable artifact of this edition: screenshot it, keep it on hand, and run it against every vendor case that crosses your desk. The five questions:
What are the model's inputs?
What are its error bounds?
Where and when did the training data come from?
Is there independent validation beyond the vendor's own case?
Who is accountable when the recommendation is wrong?
This checklist is WillCo's own diligence framing (WillCo analysis): informed by, but not identical to, the five evaluation questions de Wardt posed in the panel. The Corva Oman and Aquila cases are both worth evaluating; this checklist is how you evaluate them on the same basis.
2️⃣ Recommendation 2 (Before your next predictive-workflow trial) - Define the Decision, Not the Dashboard
Before you trial any predictive workflow, name the decision it must change. For bit-wear estimation or anything like it, specify three things in advance:
The exact decision it will inform (for example, the trip-or-continue call at a bit-trip point).
The baseline you are measuring against.
How you will attribute any rig-time saving.
And define that decision as a cost comparison, not a trip you avoid. The trip-or-continue call is techno-economic: continuing drills the remaining footage at the current, declining ROP with no trip cost; tripping pays for a round trip but restores a higher ROP. Specify up front that the model must inform the lower cost per foot to the next section point, and name the inputs it has to sharpen: remaining footage, the current bit’s ROP trend against a fresh bit’s expected rate, the rig spread rate, and full round-trip time. (WillCo analysis.)
A prediction that does not change a decision produces no value regardless of its accuracy.
3️⃣ Recommendation 3 (For your automation roadmap) - Design the Human's Role
Treat human oversight as a system component to design, not a phase to remove. For each control loop you are automating - trajectory, dysfunction, ECD, MPD - define:
What the operator sees.
The conditions that trigger intervention.
Who holds accountability for the outcome.
The IADC panel's data-quality point makes this a reliability requirement, not a matter of preference.
🎯 Competitive Intelligence
🔮 Corva - Predictive Workflows and a Basin-Scale Operator Relationship
Press them on:
Independent validation of the bit-wear figure.
What a program-level (not per-license) engagement actually costs over a full basin campaign.
Corva paired a quantified case (Oman bit-wear, up to 1.2 rig days per decision) with a renewed, program-level operator relationship (YPF, Vaca Muerta) in the same window. (WillCo analysis) That combination positions it as moving from point applications toward an institutionalized predictive-decision layer inside an operator's program - in WillCo's commercial interpretation, a stickier position than selling a single workflow.
🤖 Aquila Engineering - Domain-Grounded AI for Frontline Troubleshooting
Press them on:
Exactly what corpus Iris is grounded in.
How each answer is validated.
How a wrong answer is caught before an engineer acts on it.
Aquila competes on curated, domain-specific knowledge for time-critical troubleshooting rather than general-purpose AI. (WillCo analysis) Where operators are rightly wary of confident-but-wrong answers, grounding and accountability of the knowledge base are the defensible differentiators.
🏛️ The Governance Voice - IADC Advanced Rig Technology
For your shortlist: treat ART's published positions as a cross-industry reference point you hold every vendor's autonomy claims against, not a vendor to shortlist.
The IADC ART community remains where the industry negotiates the human-machine boundary in the open, this window through a panel arguing human oversight is an engineering requirement. (WillCo analysis) For operators, ART outputs are a useful, non-vendor industry reference point when setting internal standards for what autonomy is allowed to decide unsupervised.
👀 What We're Watching Next
An autonomous in-line rheometer for live ECD management, now confirmed on the program (SafeKick and Seadrill). If you run MPD, this is the piece to track. SPE 233721, "Closing the Loop: A Novel Autonomous In-Line Rheometer for Real-Time ECD Management," is on the technical program of the IADC/SPE Managed Pressure Drilling and Underbalanced Operations Conference (London, Sep 15-16, 2026), which overlaps this publication week. Authored by SafeKick and Seadrill, it describes a compact in-line rheometer with an autonomous pumping system that generates a rheogram every five minutes and feeds it directly to the hydraulic model, moving fluids control from periodic manual mud checks toward a closed loop. The paper reports laboratory validation and a field installation on a deepwater drillship. Conference-program-verified; field-performance results not yet independently reviewed. Full provenance in References item 5.
D-WIS interoperability follow-through from Edition #16. Watch for the next workgroup output on the H&P-led effort to standardize how steering commands pass between rig systems. If that interface layer consolidates, predictive and AI tools stop being locked to one vendor's stack and can be trusted and swapped across your whole fleet - which is what makes the decision-trust questions in this edition matter at scale rather than one rig at a time.
📚 References and Sources
Corva predictive bit-wear workflow, Oman - Corva case study. https://www.corva.ai/case-studies/corvas-predictive-bit-wear-workflow-saves-up-to-60k-per-bit-trip-decision-in-oman (WillCo cites the "up to 1.2 rig days per bit-trip decision" figure stated in the case-study body; the dollar figure appears only in the URL and is not substantiated in the text, so it is not cited.)
YPF renews real-time well-construction collaboration with Corva, Vaca Muerta - Corva blog, Sep 3, 2026. https://www.corva.ai/blog/ypf-corva-real-time-well-construction - corroborated by YPF corporate news (novedades.ypf.com), dated Sep 1, 2026 per the captured page, within the same news window; still characterized as directional operator PR, not an independently benchmarked source. https://novedades.ypf.com/ypf-corva-renovacaciondelacuerdo.html
Aquila Engineering "Iris" multi-agent AI platform, offshore troubleshooting - Drilling Contractor, Sep 8, 2026. https://drillingcontractor.org/aquila-ai-platform-speeds-up-offshore-troubleshooting-79881
AI still requires human expertise to close the loop, says industry panel (IADC Advanced Rig Technology) - Drilling Contractor, Sep 9, 2026. https://drillingcontractor.org/ai-still-requires-human-expertise-to-close-the-loop-says-industry-panel-79890
IADC/SPE Managed Pressure Drilling and Underbalanced Operations Conference 2026, technical program - IADC conference page (London, Sep 15-16, 2026). https://iadc.org/event/iadc-spe-managed-pressure-drilling-underbalanced-operations-2026/ (Program lists SPE 233721, "Closing the Loop: A Novel Autonomous In-Line Rheometer for Real-Time ECD Management," authored by SafeKick (Mauricio Santos, Helio Santos, Jason Hannam, Marlon Moura) and Seadrill (Russell Stewer, Alec Spedding). The abstract describes a rheogram generated every five minutes, direct hydraulic-model integration, laboratory validation, and a field installation on a deepwater drillship. Field-performance results are not yet independently reviewed, so the item is treated as a watch item.)
Corva company insights index (source-dating reference for the Oman case study). https://www.corva.ai/company/insights
Prior WillCo Digital Drilling Intelligence editions and archive - https://willcodrilling.com/blog
WillCo analysis, forecasts, and competitive interpretation in this edition are labeled inline as "(WillCo analysis)" and are the editorial view of WillCo Drilling Consulting, not statements of fact from the sources above.
📖 Glossary of Acronyms
About this report: The WillCo Digital Drilling Intelligence Report is a weekly synthesis of verified developments in digital drilling and well construction, published by WillCo Drilling Consulting. Analysis, forecasts, and recommendations reflect WillCo's independent engineering perspective.
Comments