The Challenge
Glacier Analytics operates in one of the most document-heavy corners of the US energy and real estate market: mineral rights intelligence. Their raw material is not clean data — it is decades of handwritten deeds, county records, and legal descriptions written in “metes and bounds” language that predates modern surveying.
Every deed that entered the pipeline took a trained analyst 6 to 8 hours to process manually. Analysts had to decipher handwriting, extract royalty percentages, reconstruct ownership chains, and translate archaic land descriptions into usable geography. At that pace, the backlog was structural, not temporary.
The cost of manual processing showed up in three places:
- Speed: reports took 3 weeks to become actionable — in a market where lease opportunities are won by whoever moves first.
- Accuracy: royalty calculations carried a 12% error rate, an unacceptable risk when errors flow directly into financial commitments.
- Scale: throughput was capped by headcount. Growing meant hiring and training more analysts, which meant months of ramp-up before any new capacity came online.
Glacier’s leadership did not want a marginal improvement. They wanted the document pipeline to stop being the bottleneck of the business.
Our Solution
OmniMinds deployed a 6-person AI-augmented pod — senior engineers with architecture backgrounds from IBM, TCS, and DXC, working alongside AI agents that handled the repetitive extraction and validation load. The engagement ran 4 months, scoped to outcomes, not hours.
The system we delivered has four connected components:
Fine-tuned LLM extraction. We fine-tuned large language models specifically on deed language and handwriting patterns to extract royalty percentages, grantor/grantee details, and legal terms from scanned documents. A RAG layer grounded every extraction against Glacier’s existing records, so the model checked its answers instead of guessing.
Automated GIS mapping. Metes-and-bounds descriptions — “beginning at the oak tree, thence north 40 chains” — were parsed and converted into mapped parcels automatically. What used to require a human with survey expertise became a Python pipeline feeding directly into Glacier’s GIS layer.
Ownership-chain tracking. We built a tracker that reconstructs the full chain of title across decades of transactions, exposed through a complete API so Glacier’s downstream tools consume ownership data programmatically rather than through spreadsheets.
Satellite imagery processing. Imagery analysis was integrated into the pipeline to cross-reference land use against ownership records, surfacing lease opportunities that pure document analysis would miss.
Every extraction above a confidence threshold flows straight through; anything ambiguous is routed to a human analyst with the model’s reasoning attached. Analysts stopped doing data entry and started doing judgment.
The Results
The numbers within the first quarter of production use:
- 90% faster deed processing. Documents that took 6–8 hours now clear the pipeline in a fraction of the time.
- 13× daily throughput. The same team processes thirteen times the daily volume — without a single new hire.
- $950K saved annually with zero critical errors. The 12% error rate in royalty calculations is gone from the automated path, and the direct labor savings compound every month.
- $8.5M in new leases identified in a single quarter. This is the number that matters most: faster, more accurate intelligence translated directly into revenue opportunities Glacier could act on before competitors even had reports in hand.
The 3-week reporting delay collapsed to same-day. Glacier’s analysts, freed from manual transcription, now spend their time on the high-value work the business actually hired them for: evaluating opportunities.
Why It Worked
This engagement is a clean illustration of the OmniMinds model. We did not staff a bench of billable hours against a vague statement of work. We committed to specific outcomes — throughput, accuracy, turnaround — and built an AI-augmented pod to hit them.
The pod structure mattered. Six senior humans set the architecture, trained the models, and designed the confidence-based routing; AI agents carried the volume work of extraction, validation, and cross-referencing. Neither alone gets to 13× throughput in 4 months. A pure automation vendor would have shipped a brittle OCR pipeline; a pure staffing firm would have shipped hours.
It also mattered that the team was senior from day one. Fine-tuning LLMs on degraded handwritten legal documents, parsing metes-and-bounds geometry, and wiring it all into GIS and satellite imagery is not junior work. OmniMinds’ Top 1% Expert-Vetted engineers — with a 100% Job Success record across 34+ projects — meant Glacier got architects, not resumes.
Brian Hill’s verdict, after 4 months and a transformed pipeline: results, delivered on time. That is the standard every OmniMinds engagement is priced against.
“OmniMinds has a broad range of experience and skills… They delivered great results in a timely manner. I highly recommend them.”