The document problem in tower portfolios
A tower portfolio runs on documents. Ground leases, tenant and colocation agreements, amendments, renewals, easements, structural reports and site drawings accumulate over decades of builds and acquisitions. Most of them sit in folders as scanned PDFs, and what the site database says they contain is often a summary someone typed years ago.
Ask a routine question of that pile — which sites have a renewal falling due, where does the recorded rent disagree with the lease, which agreements still allow free assignment — and the honest answer at most operators is a project: pull the files, read them, hope the sample generalizes. The questions are ordinary. The reading is not.
The industry has barely started reading this material with machines. As reported in Inside Towers, only 2–3% of the industry’s documents have ever been read by AI. The databases built on top of those documents show the strain.
“We have seen across different portfolios the accuracy of these legacy databases is under 65 percent,” MYX CEO Yavor Mihailov told Inside Towers.
None of this is unique to towers — 85% of enterprise data is unstructured — but tower documents punish generic tools more than most. Terms hide in exhibits, amendments override the base lease, and the same clause reads differently on the ground-lease side and the tenant side. “The Towerco space is quite specific. There’s a lot of niche expertise and knowledge that is required to truly understand,” Mihailov told the publication.
Generic tooling struggles with exactly that specificity. An old ground lease amended several times, with an exhibit that redefines the rent, will not give up its current terms to a keyword search, and template OCR breaks the moment a layout changes. General-purpose chat AI is not the answer either. Mihailov’s framing in Inside Towers: “ChatGPT is an amazing tool, but it’s very generic.”
The practical consequence: rent rolls that disagree with the contracts behind them, renewal dates discovered late, and portfolio decisions made from samples because reading everything by hand was never an option. Every one of those problems traces back to the same root — nobody has read all of the documents since the day they were signed.
250,000 documents in under 48 hours
Reading everything is the point of MYX Analyzer. In one engagement, as reported in Inside Towers, the platform processed approximately 250,000 documents in under 48 hours.
Mihailov told Inside Towers: “You’re never going to read that with a human, but we read 100 percent of it with the Analyzer.”
The same engine is self-serve: process hundreds of documents overnight with enterprise batch processing. Batch runs show an upfront cost estimate before submission, so scale never means surprise spend.
Progress stays visible while a run executes. Real-time monitoring streams results field by field, driven by webhook updates rather than page refreshes, so the first extracted values appear long before the last document finishes.
During an acquisition, that scale changes what diligence means. The whole data room gets read instead of a sample, and the rent roll you underwrite is rebuilt from the contracts themselves rather than inherited from the seller’s database. See how teams use it for due diligence data-room extraction, and work the deal itself against the telecom tower due diligence checklist.
Watch: Lease reviews with the MYX AI Document Analyzer (1:24), from the MYX Demo channel on YouTube.
What bad lease data costs
The gap between what contracts say and what databases record has a price, and the Inside Towers article puts numbers on it.
Missed tenant revenue per site per year, as reported in Inside Towers.
Across a 10,000-site portfolio, that could translate into roughly $2 million annually, per the same article.
The mechanisms are ordinary: an escalation that was never applied, an amendment that never reached billing, equipment on the tower that no agreement covers. Each one is small. Multiplied across thousands of sites and years of lease term, they compound.
Cleaning this up after the fact is a documented pattern, not a theory. One customer — described in Inside Towers only as “a leading global tower company” — put its existing records through the platform. “They liked the tool so much that they hired internal people to get trained on it,” MYX Chief Revenue Officer Dean Bezlov told the publication. “They are now working and backfilling and checking all of their data through the AI analyzer.”
The fix, in other words, is not better sampling. It is reading everything and checking the database against what the contracts actually say.
Escalation clauses are where small errors grow fastest. A fixed percentage misremembered by half a point, or an index clause applied from the wrong base year, quietly reshapes a site’s cash flow over a decade. Model a clause yourself with the free lease escalation calculator — then extract the real ones from your leases instead of retyping them.
Tower-specific fields: colocation, escalation, renewals, assignment
MYX Analyzer extracts 500+ data fields, and tower teams use the ones that decide site economics: rent and payment terms, escalation type and index, renewal options, termination rights, assignment language, and colocation terms on the tenant side.
Ground leases and tenant leases are read differently, because they answer different questions — what a site costs versus what it earns. As Mihailov put it in Inside Towers: “We have a specific workflow for ground leases. We have a specific workflow for tenant leases.”
Ground leases: what a site costs
A ground lease answers what a site costs to keep: the rent you pay, how and when it escalates, when the term ends, which renewal options exist, and what the landowner can do on a sale or an assignment. Those answers rarely sit in one clause. They accumulate across amendments, and the current position is the sum of the whole chain — which is what a ground-lease workflow has to read.
Tenant and colocation agreements: what a site earns
The tenant side decides revenue: colocation rent per agreement, escalation terms tenant by tenant, amendment history as carriers change equipment, and renewal and termination rights that shape how durable that revenue is. This is where unread terms become the missed tenant revenue Inside Towers describes.
A worked example
Here is the shape of the output — synthetic data, shown as an illustration, not customer data:
| Field | Extracted value | Confidence |
|---|---|---|
| Document | Lease_FR_0847.pdf | SOURCE |
| Monthly rent | EUR 1,250.00 | HIGH |
| Lease term | 12 years (144 months) | HIGH |
| Escalation | INSEE CPI + 1.5% | HIGH |
| Renewal | Automatic (5x 3yr) | MEDIUM |
| Assignment | Freely Assignable | HIGH |
Every extracted field carries a confidence indicator (HIGH, MEDIUM or LOW) and a document reference back to the page and section it came from, so a reviewer checks the source in seconds instead of rereading the lease. Review happens in a spreadsheet grid — rows are documents, columns are extracted fields — and any cell, row or column can be rerun.
The example is French for a reason: the Analyzer supports 40+ languages, and European tower portfolios rarely come in one.
Extraction is configurable per portfolio. Each field has its own extraction prompt, each table carries its own project context, and media resolution settings handle high-detail material such as drawings. For the hardest fields there is MYX MAX, a dedicated tier.
Leases are the start, not the whole job. The rest of the site file runs through the same grid, and tower teams use it for:
Each of those is the same mechanism — documents in, a field schema out, every value referenced back to its source. Legacy site files are the deepest version of the problem, and they have their own guide: telecom site audit software for legacy portfolios.
For the abstraction workflow itself — field setup, batch runs, review and export — see AI lease abstraction; for the tower-specific version, amendment chains and colocation riders included, see cell tower lease abstraction.
Works with your site-management stack
Extraction is half the job. The data has to land where sites are actually managed — your site-management system, your rent roll, your finance tools. A field value that lives only inside an analysis tool helps the analyst; it starts helping the portfolio when it reaches the systems that bill tenants, pay landlords and flag renewals. Running that loop continuously — renewals, escalations, audits — is telecom lease management, covered in its own guide.
Documents come in through SharePoint and bulk folder import. Data goes out as Excel or CSV export with custom templates, and Pro and Business plans include API access for automated pulls.
For heavier lifts, you are not left alone with an export button. As reported in Inside Towers: “During acquisitions or large portfolio reviews, MYX engineers often work directly with customers to configure integrations and workflows.”
The same article notes: “For organizations with strict data requirements, the platform can be deployed within the customer’s own environment and in compliance with standard enterprise security measures, including zero AI model training on customer data.”
The delivery model matches: “For enterprise clients, the company usually deploys its software together with forward-deployed engineers.” Inside the platform, team collaboration is built in — shared workspaces with permissions keep review in one place.
If your sites live in Sitetracker, there is a dedicated path for getting extracted lease data in — see the MYX + Sitetracker integration.
“The value is immediate.” Dean Bezlov, MYX Chief Revenue Officer, in Inside Towers
FAQ
What tower documents can MYX Analyzer read?
Ground leases, tenant and colocation agreements, amendments, due diligence reports, technical documents and drawings, and legacy site files. Supported file types are PDF, JPG and MP4 — documents, images, and video — in 40+ languages.
How long does a large tower portfolio take to process?
Batch processing is built for portfolio scale: process hundreds of documents overnight with enterprise batch processing. In one engagement, as reported in Inside Towers, approximately 250,000 documents were processed in under 48 hours. Every batch run shows an upfront cost estimate before submission.
How do we verify what the Analyzer extracts?
Every field carries a confidence indicator (HIGH, MEDIUM or LOW) and a document reference to the page and section it came from. Review happens in a spreadsheet grid where rows are documents and columns are fields, and any cell, row or column can be rerun. As MYX Chief Revenue Officer Dean Bezlov put it in Inside Towers: “All of this can be done at scale by the AI, and then you just need to sign off as an expert.”
How does extracted lease data get into our site-management system?
Data exports to Excel or CSV with custom templates, and Pro and Business plans include API access. As reported in Inside Towers: “During acquisitions or large portfolio reviews, MYX engineers often work directly with customers to configure integrations and workflows.” There is a dedicated integration page for Sitetracker users.
Read your whole tower portfolio
Start with your own site documents. Upload leases, amendments and drawings, and see structured fields with confidence indicators and source references in the 30-day free trial.
30-day free trial · Card required — first charge only when the trial ends · Cancel anytime