In development
Economic intelligence
for corporate tax.
The inversio Engine reads annual securities filings, extracts the tax and business data buried inside them, and turns it into analysis a professional can stake a decision on. Not a summary. Not a search index. Every figure it publishes resolves to the filing text that supports it, or it is not published.
The product
inversio Engine
Corporate tax disclosure is written to satisfy a filing requirement, not to answer a question. The material facts are real and public — they are simply buried in narrative, spread across years, and phrased so that two filings describing the same position rarely describe it the same way. The Engine is built to read that disclosure at the standard a tax professional would apply to it, and to show its work every time.
-
Evidence-anchored extraction
Reads annual filings — Form 10-K as the core corpus, with controlled and explicitly guarded Form 20-F support — and captures each tax position together with the exact source text behind it.
-
Governed analysis
Positions are evaluated against approved formulas and expert-authored tax mechanics held in a versioned registry. Analysis is a governed artifact, not a heuristic that drifts between releases.
-
Auditable provenance
Every published value resolves through a provenance ledger to a source anchor or a recorded derivation path. Findings carry an explicit evidence basis, so a reviewer can see how strong the support actually is.
-
Deterministic by construction
No language model runs at inference time. The same filing produces the same answer on every run — a property you can test, and one that matters when the output has to be defended.
-
Fail-closed
Where the evidence is missing or unstable, the Engine returns nothing and says so — rather than filling the gap with an estimate and presenting it in the same typeface as a fact.
-
Built for review
Issuer library, analytics, peer comparison, and generated reports — each surface built so a finding can be traced from headline to underlying filing language without leaving the product.
Proof
The filings are public. The information inside them is not.
Not in a form you can compare across companies, or across years, or trace back to the sentence it came from. That layer did not exist at this scale. The inversio Engine builds it — and once it exists, professional judgment has something to stand on.
Proof over probability.
A language model estimates. The inversio Engine establishes what the filing supports — the potential strategy identified, assessed under the law applicable to that filing year, classified so it compares across companies and years, and anchored to the sentence it came from.
An estimate is something you have to defend.
Proof defends itself.
The filing says
“A valuation allowance is recorded when it is more likely than not that some portion or all of the deferred tax assets will not be realized. The Company considered all available positive and negative evidence, including cumulative losses in recent years, and determined that a full valuation allowance against its U.S. federal deferred tax assets remained appropriate as of the balance sheet date.”
Composite language, not a specific filer.
What the Engine records
- Strategy
- Valuation allowance — full, U.S. federal Potential; identified from disclosure
- Evidence
- Source-anchored
- Quote
- Retained verbatim Located to the exact position in the source document
- Authority
- Assessed under the tax law and accounting standards applicable to that filing year
- Boundary
- Concludes about the filing, not about you What to do with it is your decision.
Simplified for illustration. A live finding carries considerably more — for example, how the sector treats the same position, how this filer’s treatment has moved against prior years, and the analysis that follows from both.
Comparable by construction
Two filings describing the same position rarely describe it the same way. A model returns each in its own words, and words in different forms cannot be set side by side. Every finding is classified identically for every company — which makes peer benchmarking and year-over-year change possible at all, rather than merely difficult.
The law that applied, not the law today
A position disclosed in 2019 is governed by different authority than the same position in 2025. A model applies whatever it absorbed, whenever it absorbed it. Each filing year is assessed under the law applicable to that year.
The same answer in eight months
Ask a model the same question twice and you may get two answers. Change its version and you certainly will. Nothing here is generated. The same filing returns the same finding indefinitely — and that is testable, not promised.
Nothing leaves
No filing text, no client context and no prompt is sent to an outside model at inference. For a firm handling privileged matters, that is a precondition rather than a preference.
Audience
Know which potential tax strategies a company may be pursuing — and what its filings say about each one.
The inversio Engine reads years of annual filings and surfaces the potential tax strategies and sub-strategies a company may be pursuing, each one attached to the disclosure language that supports it. The judgment stays yours — but you start from evidence, not a guess.
-
Corporate tax departments
Benchmark your positions against comparable filers: which potential strategies peers are pursuing, how they disclosed them, and where your own treatment sits against the field — with the source text behind every comparison.
-
Transaction diligence
Map a target’s tax profile before the bid: which potential strategies it depends on, where exposure concentrates, and what the filings actually say — across its full filing history, without reading each one by hand.
-
Advisers and analysts
Answer “what are others doing here?” from evidence rather than recollection. Firms covering more issuers than any team can review manually, where every claim has to survive a partner’s read or a client’s challenge.
Across all three, the starting point is the same: filings most people do not have time to read line by line, turned into potential strategies with their evidence attached. What you conclude — and the professional judgment behind it — stays yours.
Approach
Three commitments we do not trade away for coverage.
A tax platform can always look more capable by loosening what it requires before it will speak. We have taken the opposite position, deliberately, and we hold it under engineering gates rather than good intentions.
-
I
Determinism over inference
A number a professional relies on has to be reproducible. Ours is computed, not generated, and the same inputs return the same result indefinitely.
-
II
Provenance over assertion
Confidence is not a substitute for evidence. Each claim carries the route back to its source, and the strength of that route is stated plainly rather than smoothed over.
-
III
Silence over a plausible guess
The costly failure in this field is not an absent answer — it is a confident wrong one. When the filing does not support a finding, we would rather show nothing.
Founders
Who is building this.
Tax practitioners and scholars who have been working on these questions together since 2019. The domain expertise predates the company.
-
Julianne Jones
Co-Founder & CEO
Director of the University of Akron's School of Accountancy and Professor of Practice in accounting. Previously a tax manager in practice; she is a CPA with a JD and a Master of Taxation, and conducts research on federal and state taxation.
-
Doron Narotzki
Co-Founder & CTO
Full Professor of Tax and Business Law at the University of Akron's School of Accountancy, where he directs the Master of Taxation program. He has published more than seventy articles on corporate and international taxation and practiced international tax at PwC. He originated the approach behind the inversio Engine.
Contact
inversio Engine is not yet publicly available.
We are building toward a limited early-access release. If you work in corporate tax, transaction diligence, or financial data — or you would like to be told when access opens — write to us directly. Every message is read by a founder.