Ixana · Patent operations

Patent program operations

Fifty-plus filings across six product lines, owned end to end — filing decisions made on the roadmap's own calendar, no statutory date ever computed in-house, and a drafting clock on which every date is a cited message.

Shipped internally

Role: Product manager — patent programme owner and Wi-R product line PM
Team: Founder/CTO on strategy, outside counsel on execution, 1 engineer, with the patent paralegal as domain owner
Stage: Live, running across three outside firms

Live demo: five surfaces over a synthetic portfolio
No statutory date computed anywhere in the system
Drafting clock: no date without a cited message

What gets taped out decides what is patentable.

Problem. Ixana files ~25 applications a year off five-month tapeout cycles — seventy-five people when this began. Roadmap and patent estate were one decision surface run as two, and a matter's state lived in somebody's inbox: a disclosure past its filing date was discovered, not decided.

What I did. I owned both sides: the patent estate end to end, and the six Wi-R programmes those filings protect. Underneath, a mail-triggered spine routed every document to one matter identity, and a nine-stage drafting clock read each matter's dates from mail, calendar and chat, refusing any it could not cite.

Result. Fifty-plus filings across six product lines. Nothing irreversible is automated: routing, reminders and a date read from a message, yes; a date, scope or approval decided by a machine, never. The clock's first signal, n stated: median T0 to first draft 39 → 20 days.

01Why this, and why now

What gets taped out decides what is patentable. What is already filed decides what we can publish or demo. A disclosure landing before its filing destroys protection nobody can recover — so the roadmap and the estate are one decision surface.

I sat on both sides: 50-plus filings owned end to end — what to protect, how broadly, when to file — while running the six Wi-R programmes they protect. Disclosure date and filing date come off the same roadmap; until this programme different people set them.

Options sizing, costed against the ~25 applications a year the company files. All four buy docketing, which counsel already does properly and is not the thing that was missing — so the cheapest option won on the same argument that would have won at any price: none of them links a filing to the tapeout that makes it possible.

Twenty-five applications a year is large for seventy-five people, landing on the few also taping out silicon every five months. Spending engineering hours to save legal ones is the wrong trade — so the guardrail is inventor time.

02The problem as people experienced it

I did not ask anyone how it felt. I took every docketed action over twelve months — 214 — and for each counted the distinct places a person had to open to answer one question: what is owed here, by whom, by when.

What they did	Where it broke	Evidence
Asked what was owed on a matter	The answer depended on which of us you asked	The three roles disagreed on 23 of 214 actions
Searched the mail spool	Firms correspond under an informal name, the docket carries the formal title	31 of 54 matters had two names in active use
Opened the tracker	Stage was current; the next action and its owner were not	19 of 54 had no owned next action at all
Waited on an inventor	Nobody had told them a date depended on them	Median 5 weeks from request to disclosure

214 docketed actions over twelve months, walked and timed separately for each role. Counsel answered fastest from the fewest places, because the firm’s docket is one system that is genuinely maintained — the whole cost sat on our side of the boundary, which is what made this ours to fix rather than something to ask them for.

The reframe came from the 23 disagreements. Three people read the same mail spool and built three different states, because a matter's state was never written down. What was missing was a place for the answer to live — later, a clock that read the spool so nobody had to.

03My role and approach

I was the interface: the founder/CTO wanted to know what was protected, engineering wanted a disclosure to cost little of the week, counsel wanted one authority.

Key decisions

Anchor filing decisions to the tapeout, not the write-up. Every cycle is a disclosure risk and a priority date, so I decided on the roadmap's calendar.
Breadth is a roadmap question, not a legal one. Counsel could say what was claimable; only the roadmap knew what we would still build.
Automate only where being wrong is recoverable. Routing, reminders and a date read from a message can be fixed the same afternoon. A scope a machine decided cannot.
One matter identity, owned by us. Three firms, three naming habits; until resolved to one identity, nothing counted.
A date is a cited message, or it is not a date. The clock extracts from positive evidence, chooses by rule, and drops any reason it cannot cite.

Filings I owned, by product line and half-year (n=52). Markers are each line’s first public disclosure, taken from the product pages themselves. Filings cluster in the half before a line goes public — clearest on the two NFE parts, where the run-up is unmistakable. YR23 is the exception for a mundane reason: it shipped three months into this window, so most of its protection was already filed before the chart starts. That is the calendar working as the strategy, and I could only run it that way because I was in both rooms. The three source workbooks the portfolio dashboard is built from hold the whole estate, not just this window: 248 rows, 207 distinct patents once continuations and re-filings are grouped into families. The 52 here are the filings I owned in this window.

From the quarterly coverage review — one page, per product line

Every line gets one of four calls, agreed with the founder/CTO in the same hour

File. The line is still moving and someone else could plausibly get there first. Four of the six lines carried at least one of these in a typical quarter.
Defer. Claimable, but the roadmap has not committed. Deferred items are re-read next quarter rather than dropped — several carried twice before we filed.
Publish as a defensive note. Worth putting in the art so nobody else can claim it, not worth a filing. This is the call that saves the most money and gets made the least.
Abandon. The line moved and the idea no longer protects anything we are building. Proposing these was my job; nobody else in the room was incentivised to.

One page from the quarterly review. The per-line detail and every matter reference have been removed.

04What I cut

Shipped in v1

Mail-triggered extraction into matter and category
One internal matter identity across three firms
Lifecycle view over six stages, plus an explicit unclassified stage
Nine-stage drafting clock: dates read GET-only from mail, calendar and Teams, measured against the T0 targets
Stage cards — current status and reasons for delay, each citing the messages it rests on
Provenance and review queues for every machine choice
Internal readiness dates, each carrying its provenance
Invoice lines matched against the quoted schedule
Invoice approval / upload invoices — approval waits on a person opening the PDF, and upload covers what the mail spine missed
Weekly reconciliation against counsel’s docket
Deduplicated portfolio counting across three source workbooks
A combined master view derived from its sources, never authored
An annotation column preserved untouched through every rebuild

Deferred

Inventor-facing disclosure portal
Family-level cost roll-up
Category coverage measured against the roadmap

Cut

Machine-drafted claim language
Automated freedom-to-operate scoring
Any statutory date computed by us
Write-back to the source workbooks
A model choosing a date, or a reason for delay it cannot cite

Machine-drafted claim language was the hardest cut. It fails two stated constraints: anything a model proposes becomes part of the record of what the applicant considered, and a wrong claim scope is unrecoverable once filed. But I stopped it because drafting was never the bottleneck — attorney turnaround ran in days, inventor input in weeks. Automating the fast half of a slow process is the most expensive way to find the queue.

What handled the gap: structured disclosure capture front-loading the fields counsel asks for anyway, plus a prior-art shortlist that is cited and never written. The honest cost is that it saves no legal spend — only the schedule moves.

From the programme note — what we will not automate, v1.1

Three categories stay with a person, permanently, and not as a first-version simplification

Anything statutory. We compute no bar date, no priority date and no response deadline. Counsel’s docket is the record; we mirror it and mark it as theirs.
Anything unrecoverable. Claim language and filing scope. A wrong scope survives to grant and cannot be argued away afterwards.
Anything that becomes evidence. Machine-generated assessments of third-party rights. What the applicant considered is discoverable, and a generated opinion is worse than no opinion.
Anything that silently corrects a record. The reconciliation found 32 rows with real errors and the service still does not fix one of them. A tool that quietly amends a legal record is a tool nobody can testify about; it reports discrepancies and a person applies them, which costs twenty minutes a month and keeps every change attributable.
Anything a model asserts without a citation. A reason for delay that cites no message is dropped before it is shown, and a match the model suggests waits for a person. The model writes prose from the record; it never writes the record.

One page from the programme note. The examples naming live matters have been removed.

05How I got it agreed

The objection came from counsel's docketing partner, and it was the best one raised. Their docket is the legal record; a system deriving dates from email is a second source of truth the firm cannot indemnify, and a founder acting on our date is harmed by something the firm still answers for.

I could not win that, so I removed what she objected to: the system computes no statutory dates, mirrors the firm's docket as sole authority, and calculates only internal readiness dates.

She agreed, then asked for three things: a provenance stamp on every date, internal dates drawn so they cannot be mistaken for statutory ones, and no reminder without the firm's own date beside ours. The clock inherited all three.

06What was built

A mail-triggered spine, a workbook sync beneath it, six surfaces over both. Documents from any of the three firms are matched to one matter identity, given a stage and a category, and written into one record. The newest surface is the drafting clock: nine stages per matter, each date a message the reader can open, measured against targets fixed in advance.

The drafting clock grid: synthetic matters as rows, nine stage columns from drafting starts to filed, each cell carrying a date and beneath it the days taken against the target in green, amber or red; several rows carry an In progress pill where a review round or a first draft is still open, with the elapsed days running beneath it; the header chips read 108 matters on the roster, 0 dates without a cited source and 6 stages past target today.

1A date is a cited message, or the cell stays empty. Every date is read from the mailbox, the calendars or a group chat, and the reader can open the message behind it. Twelve roster matters show nothing because nothing in the record says a stage happened — a dash, not an estimate.
2Targets were fixed before the first matter was read. T0 is the draft reaching the firm; the offsets are the plan's own. The colour is how far over the target, never whose fault — the fault lives on the card, with its citations.
3In progress is a definition, not a guess. The stage after the last completed one, until filed. An open review round is not complete, so its cell carries no date and the elapsed days keep running against the target.
4Every machine choice has a human undo. A winner was chosen by rule; on the provenance view a person can reject it, select another candidate or set a date by hand. Each is a decision with a name and a note, and the grid recomputes.

A working recreation of the drafting clock, not a mock-up: the nine-stage model, the T0 offsets, the selection order, the minimum confidence per stage and the citation rule are the real ones. Every matter, reference name, firm, person, subject line and message is synthetic. The deployed portfolio demo linked above carries the other five surfaces.Open the clock, then click a red cell ↗

The spend view of the spend, reasons and routes surface on the synthetic portfolio: a total-invoiced-to-date figure of $61,900 across 70 invoice lines and 25 invoices, five tiles beside it for matters with spend, law-firm fees against government fees, spend never quoted, actual against quoted, and most expensive; a period switch, firm, route and colour selectors, a view switch reading Spend, Variance vs quote, Table, Timeline, Gantt, Changes, Routes, Negotiation, Firms, Exceptions; a legend of six fee categories with their totals; and beneath them the matters as horizontal bars ranked by total, stacked by category, with the amount and its variance against quote at the end of each bar, a PCT chip on the PCT-route matters, a green dot on the granted ones and a red triangle on the three billed above quote. — **Spend, reasons and routes: spend by matter**One ranked bar per matter, stacked by stage of work and never lumped per matter, so an invoice line is checkable against the engagement; under each total sits the matter’s variance against the firm’s quoted price, and ▲ marks a line billed above it. Clicking a bar opens every invoice line with the reason it was charged. The same surface carries nine more views — the ledger itself, variance against quote by fee stage, spend by month, one lane per matter, what is sitting unanswered, the US-only and PCT routes with a forward national-phase budget, a negotiation sheet with editable targets, a like-for-like firm comparison, and the exceptions and open questions. Every date on it is ours or the firm’s docket entry mirrored as such — none is computed.Open spend, reasons and routes ↗

The stage card for the firm's first draft on a synthetic matter, open beside the grid: a facts list with the date, the previous stage, a 10-day target, 36 days taken and 26 over, the owner, the source and the rule the date came from; a current-status line; two reasons for delay tagged Firm and Open decision, each followed by citation chips M6, M7, M8 and M5; and the cited messages listed beneath with date, source, side, sender and subject. — **Stage card: reasons for delay, cited or dropped**One cell, opened. The facts arrive at once — dates, target, days over, the rule and message the date came from. The status and the reasons for delay are written by a model from the messages in the stage window, each reason tagged Firm, Internal or Open decision and citing the message ids it rests on; a reason it cannot cite is dropped before anyone sees it, and the card says so when that leaves it empty.Open this stage card ↗

**Portfolio dashboard**The founder’s view — composition by stage and by technology category, and the next actions owed. Unclassified appears in the table in its own right rather than being absorbed into the total, so a gap cannot hide inside a rounded number.Open the dashboard ↗

**Lifecycle tracker**Every matter in one of six stages, with an explicit unclassified stage beside them. Overdue is a property of the internal action date and never of a statutory one, so nothing red here can be read as a legal deadline.Open the lifecycle tracker ↗

**Invoice approval**Each line checked against what that firm quoted for that stage. The amber block is the one that matters: the system made the match itself, says so, and will not let anyone approve until a person has opened the PDF and confirmed it. The upload half of the same section runs the same extraction for the few invoices the mail spine missed.Open invoice approval ↗

**Upload invoices**The other half of the same section, for the few invoices the mail spine missed. Firm and matter are chosen by hand because they are what ties an invoice to the portfolio; every cost line on the PDF is then read and matched against the quoted schedule exactly as a mailed invoice would be.Open invoice upload ↗

The other surfaces over the same spine, each link opening the screen its image shows — five on the deployed demo's synthetic portfolio, and the stage card from the drafting-clock recreation.

Surface	Who uses it	The decision it carries
Mail-triggered spine	Nobody — it runs underneath	What can be automated at all: routing yes, dates never
Workbook sync	Nobody — it runs underneath	What one patent means: a family counts once, and the master is derived, never authored
Drafting clock	Me weekly; each firm, read-only	Which stage is past its target, by how much, and — from the record — why
Provenance and review queues	Me, when a rule stops	Whether a machine's match or date stands, decided by a person with a note
Spend, reasons and routes	Me at quarter end; each firm, read-only	Whether a matter is worth what it is costing — by stage, against the quote, and for what the rest of its route will cost
Lifecycle tracker	Me weekly; the founder before a review	Which matters are past their internal action date
Portfolio dashboard	Founder/CTO, and counsel on the monthly call	Where the estate is thin against the roadmap
Ask the portfolio	Engineering leads, ad hoc	Whether a question needs me in the room at all
Invoice approval	Me, then the founder	Whether the work billed matches the engagement
Invoice upload	Me, for anything the mail spine missed	Which matter an invoice belongs to, decided by a person

System configuration — mail-triggered extraction into 6 lifecycle stages and 5 technology categories, plus an explicit unclassified stage · one internal matter identity mapped across 3 outside firms · portfolio counts deduplicated by family across 3 source workbooks, taking each family’s most recent row · eTag then content-hash change detection on a 60-second poll · master view rebuilt in place through the workbook API with annotations preserved and every write verified · spend held as an invoice-line ledger — 1 row per line, each carrying its fee type, who keeps the money, the jurisdiction, the firm’s quoted price and a written reason · variance reported on law-firm fees only, government fees passed through · national-phase deadlines mirrored from the firms’ dockets, never computed · forward budget from the firms’ country quotations, labelled estimate · invoice lines matched individually against the quoted schedule · deadline colouring at 60 days, 6 months and 1 year · no statutory date computed anywhere in the system · nine-stage drafting clock with T0 offsets −10, −5, 0, +5, +15, +20, +21, +22 and +24 days, read GET-only from one mailbox, the calendars and the group chats · alias match at an 85-point threshold · winners chosen by rule in chronology order with a minimum confidence per stage · stage cards cite message ids or drop the reason · 0 writes to any source

07Tradeoffs

Tension	What I chose	What it cost
Cycle time vs inventor time	Front-load the fields counsel asks for anyway; never chase for more	Some disclosures reach counsel thinner than they could be and come back with questions
Coverage vs discoverability	No automated freedom-to-operate scoring, on the same reasoning that cut drafting	We have no systematic read on third-party risk; it stays a question we ask counsel, one product at a time
One view vs one authority	Mirror counsel’s docket; compute only internal readiness dates	Two date systems on one screen, and a permanent explaining job for anyone new
Automation vs attributability	Machine-made links stay marked until a person confirms them	Every invoice needs a human minute before approval, and a backlog forms whenever I travel
Correctness vs a record that can be testified about	Report discrepancies, never correct a source workbook	Known errors sit uncorrected until a person applies them, sometimes for weeks
A derived view vs someone’s working surface	Derived columns rebuild every cycle; the annotation column is preserved untouched	The sync halts whenever a rebuild would drop an annotated row — twice so far, both justified
Evidence-only dates vs coverage	A stage is dated only by a message that says it happened	Stages that live in chat are found for barely a third of matters; the grid shows the gap rather than an estimate

08Impact and outcomes

We agreed before build that this failed if any date the system showed disagreed with counsel's docket at the weekly reconciliation twice running. It has not happened. The guardrail was inventor minutes per disclosure: cycle time bought with engineering hours is not a win. The clock adds one early signal.

14w → 6w

Median disclosure to filing

39d → 20d n=10

Median T0 to the firm's first draft, against a 15-day target (drafting clock)

19 → 2 of 54

Active matters with no owned next action

95m → 40m

Inventor minutes per disclosure (guardrail)

Dates disagreeing with counsel's docket · source sheets modified

0 of 284

Dated drafting stages without a cited source message

Share of open matters still waiting, by how long the next action had been owed. Before is the starting population (n=38); after is the same measure over the latest twelve months (n=54). The curves separate almost entirely inside the first month and then fall at a similar rate, which says plainly that the system fixed how fast work gets picked up and did nothing at all for the long tail — and that tail is inventor availability, not process.

How we counted. An active matter is a filing with at least one open action owned by a named person and dated inside ninety days. It excludes granted patents with no maintenance action due, abandoned matters, and unconverted provisionals past their window. We count the matter, not the document — a family of four counts once.

09What I'd do differently

I built the mechanism before I wrote down the strategy. For a year the filing decisions lived in my head and in conversations, while what got documented was the plumbing. How a tapeout maps to a filing decision is still the part I cannot hand over.

And I would have started the clock a year earlier. The overrun it shows most — the firm's first draft — was the one everyone felt and nobody measured; the tail past sixty days is inventor availability in a tapeout crunch.