In development · Selecting design partners for the pilot
Every notice read correctly.
Every deadline with its authority.
NoticeRight identifies a tax notice, extracts the facts that matter, states the deadline and the rule it comes from, and drafts a response for your review.
You decide the position. You sign and send. The software never transmits anything to a taxing authority.
Built inside Ledger Right, a working accounting practice — from three years of notices we handled ourselves.
Identified
IRS CP2000
Proposed changes to your tax return · underreporter
Tax year
2023
Taxpayer ID
•••-••-4417
Proposed increase
$6,240
Notice date
14 Mar 2026
Authority: 30 days from the notice date, per the notice’s own terms. If the window closes without a response, a statutory Notice of Deficiency (CP3219A) may issue.
Draft is watermarked DRAFT until a named person approves it.
Representative output. Figures are illustrative.
The problem
The notice is not the hard part. Everything around it is.
Notice work is high-consequence and low-leverage. It arrives unpredictably, it rarely bills cleanly, and the knowledge needed to handle it well lives in the head of whoever happened to do it last time.
- The same notice types get researched from scratch, by a different person each time.
- A CP504 says “levy” and reads like the end of the road. It carries no CDP right. An LT11 does — and that distinction is a deadline with teeth.
- Deadlines get copied off a letter without the rule behind them, so nobody can tell which are jurisdictional and which are courtesy.
- Response letters get adapted from whichever old letter was easiest to find, not from the one that fits the position.
- Penalty relief gets claimed under the wrong provision — First-Time Abate does not reach accuracy penalties under §6662.
- State and local notices have no playbook at all. They take the longest and pay the worst.
How it works
One path, seven steps, no branches.
A notice goes in. A reviewed, signed-off response comes out. Each step has a rule it will not break — printed here because the rules are the point.
01
Upload
Drop in a PDF or a phone photo of the notice. Multi-page notices are read as a whole.
Rule: full SSNs are redacted at ingestion, before the first write and before any model call. Only the last four are kept.
02
Identify
The notice is matched against the playbook library — not just by its form number, but by the markers that separate it from the notices it is most often confused with.
Rule: below the playbook’s confidence threshold, or on any negative-marker hit, it routes to manual review. It does not guess.
03
Extract
Only the fields the matched playbook actually declares — tax year, periods, proposed amounts, the response address printed on the notice itself.
Rule: a missing required field blocks drafting and names the field. The response address is read from the notice, never substituted from a stored default.
04
Deadline
The date is computed or read per the playbook, and shown with the rule it comes from — including what happens if the window closes.
Rule: the deadline renders even when drafting fails. It is the highest-value output and it does not depend on the rest working.
05
Draft
You choose the response strategy, then a letter variant. Both are filled from the case data. A separate plain-language client letter is generated alongside it.
Rule: rates and thresholds come from dated reference tables. If a tax year is not in the table, the draft is blocked rather than extrapolated.
06
Review
Notice and draft side by side. Edit freely. Approval is an explicit action by a named person, and it is logged.
Rule: every generated document carries a DRAFT watermark until that approval is recorded.
07
Export
PDF and DOCX on your letterhead, ready for your signature and your envelope.
Rule: NoticeRight never transmits to a taxing authority. There is no integration that files, sends, or represents. A human signs and sends.
What’s different
Anyone can build the upload form. The library is the product.
The application around it is scaffolding by comparison. What takes years to build is a set of authored notice playbooks grounded in correspondence that was actually sent, and outcomes that actually came back.
Authored from real files, not scraped from the web
Every playbook is written from three to five real specimens of that notice, never from memory and never from a single example. Each one carries what the notice means to a preparer and — separately, in a different register — what it means to the client on a stressful phone call.
The response strategies are branches from cases that were actually worked, and each branch carries a “watch for” note. Generic advice is worthless here.
It knows what it can confuse
The failure that kills a product like this is a confident draft for the wrong notice. So every playbook is authored with negative markers first — the things that must not be present — and an explicit list of the notices it is confusable with.
CP2000 against CP2501. CP504 against LT11. Separating those pairs is the actual problem; general accuracy is the easy part.
Every record says when it was last checked
Each playbook carries its own provenance: status, sources, the date it was last verified, the date its review falls due, and any question that could not be confirmed. A record whose review date has passed cannot drive a draft.
In a compliance product, a rule that cannot say when it was last checked is a liability. The unknowns are visible rather than hidden.
State and local, modelled to be tractable
Pennsylvania routes thousands of municipalities through a handful of appointed Act 32 collectors. Ohio concentrates in RITA and CCA. Model the collector, and one record covers every municipality it serves.
The same applies to states: across all fifty, notices collapse into a small number of recurring shapes. Author the shape once, attach the jurisdiction record for the address and the appeal rights, and guided drafting works anywhere.
50 states × every form number = a career
Coverage
Starting where the volume and the consequences are.
The IRS escalation chains were authored first, and both close end to end — because a playbook that warns you what issues next is only trustworthy once that record exists.
Where this actually stands. The library is being authored now, and no record drives a client-facing draft until it has been verified against current authority. That gate is deliberate and there is no flag to bypass it. If you are evaluating NoticeRight, evaluate it as a product in development — because that is what it is.
Standing constraints
What it will never do.
These are not settings. They are constraints written into the product, and they are the reason a professional can put their name on what comes out of it.
-
It never transmits to a taxing authorityThere is no integration that files, e-sends, or represents. It drafts. You sign and you send.
-
Every document is watermarked DRAFTUntil an explicit, logged approval by a named person. There is no silent path to a finished letter.
-
Every deadline displays its authorityNo exceptions. A date without the rule behind it is not information a professional can act on.
-
Low confidence routes to review, never to a guessDeclining to classify is a good outcome. A confident wrong answer is the worst one available.
-
Full SSNs are redacted at ingestionBefore the first database write and before any model call. Only the last four are retained.
-
Drafting confers no representation rightsCircular 230 and AFSP limits apply to you and vary by credential. The software does not change them.
Who it’s for
Firms where notices are a recurring workflow, not a rare event.
Solo and small-firm preparers
Where a single CP2000 can absorb an afternoon, and there is nobody down the hall who has seen one recently.
Firms with real notice volume
Where the same notice types recur often enough that re-researching each one is a measurable, repeated cost.
Practices doing state and local work
Where local income-tax and state notices are the least documented, the slowest to handle, and the hardest to bill.
Design partners
A small number of firms, working with it before it is finished.
The library only gets good against real notices from more than one practice. That is the whole reason for a pilot, and it is an honest trade rather than a marketing offer.
What you get
- Free access throughout the pilot, and preferential pricing afterwards
- Direct influence over which notice types and jurisdictions get authored next
- Onboarding and support directly from Damien, not a support queue
- An honest read on whether this saves your firm time — including if the answer is no
What we ask
- One introductory call so we understand how notices actually move through your firm
- Candid feedback on the drafts — particularly where they are wrong
- Optionally, redacted specimens of notices you have handled, to widen the corpus
- Periodic check-ins over the first few months of the pilot
Questions
The trust questions, first.
Does the software make the tax decision?
No. It identifies the notice, pulls the facts, computes the deadline with its authority, and produces a draft. You decide the position, you edit the response, and you approve it. The product informs a professional; it does not act for one.
What happens when it gets the notice type wrong?
The design goal is that it declines rather than guesses. Every playbook carries negative markers and a confidence threshold; a hit on the former or a miss on the latter routes the notice to manual review with no draft produced.
That will not be perfect. It is why every draft is watermarked until a named person approves it, and why the notice and the draft sit side by side during review.
How is client data handled?
Full SSNs and EINs are redacted at ingestion — before the first database write and before any model call — and only the last four characters are retained. Client data is not used to train models.
Details are in the privacy policy. If you are a preparer, note that your own §7216 obligations apply to any disclosure you make to a service provider, and you should confirm your position with your own adviser.
Does this replace my tax software or practice-management system?
No. It does one job: turning a notice into a correctly identified case with a defensible deadline and a reviewable draft. It does not prepare returns, it does not keep your books, and it does not manage your engagements.
When will it be available, and what will it cost?
There is no launch date to give you yet, and quoting one before the library is verified would be exactly the kind of claim this product should not make. Pricing will be set after the pilot, with design partners getting preferential terms.
If you want to know when that changes, the form below is the way to hear about it.
Who is building this?
NoticeRight is a Ledger Right product, built by Damien Scott out of an active accounting practice. The playbooks are authored from notices the firm handled itself, along with the responses sent and the outcomes that came back — which is why the guidance in them is specific rather than generic.
Request access
Tell us how notices move through your firm.
Design-partner places are limited and prioritised by notice volume and jurisdiction mix. If the pilot is not a fit, you will go on the list for launch instead.
- No obligation and nothing to install
- Damien reads and replies to these personally
- Your details are used to contact you about NoticeRight, nothing else
- Never sold, never shared with third parties