Production AI, built and run for you
Read every call. Answer every review. Appeal every denial.
SASID puts production AI on the customer conversations and claims you already have, then runs it with you. Live in 4 to 8 weeks. One fixed price. Working proof on your own data before you pay a dollar. Pick the leak that is yours and the page rebuilds itself around it.
Free 30-minute call. Written roadmap in 48 hours. If the numbers say do not build, we will say so.
- 7 systems live across 5 industries
- Live in 4 to 8 weeks
- You own all the code
Which leak is yours? The page rebuilds around it.
The problem
The calls, reviews, and denials you already have are the revenue you are not collecting
A QA team hears 2 to 5 percent of calls. Managers answer a third of reviews. Revenue cycle teams appeal the denials that are big enough to be worth 45 minutes of clinical time and write off the rest. None of that is a technology gap. It is arithmetic: there are more conversations and claims than there are people to read them, so the work gets sampled, and the revenue in the unsampled part is written off as if it were inevitable.
Which leak is yours?
Choose your operation above the fold and this page rebuilds itself: the problem in your numbers, the workflow that fixes it, a calculator with your defaults, and the proof from the closest system already in production.
How it works
The system drafts. Your reviewer signs.
Cooperating agents do the reading and writing. Your nurse or denials specialist approves every letter, and the edits feed back into accuracy. Nobody is replaced, and nobody spends an hour typing.
- 01
Denial arrives
The denial and the linked encounter are pulled from your EHR over read-only FHIR, or from your 835 remits and a chart export where FHIR is not yours to grant (Epic Community Connect hosts, athenahealth, eClinicalWorks). Nothing leaves your environment except the model call, under BAA, through a private endpoint.
- 02
Record and policy are read
One agent builds the clinical argument from the chart. Another matches the payer's own policy language and the medical-necessity criteria that apply.
- 03
The letter is drafted, with citations
A complete, payer-formatted appeal with inline citations to the record and the policy, ready in about two minutes instead of 30 to 60.
- 04
Your reviewer signs, the appeal is filed
Your nurse or denials specialist reviews and approves. Edits are tracked, accuracy is measured against them, and the appeal goes out inside the filing window.
The same engine runs three more workflows for provider groups
Patient-access call coverage
Every recorded call scored for compliance and empathy, with evidence quotes, instead of a 5 percent sample.
Plain-English reporting over denial data
Ask "which payer denied the most inpatient claims last quarter" and get a cited answer from your own data.
Patient review response
Every public review answered in your voice within minutes, per location, with staff approval where it matters.
How it works
The system drafts, scores, or answers. Your staff approve what matters.
Each workflow below is shipped and measured in production. Start with the one that matches your leak, and add the next when the first is paying for itself.
Denial appeals
Insurance denials turned into complete, cited appeal letters from the medical record, reviewed by your staff and filed inside the timely-filing window. At 2 minutes an appeal you fight every denial, not just the large ones. Deployed inside your HIPAA environment under a BAA.
Call quality scoring
Every recorded call scored against your scorecard within the hour, with the evidence quote and a coaching note per agent. Scores are calibrated against your senior QA analysts and the agreement rate is reported monthly.
Review response engine
Every Google, Yelp, and Facebook review answered in your brand voice within minutes, per location, with a manager approval step for anything under three stars. Replies reference the actual visit from your job record.
Lead scoring from signals
Buying intent detected in reviews, calls, and CRM activity, pushed to sales as a ranked queue.
Plain-English data queries
Sales and operations teams query complex systems in natural language with validated, correct results.
Onboarding copilot
A conversational guide over your product that gets new customers to value in days instead of weeks.
The cost of doing nothing
Put your own numbers in.
A few sliders, no email required. The estimate uses your inputs and the stated assumptions, nothing hidden, so you can check the math before you talk to anyone. Bring the numbers to the call and the roadmap does this math for your group, in writing.
Denial appeals
CFO or VP of revenue cycle at a hospital system or physician group
Median was 2.7% in 2025 (Kodiak Solutions, 2,300+ hospitals). Edit it if you know yours.
Your call. About one physician in five says their practice always appeals an adverse prior auth decision (AMA, 2025).
Used with net revenue to pick the tier that fits you.
- Overturn rate on appeals you file today:
- 70% Premier Inc., 2025
- Applied to denials you have never fought:
- 35%, half the rate, because the easy ones are already being appealed
Final denials written off every year
- $4,050,000Final denials written off per year
- $2,430,000Of which never appealed
- $425,250Recoverable at the stated assumptions (cannot exceed the write-off)
- $95,000Year-one cost of the tier that fits your volume (Appeals Engine)
The estimate starts from what you already wrote off, not from initial denials, and it applies half the industry overturn rate to denials nobody chose to fight. Bring your own final denial number to the call and the roadmap redoes this math for your group, in writing.
What you get
What you get, and what each piece is for
- 01
The appeals engine, live in 4 to 8 weeks
Record retrieval, clinical evidence, payer policy, and letter generation, with reviewer sign-off, deployed in your EHR environment. In production this took under 2 minutes per appeal against 30 to 60 by hand, and the hospital group recovered millions in denied inpatient claims.
What it kills: Denials expire unappealed because nobody has the hours.
- 02
Working proof on 50 of your own denials before the fee
We run fifty of your real denials, under NDA and BAA, and your reviewer scores the letters. You get a written go or no-go, and "do not build" is a real answer we give.
What it kills: Pilot purgatory and vendor slideware.
- 03
A 48-hour roadmap with your denial math
Architecture, integration path, timeline, price, and your own numbers: denial rate, overturn rate, unappealed volume, and expiring deadlines.
What it kills: "Is this worth it?" and "will IT allow it?"
- 04
Week-one data audit and a one-page readiness checklist
Written findings on record access, denial exports, and gaps, plus the list of what we need, from whom, for about one hour a week.
What it kills: "Our data is a mess" and hidden effort on your side.
- 05
Your reviewer signs every letter
The system drafts; your nurse or denials specialist approves. Edits feed back into accuracy. Staff move from an hour of typing to minutes of review.
What it kills: Clinical and compliance fear, and staff resistance.
- 06
Monthly results review
Coverage, accuracy against reviewer edits, cost per appeal, cycle time, and dollars in flight, on one page you can put in front of the board or the owners.
What it kills: "We cannot measure the impact."
- 07
Quarterly executive briefing and a diligence pack
What shipped, what it earned, what is next, plus the architecture, audit trail, evaluation results, and ownership documents a buyer or lender will ask for.
What it kills: "What is our AI strategy?" and quality-of-earnings scrutiny.
- 08
You own all of it
Code, documentation, runbooks, and training, from day one. If SASID disappears, the system does not.
What it kills: Bus factor and lock-in.
The 90-day guarantee
Before we build, we agree in writing on two numbers: the share of denials the engine drafts a letter for (coverage) and the share of drafted letters your reviewer accepts with under five minutes of edits (accuracy). If either number is below target at launch or at any point in the first 90 days, for any reason on our side, we fix it at no charge and the monthly fee pauses until it is met. If it is not met within 30 days of your written notice, you may end the term, owe nothing further, and keep all code, documentation, and runbooks. The implementation fee is earned at launch and is not refunded.
Your side: read access to the systems in the readiness checklist within 10 business days, one named champion for about an hour a week, and reviewer feedback on drafted letters within 5 business days. Full wording is in the agreement.
Onboardings are scheduled monthly and limited to two or three at a time so every launch gets the engineer's full attention. A signed agreement holds a date.
Run it on 50 of your denialsProof
Shipped, measured, in production
Seven AI systems the founder built and shipped across five industries, as the in-house AI lead and as a consultant. Five are documented below with the client type, the system of record, and the measured result; two are under NDA. No testimonials appear on this site until a client agrees to be named.
- 5 weeks
- kickoff to production for the appeals engine
- 2 min
- per appeal, was 30 to 60
- Millions
- in denied inpatient claims recovered
- 7
- production systems, 5 industries
Why I built this
The engineer on the first call is the engineer who ships your workflow.
I have been building software for thirteen years and shipping production AI for the last five. The pattern I kept seeing was the same in every industry: a pilot that impressed everyone in the demo and then quietly stopped moving. Nobody owned it. Nobody measured it. It never touched a real workflow.
The turning point was being hired as the only AI engineer at a company that needed a platform every product team would depend on. There was no team to hand it to and no room for a pilot. It had to run. It ended up handling more than 30,000 reviews a day across 200 locations, and lead scoring on those signals added seven figures of recurring revenue.
Then came healthcare. Nurses were spending 30 to 60 minutes writing each insurance appeal, and the backlog meant many denials were never appealed at all. The system we built drafts a cited, payer-formatted appeal in under two minutes, inside a HIPAA environment, with a nurse signing every letter.
I started SASID so a provider group does not have to build a twelve-person AI team to get this. I build it, I measure it, and I stay.
Shahrukh Siddiqui, founder and engineer, SASID AI
- 13+ years engineering
- 5+ years production AI
- Head of AI, platform to product
- HIPAA, SOC 2, ISO 27001 environments
If something happens to me
You own the source, the architecture docs, and the runbooks from week one. Your team is trained before launch. The platform runs in your cloud, on your accounts, and the runbooks are written so that any senior engineer can operate the monthly review from the documentation alone. No part of your workflow depends on a person you cannot replace.
Where the work was done
- Vertex Inc, 2024 to 2026. AI engineering across healthcare, cybersecurity, and customer-service platforms.
- Persefoni, 2022 to 2024. Senior engineer, climate accounting platform and its AI copilot.
- Bit Builders, 2017 to 2021. Lead engineer, banking platforms for national banks.
- Schlumberger, 2014 to 2017. Web developer, public sites and internal systems.
- B.S. Computer Science, University of Houston
Where the engineer has shipped before
Employers and consulting clients over a 13-year career. These are not SASID clients and no endorsement is implied. All trademarks belong to their respective owners.
Pricing
One implementation fee. One monthly fee. No hourly meters.
One clinical appeals nurse costs about $94,500 a year. One unappealed inpatient denial can exceed $14,000. This is priced against those numbers, not against hours.
Platform
Multi-facility, inside your compliance boundary.
from $400,000 over 12 months
Includes up to 150,000 denials a month across all facilities
Written quote in the 48-hour roadmap. No assessment fee.
- Every workflow across facilities and entities
- Dedicated environment in your cloud, BAA signed
- Fractional Head of AI seat
- Quarterly executive briefing and diligence pack
- Security review support for your IT and compliance teams
- Everything in Revenue Cycle
Revenue Cycle
RecommendedAppeals plus two more workflows across the revenue cycle.
$204,000 over 12 months
Includes up to 50,000 denials a month (calls or reviews on the adjacent workflows)
- The appeals engine
- Two of: patient-access call coverage, plain-English reporting over denial data, patient review response
- A plain-English analyst over your own denial data
- Biweekly roadmap sessions
- Priority engineering response
- Everything in Appeals Engine
Appeals Engine
Every appealable denial, appealed.
$95,000 over 12 months
Includes up to 10,000 denials a month
The smallest tier. A group under 3,000 denials a month gets the same engine at the same price, well inside the limit.
- The appeals workflow, Epic or Cerner over FHIR
- Reviewer sign-off on every letter, edits tracked
- Weekly releases into your environment
- Monthly results review with the engineer
- Coverage, accuracy, and cost per appeal reported
- 90-day performance guarantee with term exit
Included in every plan
- You own all source code, documentation, and runbooks from week one
- Working proof on your own denials before the implementation fee is committed
- Your nurse or denials specialist approves every letter, with an audit trail
- PHI stays in your cloud tenant; the model call goes through a private, HIPAA-eligible endpoint under BAA
- Cost per appeal, coverage, and accuracy reported to you monthly
- Model routing and fallback across providers, PII redaction, and guardrails
Every plan carries the 90-day guarantee with term exit: two agreed numbers, fee pause if missed, and the right to end the term after 30 days' notice.
Usage above the plan limit is passed through at model cost with the invoice attached. Your monthly review shows cost per appeal, so there are no surprises.
12 months standard, starting at launch. A 36-month partner term with a locked price is offered after the 90-day guarantee period, once you have a number. Contracts are assignable to an acquirer at no charge.
All prices in USD.
Other operations
The same engine, pointed at other leaks
SASID is built for provider groups first. The workflows below are shipped and measured for operators in adjacent industries, and each has its own page.
Questions revenue cycle leaders ask
An AI appeals system built and run inside your EHR environment by the senior engineer who built the original, plus the monthly operation of it. The implementation fee covers integration with Epic or Cerner over FHIR, the payer policy corpus, evaluation against your reviewers, and launch. The monthly fee covers hosting in your tenant, model costs within plan limits, monitoring, tuning, the monthly results review, and the engineering time in your plan.
Get started
Send us 50 of your denials. Get the math back in writing.
Thirty minutes on a call, then a written roadmap within 48 hours: the workflow, the systems it connects to, the price, and what doing nothing costs at your volume. Then fifty of your real denials, under NDA and BAA, drafted and scored by your reviewer before any fee.
- Free 30-minute call
- Written roadmap in 48 hours
- Proof on 50 of your denials before any fee
Book the 30-minute call
Thirty minutes. You bring the leak and the volume; within 48 hours you receive a written roadmap covering the workflow, the integrations, the price, and the cost of doing nothing at your numbers.
- 0130-minute call
You bring the leak and the volume. We map it to one workflow and the system it lives in.
- 02Written roadmap in 48 hours
The workflow, integrations, timeline, price, and the cost of doing nothing at your numbers. If the numbers do not support building, the roadmap says so.
- 03Proof on 50 of your denials
Fifty real denials (or two hundred calls, or a week of reviews for adjacent workflows). You see the letters and the accuracy before any fee is committed.
- 04Live in 4 to 8 weeks
Weekly releases into your environment, then the 90-day guarantee clock starts.
Prefer to write it down?
Starts are scheduled monthly. A signed roadmap holds a date.
Get the AI readiness checklist
The 12-point checklist used to qualify every SASID engagement, appeals included. Score your group in five minutes and find out whether the workflow will ship or stall, before you spend anything.
One PDF, no newsletter, no spam.