
Every claim checked before a handler opens it.
Incoming claims checked for missing documents and routed to the right team. Files the agent is unsure of wait for a claims handler.
The same case on the Gatehouse siteGoverned AI agents
We study your queue, deploy AI agents under rules you sign, and run them with you. Anything risky stops for your approver, and every action goes on a record you keep.
Pick the one that looks like yours. Each shows the jobs, what the agent does in each, and the person it stops for.
Agents work every invoice, reconciliation and vendor change in your AP queue. They stop for the person your policy names before anything posts.
Match, code, propose release within tolerance
Your controller: any variance outside tolerance, any duplicate, any missing PO
Tie out, list breaks, draft the reconciliation
The accountant who signs the reconciliation. The agent posts no adjusting entry
Log the request, gather the detail on file, draft the change
The AP lead, on every bank detail change, with a call-back to the number on file
Agents work your prior auth, denial, referral and intake queues. They stop for a person you name before anything reaches a payer or a patient.
Assemble the request, attach documentation, track status
The auth coordinator before anything is submitted; a clinician on anything clinical
Sort denials, gather documents, draft the appeal
The billing lead before any appeal or write-off leaves; the agent adjusts no balance
read the referral, check coverage, propose a slot, request missing information
The front-desk lead on anything the intake rules do not cover; a clinician on anything clinical
Agents for the shared inbox, the orders, the bills and the follow-ups that pile up with nobody to hand them to. You still sign the calls that matter.
Classify, draft from the approved answer set, open tickets
The owner or operations lead, on any refund, complaint or price
Flag exceptions, draft customer updates, propose a replacement within the limit
The owner, on any refund or replacement over the limit in the manifest
Capture, code, match, propose payment
The owner, on every payment release and every new vendor
AI agents that clear your security review, change control and internal audit before they touch a queue. Signed rules. An approver from your own team. Your cloud if you want it. A record your auditors can replay.
Exactly what the manifest lists, and nothing else
The approver the business unit and security agree on, for every action outside the list
Match, code, propose release within tolerance, run onboarding checks
The controller of the entity, on any variance outside tolerance; the AP lead on every bank detail change
Answer from the approved policy set with the clause cited, open tickets, route by name
The policy owner on anything outside the set; IT on every access change
Read this once: no customer names appear on this site. The cases are composites. The screens are simulated. Each one is labelled.
The people who deploy your agents are the people who run them.
We measure one process before anything is built: how much comes in, how much goes wrong. You keep the write-up, whatever it says. Sometimes it says: not yet.
Your team and ours write the rules. You sign them. The agent works inside them and nowhere else, in your cloud or ours.
We run what we deploy, day after day. Hard calls go to your approver. The record stays yours.
Every deployment runs on Gatehouse, the control software we built. Your agents act. You stay in charge.
What the agent may read, touch and spend. One signed file.
Every action, before it runs. Not named means refused.
Anything over the line waits for a person you name, evidence attached.
Every run, stop and decision. It exports, and it reads without Gatehouse.
Hire someone. Give your team ChatGPT. Build it yourself. Or hand it to us. Pick a question and read across.
A hire, or overtime on the team you have.
No
They ask someone
Notes, if any were kept
What they knew leaves with them
Volume, and holidays
A chat window, and someone pasting into it.
No. A person has to paste
It answers anyway
No record you can hand over
Nothing to keep
Consistent answers, and proof
Your own engineers, AI models and on-call rota.
Yes
Only if you built a hold
Logs, which seldom count as evidence
You keep it all, and the upkeep
Upkeep, once the build is done
Agents we build and run for you, under rules you sign.
Yes
It stops and waits for your approver
A record with the evidence and who approved
You take the record, in open formats
A process nobody wrote down
5 of the questions. The full table has more, and when to pick something else.
The full comparisonClaims, invoices and policy questions. Each one runs on Gatehouse today.

Incoming claims checked for missing documents and routed to the right team. Files the agent is unsure of wait for a claims handler.
The same case on the Gatehouse site
Every supplier invoice matched to its PO and receipt. Clean ones drafted in the ERP. The rest go to the AP lead, sources attached.
The same case on the Gatehouse site
Internal policy questions answered from your policy library, clause cited. Anything the agent is unsure of goes to a support lead.
The same case on the Gatehouse siteWritten by the people who run them. The newest three.
An AI agent for prior authorization assembles the request from the record, works the payer portals and status calls, and drafts the appeal. It does not decide what is medically necessary, and it does not submit on its own. What it should do, where it stops, and what to ask a vendor.
7 min readAccounts payable is the most common first AI agent and the one where build or buy matters most. The model is the easy part. Exceptions, bank-detail fraud and the audit file are what you are choosing between.
7 min readA PO invoice matching agent finds the purchase order and receipt behind each supplier invoice and checks price and quantity against your tolerances. What it should do, the invoices it must hold for a person, and what to ask before you buy one.
6 min readWe deploy AI agents on your work and run them for you. Every deployment runs on Gatehouse, the control software we built. You get signed rules, an approver for hard calls and a record of every run.
Gatehouse is Surehand's control software for AI agents. It checks each action against your signed rules before it runs. Hard calls go to your approver. Every run ends in a sealed record. Gatehouse on this site
You bring one process. We measure how much comes in and how much goes wrong. You keep the write-up, whether or not we go further.
It stops. Your approver gets it, with the evidence. Nothing moves until they decide. Their decision is saved under their name.
Anything over the limit stops and goes to your approver, with the evidence. The stop and the decision both go on the record.
Tell us what comes in, who handles it and where it breaks. A person replies, usually inside two business days.
support@surehand.io