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
Agents that do the work. People who sign the hard calls.
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.
Where it runs
Your queues. Your systems. Your sign-off.
Pick the one that looks like yours. Each shows the jobs, what the agent does in each, and the person it stops for.
01 / 04 · Finance
Agents do the matching. Your controller signs.
Agents work every invoice, reconciliation and vendor change in your AP queue. They stop for the person your policy names before anything posts.
- 01
Invoice matching
The agentMatch, code, propose release within tolerance
Stops forYour controller: any variance outside tolerance, any duplicate, any missing PO
- 02
Close and reconciliations
The agentTie out, list breaks, draft the reconciliation
Stops forThe accountant who signs the reconciliation. The agent posts no adjusting entry
- 03
Vendor master changes
The agentLog the request, gather the detail on file, draft the change
Stops forThe AP lead, on every bank detail change, with a call-back to the number on file
02 / 04 · Healthcare
The queue moves. The clinician still decides.
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.
Works in
- Epic
- athenahealth
- Payer portals
- Fax inbox
- athenahealth billing
- Clearinghouse
- 835 remits
- 01
Prior authorisation
The agentAssemble the request, attach documentation, track status
Stops forThe auth coordinator before anything is submitted; a clinician on anything clinical
- 02
Claims and denials
The agentSort denials, gather documents, draft the appeal
Stops forThe billing lead before any appeal or write-off leaves; the agent adjusts no balance
- 03
Referrals and intake
The agentread the referral, check coverage, propose a slot, request missing information
Stops forThe front-desk lead on anything the intake rules do not cover; a clinician on anything clinical
03 / 04 · Small and mid-size businesses
The queue gets worked. You still sign.
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.
Works in
- Gmail or Outlook shared inbox
- HubSpot
- Help desk
- Shopify
- Carrier tracking
- Warehouse or 3PL portal
- QuickBooks or Xero
- 01
Shared inbox triage
The agentClassify, draft from the approved answer set, open tickets
Stops forThe owner or operations lead, on any refund, complaint or price
- 02
Order exceptions
The agentFlag exceptions, draft customer updates, propose a replacement within the limit
Stops forThe owner, on any refund or replacement over the limit in the manifest
- 03
Bills and invoices
The agentCapture, code, match, propose payment
Stops forThe owner, on every payment release and every new vendor
04 / 04 · Enterprise
Built to pass the review. Not to skip it.
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.
Works in
- SAP
- Oracle
- Coupa or Ariba
- Vendor master
- AP inbox
- ServiceNow or Jira Service Management
- Confluence or SharePoint
- 01
The pilot a business unit already runs
The agentExactly what the manifest lists, and nothing else
Stops forThe approver the business unit and security agree on, for every action outside the list
- 02
Shared-services AP and procurement
The agentMatch, code, propose release within tolerance, run onboarding checks
Stops forThe controller of the entity, on any variance outside tolerance; the AP lead on every bank detail change
- 03
Employee policy and help-desk answers
The agentAnswer from the approved policy set with the clause cited, open tickets, route by name
Stops forThe policy owner on anything outside the set; IT on every access change
Govern the agent you already run
Read this once: no customer names appear on this site. The cases are composites. The screens are simulated. Each one is labelled.
How an engagement works
Study it. Deploy it. Run it.
The people who deploy your agents are the people who run them.
/01 Study
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.
/02 Deploy
Your team and ours write the rules. You sign them. The agent works inside them and nowhere else, in your cloud or ours.
/03 Run
We run what we deploy, day after day. Hard calls go to your approver. The record stays yours.
Gatehouse · the product
How Gatehouse works
Everything between the agent and your systems.
Every deployment runs on Gatehouse, the control software we built. Your agents act. You stay in charge.
How Gatehouse works- 01
Rules you sign
What the agent may read, touch and spend. One signed file.
- 02
Checked before it acts
Every action, before it runs. Not named means refused.
- 03
Your approver on hard calls
Anything over the line waits for a person you name, evidence attached.
- 04
A record that shows any edit
Every run, stop and decision. It exports, and it reads without Gatehouse.
Four ways to clear the queue
Not all of them are us.
Hire someone. Give your team ChatGPT. Build it yourself. Or hand it to us. Pick a question and read across.
/01
Hire a person
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
/02
Give the team ChatGPT
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
/03
Build it yourself
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
/04
Surehand
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 comparisonDeployments
Three kinds of work we run.
Claims, invoices and policy questions. Each one runs on Gatehouse today.
The real mismatches, found. The small ones, cleared.
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
Every answer cites its clause. Or a person answers.
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 siteArticles
What we learn running agents.
Written by the people who run them. The newest three.
NIST's ten agent tenets: a guardrail is not a control, and a tired approver is not oversight
NIST proposed ten design tenets for agent identity. Four you can test against your setup this week, and two change how you run approvals.
4 min readRPA vs AI agents for accounts payable
In accounts payable, RPA handles the steps that never change and an agent handles the invoices that do not fit. Which part of the AP process each one should own, and the controls an agent needs before it touches a payment.
5 min readSK Telecom gave its AI agents employee numbers. Keep the approver off the scoreboard.
SK Telecom's first AI Governance Report gives every agent an ID, an owner and a stop button. Copy that. Then fix the part that could tie the supervisor's pay to the agent's output.
4 min read
What does Surehand do?
We 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.
What is Gatehouse?
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
What is a teardown?
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.
What happens when the agent is unsure?
It stops. Your approver gets it, with the evidence. Nothing moves until they decide. Their decision is saved under their name.
What happens at the spend limit?
Anything over the limit stops and goes to your approver, with the evidence. The stop and the decision both go on the record.
Start here
Bring us the queue nobody wants.
Tell us what comes in, who handles it and where it breaks. A person replies, usually inside two business days.
support@surehand.io