We design and build custom AI for your finance, operations, sales and service teams, inside the ERP, CRM and banks you already run. It takes on the work that grows with your business. Your people approve what matters, and you own everything we build.
Your growth shouldn't wait on manual work. More customers, more volume and new markets add work between your systems. You don't need a new ERP to grow. You need that work to keep up. Since 2021, we have built production AI that takes it on.
- Medical manufacturing
- 60%
- less manual entry across 4 markets
- Beverage manufacturing
- 15 min
- per invoice run, down from 3 days
- Payments
- 99.25%
- of 4.6M transactions a period matched automatically
- Private equity
- 3 weeks
- earlier warning of cash-flow risks across 20+ companies
Built into the systems you already run
SAP
Microsoft Dynamics 365
Salesforce
Oracle NetSuite
HubSpot
Zendesk
Sage
Xero
DATEV
SharePoint
Snowflake
Databricks
Azure
AWS
OpenAI
How It Works
A real case: a payments operator reconciling every card settlement against its ledger. Agents read, rules check the numbers, a person approves each exception.
Every rule is written down and traced to its source, not kept in someone's head.
- 01
Work arrives
A provider's settlement file lands: 18,402 card payments in that provider's own format, every day.
- 02
A reading agent reads it
It turns each provider's format into ledger lines, with a confidence score for every field. It can read the file and nothing else.
- 03
Rules check it
Rules in code own what must be exact: counts, amounts, duplicates and the agreed fees. No model decides a number.
- 04
An investigating agent prepares the exception
One payout is €38.20 short. A second agent with read-only access gathers the evidence and drafts the reason. Finance approves.
- 05
Posted once, and logged
After approval, the workflow posts once to the ledger. Every step, rule and decision stays in the case record.
Provider settlement files
Settlement file BP-0914settlements@brisholtpay.com
Daily settlement, 14 September.
18,402 card transactions settled for €2,316,480.55 gross.
Fees deducted at source: €11,582.40. Net payout on 15 September.
References use the new format BP-yyyymmdd-n from this file.
Reconciliation runWaiting for approval
- Merchant and provider IDLedger
- Transaction countLedger
- Duplicate referencesLedger
- Payout dateBank
- Net payout amountBank
Payout €38.20 below the expected net amount (fee not in the agreed schedule)
Sent to Anna de Vries, FinanceApproveApproved by Anna de Vries, 09:43
Posted to the ledger through its API. Every step logged.
- Behind the screen
- WorkflowRuns the steps, keeps the case recordCodeRunning
- Reading agentSettlement file to ledger linesReads the file onlyDone
- RulesCounts, amounts, duplicates, feesCodeDone
- Investigating agentEvidence and reason for the exceptionReads ledger, bank, feesRunning
- Anna de VriesApproves the exceptionFinanceRunning
- PostingPosts once, after approvalWrites to the ledger APIWaiting
Modelled on the reconciliation system we built for a payments operator, matching 4.6 million transactions a period at 99.25%. Client names and figures are changed to protect confidentiality.
AI Workflows We Run in Production
Each one runs inside a client's own systems today, with a measured result.
Operations · Medical manufacturing
Multi-market commercial operations
Bredent Medical
- Goal
- Run four European markets as one business.
- What stood in the way
- Each market ran its own Dynamics 365 or legacy system, so orders and stock were retyped between them.
60%
less manual entry, and one real-time view of all four markets
IT · Public sector
IT ticket triage
National digitalisation agency
- Goal
- Route a national service desk's tickets to the right team, faster.
- What stood in the way
- No data could leave the agency's own servers, so cloud AI was ruled out.
Finance · Payments and gaming
Payment reconciliation
Payments operator
- Goal
- Keep reconciliation on time as transaction volume grows.
- What stood in the way
- Millions of transactions a period were matched largely by hand.
Industries we build for
- Manufacturing
- Logistics
- Retail
- Healthcare
- Financial services
- Insurance
- Private equity
- Payments
- Media
- Telecom
- Energy
- Professional services
- Public sector
Connected. Controlled. Owned.
Connected
It runs inside the ERP, CRM and banking systems you already have. No migration, and no new platform for your teams to learn.
Controlled
Rules decide amounts, limits and policy. A named person approves each exception, and every decision is logged for your auditors.
Owned
You keep the code, the rules, the tests and the operating manual. Your team runs the system after we leave.
Trusted by teams at
We ran four markets on systems that did not talk to each other. SIEL connected them without replacing anything, and for the first time we see all four in one view. The retyping went with it.
Partners we deliver with
Three Ways to Work With Us
Start with one workflow and extend only when the results justify it.
- 01
Workflow Assessment
Two weeks with your process owners and system records. You get a measured baseline and a clear decision on what to build first, or not to build.
Two weeks. Read-only access under NDA. Fixed fee.
- 02
First System Delivery
We build in your environment on your existing systems and validate against real cases before the system carries live work.
Scope and fixed fee agreed after the assessment.
- 03
Embedded Engineering
Our engineers stay alongside your team to extend the system or build the next one, each with its own handover.
Time-boxed, with a fixed fee per workflow.
Questions Before the First Call
What leadership teams ask us before the first call.
How do we choose the first workflow?
With a two-week Workflow Assessment. Our senior engineers map how the work runs, measure today's baseline and name the workflow where AI adds the most capacity, and what should not be built. Fixed fee, agreed in writing before we start.
How do we prove a system works before it goes live?
We validate each system against real cases agreed with your team and measure it against the baseline from the assessment. Approval steps and exception paths are tested with the people who will use them. The system goes live only when the results meet the thresholds you have signed off.
How does the AI stay under your control?
The AI reads and proposes. It does not act alone. Rules agreed with you set what is permitted. Any item below your confidence threshold, or rejected by a rule, goes to a named approver with the evidence. Nothing is written to your systems until it passes these checks. Data stays in the environment you specify, and every step is recorded for audit.
What do we leave behind at handover?
A production system running in your environment, owned entirely by your organisation: source code, business rules, test cases and documentation, with no licence to renew. Your team receives an operating runbook and training. Where continued support is preferred, our engineers remain embedded for a defined term.
What can we show an auditor?
The case record behind each item: what the AI read, which rule applied, who approved it and when. We agree with your audit team which controls they need to see, and hand over the rules in readable form, the test results on your own cases and a description of each control.
Can it run in our own environment?
In most cases, yes. We build in your cloud, region or data centre, and you choose the AI models, which can be swapped later. Where data must stay on your own servers, we run the models there as well, as we do for a national digitalisation agency. The setup is agreed during the assessment.
How is it priced?
Each step has a fixed fee, agreed before we start: the two-week Workflow Assessment first, then the build, scoped from what the assessment found. There is no licence or usage fee from us. You pay your own cloud and model costs, in your own accounts.
Tell us what you need.
A senior engineer replies within two business days.