Resolon Start a conversation

AI built into the ERP you run, and proved before you switch.

We redesign how work moves through your ERP so AI does the routine steps and people approve the rest. Old and new run side by side until every total agrees.

What we mean by AI-native ERP

Three meanings are in use. We work to one.

An AI-native ERP puts AI inside the steps of how work moves through the system, on the same records your people use, with people approving what matters.

The phrase is also used for new systems built around AI, and for a vendor’s built-in assistant. We mean the first: redesigning the work on the ERP you run, with AI designed into each step instead of sitting outside it as a separate assistant.

Start where the staff time goes.

Three ways to start

  1. Redesign on the ERP you run

    Pick the processes that cost the most staff time. We rebuild each with AI in every step, on the ERP you already have.

  2. A new ERP, with AI designed in

    Choosing or moving to a new ERP? AI use cases go into the selection and the first design workshops, so roles, approvals and data are set up for AI from the start.

  3. A build beside your ERP

    Where the ERP does not cover a job, we can build an application next to it on your own database, with AI designed in.

Where AI fits in the back office

Each one is proposed by AI and approved by a person.

  • Close and reconciliation

    Matches, accrual proposals and variance explanations for review.

  • Invoice and document intake

    Read, coded and routed, with exceptions sent to a person.

  • Purchasing and approvals

    Requests checked against policy before an approver sees them.

  • Inventory and planning exceptions

    A late delivery or a hold read, with a fix proposed.

  • Billing

    Statements built from the same records, with exceptions shown first.

  • Contracts and leases

    Terms pulled into a record a person checks.

  • Questions in plain English

    Answers from the ERP, under each person’s permissions.

  • Data upkeep

    Duplicates and gaps flagged to the record’s owner.

What the AI may do

Written down before it starts, and agreed with the people who own the work.

  1. Read

    The records its own role allows, nothing wider.

  2. Suggest

    Drafts and matches, held for someone to check.

  3. Change

    Only after sign-off, or alone where a mistake is cheap to undo.

  4. Never touch

    Closed periods, bank details, and anything else you name.

Six steps, one process at a time.

How a build runs

  1. Review

    How your ERP, data and approvals work today, and where AI would earn its place.

  2. Data foundation

    The fields the AI will read, agreed definitions, and one named person who owns each.

  3. Controls and audit trail

    Each AI user holds a role of its own, and every step is logged.

  4. Build in a sandbox

    A written list of what the AI reads, suggests and changes, and a person to approve.

  5. Prove side by side

    Old and new run together until the totals agree, then the switch comes.

  6. Hand over

    Your team runs it on systems you own, with a plan for re-checking the AI whenever the ERP or the model underneath it changes.

Redesign the process. Prove the numbers match. Then switch.

Prove it, then switch

Two lanes run in parallel: the way you work today and the redesigned way.

We compare the totals each period. The switch happens one process at a time, when you say so.

Example: paying a supplier invoice

One supplier invoice, from inbox to ledger

An AI agent prepares and checks it. A named person approves it. Only then is it posted and logged.

  • The agent prepares.Steps 1 and 2
  • A named person approves.Step 3. Nothing posts before this.
  • The ledger is written.Step 4, after approval
  • Every step is logged.Step 5

AI agent does the workNamed person makes the decisionAutomatic is recorded by the systemSample layout. No real supplier, figures or dates.

The same thing in words

  1. Prepare. A supplier invoice arrives by email or portal. The AI agent reads it and drafts which accounts it belongs to. Nothing is posted.
  2. Check. The agent compares the invoice, the purchase order and the goods receipt: what was billed, what was ordered and what arrived. If all three agree, the draft goes on to approval. If they differ, the invoice is held and sent to the buyer with the reason in plain words, for example a different price, fewer items received than billed, or no purchase order found.
  3. Approve. A named person, for example the accounts payable manager, sees the draft and all three records side by side and approves it or sends it back to the agent to redraft. Nothing posts before this.
  4. Post. Only after approval, the agent writes the entry to the general ledger. It never touches closed periods or bank details.
  5. Log. Every step is recorded by the system: who or what acted, what they did, when and why. Any entry can be traced and reversed.

Today a person does the lookup, matching and coding by hand. With the agent, the same person still approves. The values on this page are a sample layout, not real data.

Eight questions to put to any ERP vendor about AI

The questions we would ask, including of the ERP you run today.

  1. Does the AI run inside the workflow on live records, or beside it on an export?

  2. Can it act, or only answer? Who approves what it does?

  3. If you switched the AI off tomorrow, would the data work still pay off?

  4. Does an AI-drafted record wait for approval before it reaches the live books?

  5. Does the log show who or what acted, when, why, on which model, and who approved?

  6. What happens if the data changes between the AI’s draft and the approval?

  7. Can you widen what it does in steps: suggest, queue for approval, act alone under a limit?

  8. If you leave, can you take the ledger, the history and the AI steps with you?

Questions we hear first

What is an AI-native ERP?

People use the phrase for three things: a process redesigned so AI works inside each step, a system built around AI from the start, and a vendor’s built-in assistant. We mean the first. The label matters less than three tests: does the AI run inside the workflow on live records, can it act with a person approving, and does the data still pay off if you switch the AI off?

Must we replace our ERP?

No. We can add AI to the ERP you already run, and old and new run side by side until every total agrees. If a replacement would serve you better, we will say so.

Where does our data go when AI reads our ERP?

Only where you decide. Before anything connects, we settle which model provider is used, what data may leave your systems, and what the provider may keep. The AI signs in with a role of its own, so it sees what that role allows and nothing more.

Who is accountable if the AI gets an entry wrong?

A named person at your company approves each decision before it takes effect, and every step is logged, so an entry can be traced and reversed. Who approves what, and where AI may not act, are written down before anything is built.

How can we tell the AI is right before we rely on it?

We run the old way and the new way side by side and compare the totals, so nothing switches until they agree. After go-live, the AI is re-checked whenever the ERP or the model underneath it changes: last month’s work is run through the new version and the answers are compared.

Can we connect Claude or ChatGPT to our ERP?

Yes, with limits set first: a role of its own for the tool, read-only to begin with, and a written list of what it may change. Which tools can connect depends on your ERP account and its vendor’s current rules, so we check both before we start.

Do the new ERPs built around AI suit us?

Some do, for finance teams with simple operations. Some leave out inventory, manufacturing or job costing, so check the scope against the work you do before you move.

Where does a build start?

A short review of how your ERP, data and approvals work today and where AI would earn its place. You receive a ranked list of processes.

Resolon works across five practice areas. Our senior team leads each engagement and adds the specialists the use case calls for.

Senior people lead the work. Our senior team leads every engagement and builds the team each use case needs. The people who scope the work stay on it through handover, and Resolon answers for the result.

Chad Neptune

Chad Neptune

Chief Executive Officer and Chief AI Architect

Chief AI Architect, fluent in the boardroom, the architecture, and the code. 20+ years leading businesses; founded and architected Lead Validator Pro, an AI-native SaaS platform deployed in a confidential enterprise insurance pilot reaching 14,000 users. Anthropic Certified: Claude Certified Architect — Professional and Claude Certified Developer — Foundations.

Read the full profile

Pick one process. We’ll run it side by side.

Old and new run together until every total agrees. Then you decide when to switch.

Share this page