Talk to Your ERP: Rebuilding Service Operations from First Principles for the Agent Era

TL;DR: An AI-native ERP for service companies is not a chat widget on factory software. Factory ERPs force agencies and consultancies into module sprawl, training that never sticks, and endless clicks. We rebuilt Mercury’s Business Operation Suite from first principles — five connected modules, conversation as the primary interface, MCP-ready tools, OpenClaw Skills in production, Airwallex payment pre-matching, and multi-channel chase after day 29. The point is not a prettier dashboard. The point is collapsing repetitive work so humans do judgment again.

I am James, CEO of Mercury Technology Solutions. Hong Kong — August 2026

Most ERPs still assume you run a factory.

Inventory counts. Bill of materials. Warehouse put-away. Screens designed for people whose job is to operate the software.

Service companies got dragged into the same mold anyway. Agencies. Consultancies. Professional services firms. You buy the platform, then spend six months teaching people where the buttons live. Six months later they still open Excel because the ERP is slower than their memory.

We decided that was no longer acceptable — not as a slide, as an internal rebuild.

We tore the model down and put it back up for how a service firm actually makes money: win work, deliver it, buy what you need, track cost, invoice, collect, and see the books without a ritual export every Friday.

That rebuild is what I mean when I say talk to your ERP.

Why factory ERPs fail service companies

Short answer: Conventional ERP complexity optimizes for inventory edge cases that rarely appear in professional services. Service firms still pay the tax every day in training, duplicate entry, and context switching.

You feel that weight as:

module sprawl nobody asked for
training sessions that decay in a week
the same fields typed twice because Sales and Project refuse to share a brain
context-switching between “raise PO,” “log expense,” “check if this is invoice-ready,” and “where did cash land”

That is not digital transformation. That is digital cosplay with a license fee.

The real cost is not software. It is human attention spent on navigation.

In a labor-scarce market, administrative bloat will kill you. If your best people are matching bank lines and hunting invoice PDFs, you are burning scarce capital on chores a system should own.

What does a service company actually need to run cleanly?

We asked one boring first-principles question and refused to decorate the answer.

What loops must stay clean for a service firm to run?

Win the work
Deliver the work
Buy what delivery requires
Track cost
Invoice and collect
Understand the financial picture

Everything else is optional until those loops are tight.

So we limited the surface area to five modules, deliberately:

| Module | Job | |--------|-----| | Sales | Quote → order → commercial truth | | Purchase (PO) | Buy what delivery needs, with trail | | Expense | Cost that is not PO-shaped | | Project | Delivery spine, tied to sales and invoicing | | Accounting | Books that reflect reality without spreadsheet archaeology |

The non-negotiable design choice: Project is not a side app. It links directly to sales and invoicing. When that link is soft, teams invent parallel trackers. When it is hard, the handoff stops leaking.

Strip the edge-case cathedral. Keep the essential loops. Make the connections explicit. Most of the duplicate entry dies on contact.

Why is training people on ERP screens a tax?

Short answer: Every hour spent teaching menu trees is an hour not spent on client work. In the agent era, the primary interface should be conversation backed by real business logic — not another tab pilgrimage.

Here is the line that actually matters:

Training people on user interfaces is a tax on the business.

Every new hire should not need a tour through menu trees before they can raise a PO or check whether a project is invoice-ready. That is 2005 thinking with a 2026 logo.

The primary interface should no longer be a screen.

The primary interface is conversation — on the channels people already live in — backed by tools that execute real business logic.

Think of the old ERP as a fortress of forms. You train people to walk the corridors. The agent-era version keeps structure and controls, but operators speak intent at the gate and the system moves the right pieces.

Not a chatbot taped onto legacy forms. An agentic AI layer that operates the actual ERP.

How do MCP and OpenClaw turn ERP into an agent?

Short answer: MCP makes tools discoverable. OpenClaw Skills call those tools against live modules. Staff state intent in natural language; the agent executes sales, PO, expense, project, and accounting actions without forcing UI fluency.

We made the stack MCP-ready (Model Context Protocol). Capable agents can discover and call tools cleanly instead of scraping UI or guessing API folklore.

On top of that we built and pressure-tested OpenClaw Skills against the five modules. Not a hackathon demo. Stable enough for real internal use.

What staff do now:

create a sales order
raise a purchase order
log an expense
check project status against invoice readiness
pull accounting numbers

…in natural language, through OpenClaw, without memorizing which tab owns which truth.

This is the shift most “AI ERP” decks miss. They add a chat widget that still dumps you into the same screens. We inverted it:

People talk to the agent. The agent talks to the system of record.

UI still exists for audit, exception, and deep work. It is no longer the tax you pay to start the day.

How do you close the revenue loop without manual payment matching?

Short answer: Wire payments to invoices with AI pre-matching, notify humans only for confirmation, and auto-chase overdue invoices on the client’s preferred channel after a fixed threshold — for us, day 29.

Operations theater is useless if cash still depends on someone with three browser tabs and a grudge against bank portals.

We wired a direct API path to Airwallex. When a client deposits into the account, AI pre-matching runs against outstanding invoices. Staff get a notification of the proposed match instead of archaeology through statements and email threads.

If an invoice sits unpaid past 29 days, the system triggers controlled multi-channel follow-up — WhatsApp, WeChat, LINE, Telegram, email — whichever channel the client actually answers. Tone stays governed. Calendar memory does not depend on a human having a good week.

Daily business intelligence arrives through OpenClaw. Nobody has to open a dashboard shrine or export a spreadsheet to know what moved.

Stop hunting cash. Start deciding what to do when cash is late.

How should an ERP support how service firms actually sell?

Two more capabilities match agencies and consultancies — not factories:

1. Partner access + commission logic External and internal partners need correct visibility and automated commission calculation. Spreadsheet commission is how trust dies quietly.

2. Contract management for recurring revenue Retainers, subscriptions, multi-period agreements are first-class — not PDFs living in someone’s Downloads folder.

On the front of the funnel, lead generation and lead capture pull from social and messaging into internal CRM, then flow into quotation and into contract or sales order. The handoffs that usually force re-keying are exactly where revenue leaks. We removed the re-keying by making one record travel the whole path.

Lead → quote → contract/order → project → invoice → payment match → books.

That lead-to-cash chain is the product.

What changes in daily work after an AI-native ERP?

The chores that used to eat hours:

matching payments
chasing late invoices
compiling status packs
re-entering lead details
calculating commissions
bouncing between modules to keep project and invoice aligned

…are automated or reduced to a short conversation with the agent.

People spend time on judgment, client work, and decisions. The system handles the rest.

That is the only productivity story I care about. Not “AI wrote an email.” AI collapsed the coordination tax so scarce humans stop acting like middleware.

We did this inside Mercury Technology Solutions to streamline manual process and put staff back on creative and high-leverage work — the work you actually hired them for. That is digital transformation with a receipt: fewer clicks, tighter loops, more judgment hours.

The Service Ops Stack

If you want the mental model on one line:

Five modules · hard Project↔Sales/Invoice links · conversation interface · MCP tools · payment pre-match · day-29 multi-channel chase · partner/commission · recurring contracts · lead-to-cash without re-key.

Call it an AI-native ERP if you need a category. What it really is: service operations rebuilt so agents can run the loops and humans can leave the click farm.

FAQ: Talk to your ERP

What is an AI-native ERP for service businesses? An AI-native ERP is a system of record designed around service loops (sales, delivery, purchase, expense, accounting) where agents can execute business logic through tools — not a manufacturing suite with a chatbot bolted on.

How do you talk to your ERP with OpenClaw? Staff send natural-language requests on channels they already use. OpenClaw Skills call MCP-exposed ERP tools to create orders, raise POs, log expenses, check project–invoice readiness, and pull accounting figures.

Why rebuild instead of adding another dashboard? Dashboards do not fix duplicate entry, soft Project↔Invoice links, or payment archaeology. If people still need weeks of screen training before they can work, the foundation is wrong.

Stop adding dashboards to a broken foundation

If your current system still requires people to be trained on screens before they can do their jobs, another BI tile will not save you.

You are decorating a factory tool and hoping it becomes a services operating system.

Stop buying complexity designed for inventory. Start demanding systems designed for delivery, cash, and conversation.

Question the foundation. Then rebuild the loops. Then let the agent drive the loops while your people do the work only people should do.

That is what the agent era looks like when you build an AI-native ERP for service companies instead of adapting manufacturing software and calling the chatbot “innovation.”

Mercury Technology Solutions: Accelerate Digitality.