Implementing AI for DTC Operations & Accounting
Put AI to work across
operations and accounting.
Your team wants to use Claude and ChatGPT on your own business data. The blocker is rarely the model; it's the system underneath. Software with no CLI or MCP connector can't be reached by AI at all. Data sits in silos with no single source of truth. And when AI doesn't know where to fetch clean numbers, it hands you a confident answer that's quietly wrong. Fulfil is an Agentic ERP, built to be the source your AI reads from and writes to.
Sound familiar?
The real reasons AI stalls inside DTC operations and finance teams.
"Half our software can't even connect to Claude or ChatGPT. There's no MCP connector, so the AI can't see the data."
Tools without an MCP connector or open API are invisible to your AI assistant. The data stays trapped where the models can't reach it.
"I asked the same question in three places and got three different answers."
When every team pulls from its own export, spreadsheet, or dashboard, there's no shared source of truth, and no shared definitions.
"The AI gave us a number that looked right and wasn't."
Garbage in, garbage out. If AI doesn't know which table to read or how to filter drafts and cancellations, it guesses, and guesses convincingly.
"Finance calls it net revenue. Ops calls it something else. Nobody agrees on 'available to sell.'"
Different teams work off different calculations and definitions, so the numbers never reconcile and the AI inherits the confusion.
"Every report request goes to the data team and comes back a day later."
Analysts become the bottleneck. By the time the answer arrives, the decision has already moved on.
"We spent four months building our own AI tool before we gave up."
An AI layer built from scratch, disconnected from live operational data, rarely beats AI that sits directly on top of your ERP.
One source, one clock
AI is only as good as the data it reads from.
The "same question, three different answers" problem isn't an AI problem. It's what happens when sales lives in one spreadsheet, inventory in another, and finance in a third, each updated on its own schedule. Point AI at that and you automate the confusion. Fulfil holds operations and accounting in one system, so there's a single set of numbers for every assistant, dashboard, and person to read from.
- ✓Orders, inventory, purchasing, and the general ledger sit in one place, with no exports, no nightly syncs, and no reconciliation layer between systems.
- ✓Define shared metrics once, like "available to sell," "weeks of supply," and net revenue, so every team and every AI answer uses the same definition.
- ✓AI reads governed records, not stale copies, so drafts and cancellations don't get counted as real sales.
"The key is the same source, the same clock. You get to a daily automated single source of truth, and a common and shared understanding across your team."
Grunt Style embeds an "Ask Claude" button on every tool, backed by a shared data dictionary. Ask what a metric means and Claude answers from the same definitions everyone else uses.
Live orders, inventory, financials, reports & docs
AI can't see this data. Nothing to read.
Write & delete tools off by default, enable with approvals
MCP is an open standard from Anthropic. On business plans, Anthropic doesn't use your data to train models.
If it can't connect, AI can't help
Your ERP, plugged straight into Claude and ChatGPT.
MCP is like a USB plug for AI: it's how software lets an assistant reach its data. Tools without a CLI or MCP connector and no open API are simply invisible to Claude and ChatGPT. Fulfil's MCP server lets you connect once and ask questions against your live ERP in plain English, with no SQL, no exports, and no middleware.
- ✓Query the data warehouse, search orders, check inventory, and pull reports conversationally in Claude or ChatGPT.
- ✓With write access enabled, create sales and purchase orders, post journal entries, apply payments, and adjust inventory, all from a prompt.
- ✓Claude can't reach anything you can't reach yourself. Every action is logged, and nothing changes until you approve it.
Ask in plain English. Across both sides of the business.
The same connection answers operations and accounting questions, and with approval acts on them.
Operations
- "What's my current inventory for SKU ABC123, and which warehouse has the most?"
- "Show me total sales by channel for the past 30 days."
- "Which 3PL is slipping on promised ship dates this week?"
- "Create a sales order for Acme Corp with 10 units of SKU-123."
- "Adjust inventory for SKU-456 down by 5 at Main Warehouse, reason: damaged."
Accounting
- "Show me accounts receivable aging for invoices over 30 days."
- "Analyze shipping costs as a percentage of revenue by channel."
- "Help me build a monthly accrual for our 3PL invoice from Fulfil data."
- "Post a journal entry: debit Prepaid Expenses $1,200, credit Cash $1,200."
- "Apply the $5,000 payment from Acme Corp to their oldest open invoices."
Teams that used to wait a day for the data team to pull a report now get answers in minutes, without exporting a thing.
For heavier, repeatable work
Point coding agents straight at your ERP.
When the work is large, scripted, or repeated, like batch updates, multi-step workflows, and scheduled jobs, the Fulfil CLI lets AI agents run commands and pipe results without filling the model's context window with raw data. Large language models have seen billions of terminal interactions, so they use a CLI fluently.
- ✓Works with Claude Code, Cursor, Codex, and other AI coding agents.
- ✓Structured JSON output when piped, rich tables for humans, with the same role-based permissions as the rest of Fulfil.
- ✓Use MCP for conversational exploration and data-warehouse questions; reach for the CLI when the data work gets heavy.
"I am not a developer. I'm just an operator with a passion for technology. In less than 60 days, I built an entire ecosystem my whole company is using."
No developers required
Build the tools your team needs, by describing them.
Micro apps are single-page tools that run inside your Fulfil instance and read your live ERP data. You describe the workflow to Claude in plain language and it writes the code. No React, no build tools, no hiring. It's the space between a rigid saved report and a six-figure custom integration.
- ✓Grunt Style's controller replaced a $50K/year financial-close package with a micro app built in three to five days.
- ✓Caraway's accounting team built a real-time balance-sheet reconciliation app and a month-end close tracker with no engineers.
- ✓Command centers per channel, picker performance boards, demand planning, from idea to first version in minutes.
See a micro app built from a plain-language prompt.
Package the expertise
Turn recurring analysis into one-command workflows.
Claude Skills package domain knowledge and a proven framework so a repeated analysis runs the same way every time, on your actual Fulfil data. Instead of re-explaining the method each month, you run the skill and get consistent, decision-ready output, and the knowledge transfers cleanly across the team.
- ✓Fulfillment Optimization analyzes your backlog and recommends batch picking and packing waves. One brand found it was picking a kit's components 1,300 times a month and switched to pre-kitting.
- ✓Free Shipping Threshold uses your order distribution and margins to recommend where to set the bar, measured in margin dollars, not just revenue.
- ✓Write your own SOPs in as skills, so the guidance for how to act lives right inside the tools your team already uses.
Claude, three ways
Projects, Skills, and Connectors working on your ERP.
Claude Projects
Give a team a shared Project with your definitions, SOPs, and the Fulfil connection built in. Everyone works from the same context and the same numbers, with no re-explaining the business every session.
Claude Skills
Package a recurring analysis like fulfillment batching, free-shipping thresholds, or month-end checks into a reusable skill that pulls live Fulfil data and returns the same rigorous output every time.
Claude Connectors
Add Fulfil as a connector and your ERP becomes something Claude can read and, with approval, write to. Live records, report access, and documentation search, all inside the assistant your team already uses.
An AI-native ERP partner
You don't have to figure out AI on your own.
Fulfil is an OpenAI and Anthropic Partner. Our team has completed certifications including Claude Certified Architect, ChatGPT & Codex Solutions Practitioner and ChatGPT Deployment Practitioner. We've trained thousands of users on how to use Claude and ChatGPT with their ERP, on real AI use cases across operations and accounting for DTC, not demos.
Customer stories
DTC brands already running AI-native operations and accounting with Fulfil's Agentic ERP.
“We were trying to build our own AI forecasting system. Fulfil launched their Claude integration, and it's been way better than what we spent four months trying to build ourselves.”
“Being able to utilize the MCP with Fulfil and Claude was monumental. Instead of being reactive and fighting fires, we were able to identify issues, identify slowdowns, and find any problems with any 3PLs' fulfillment process ahead of time.”
“The AI piece has been a game changer. The ability inside Fulfil for our team to use plain English to build agents, build custom reports, and self-serve on the tools we need has been huge.”
“I've been able to build out a micro app within Fulfil that refreshes and pulls in the real-time balance sheet balances. I can do what I was doing in Excel inside this micro app.”
“I'd never used this stuff before less than 60 days ago, and now I have an entire ecosystem that my entire company is using. The key is the same source, the same clock.”
“My favourite feature is the MCP that lets us integrate Fulfil with Claude or other LLM tools, and then using that interface to get different reports has been super fun.”
Agentic ERP
Run it with your own AI agents.
Fulfil is an Agentic ERP. Connect Claude or ChatGPT to your live data over MCP, point coding agents at it with the CLI, vibe-code the tools your team needs as Micro Apps, and package recurring analysis as Claude Skills, all reading from one source of truth.
FAQ
Common questions.
What does it actually take to use Claude or ChatGPT with our ERP data?
What if our other software does not have an MCP connector?
How do you stop AI from giving confident but wrong answers?
Is our data safe, and does Anthropic train on it?
Do we need developers or a data team to build AI tools on Fulfil?
Which AI assistants and tools does Fulfil work with?
What is the difference between MCP, the CLI, Micro Apps, and Claude Skills?
Ready to put AI to work
on your own data?
See how Fulfil gives your operations and finance teams one source of truth, and connects it straight to Claude and ChatGPT.