The new SKU request. The Make Copy. The datasheet hunt. The commodity code guess. The pricelist lookup. The category that doesn’t quite fit. The sister item that’s almost right but not quite.
If your team adds products to NetSuite, you already know the dance.
Someone sends a request: can you add this new SKU? Twenty minutes later, after a careful sequence of clicks, lookups and copy-pastes, a new item exists. Multiply that by tens or hundreds of new SKUs every month and you’ve got a team that spends most of its week on data entry rather than the work that actually matters.
We’ve been looking at this problem closely for the past few months, and we think there’s a much better way.
The hidden cost of catalogue growth
It’s not just slow. It’s quietly expensive.
Manual onboarding is error-prone in ways that compound. Make Copy carries over fields you almost certainly wanted fresh, like barcodes and URLs. Commodity codes get entered as best-guesses by sales staff who aren’t customs experts. Class hierarchies drift because nobody enforces them at the field level. Descriptions inherit phrasing from a sister item that was itself a rushed copy from the one before.
Years of small mistakes compound into a catalogue that nobody fully trusts. And the team doing the work knows it.
We’ve also seen a quieter cost: the bottleneck. When item creation is the gating function on new product launches, the rest of the business either waits or works around the system with placeholder SKUs that never get cleaned up properly. Neither outcome is good.
What MCP changes
Earlier this year, Anthropic published an open standard called the Model Context Protocol, or MCP. It lets Claude talk directly to business systems through structured, permissioned tools.
In practical terms, an MCP server for NetSuite gives an AI agent the same read and write capabilities a purchasing user has in the UI, with the same security boundaries and audit trail. Combine that with the agent’s existing ability to search the web, parse PDFs, and reason about structured data, and the picture shifts.
A new workflow looks like this:
- A team member messages the agent: can you add this SKU?
- The agent confirms the SKU isn’t already in NetSuite.
- It searches the manufacturer’s site for the official datasheet, fetches it, and parses out the description, key features, and dimensions.
- It cross-references the internal supplier pricelist for the trade cost and web price.
- It infers the brand-dependent fields like country of origin and warranty terms from a sister item already in the catalogue.
- It drafts a complete NetSuite payload and presents it back to the requester for approval.
- On a green light, it creates the record.
What used to be twenty minutes of careful clicking becomes a five minute conversation. The human stays in control. The catalogue gets better data because the source is the manufacturer datasheet, not a hurried copy-paste.
Packaging the workflow as a Claude Skill
Anthropic recently introduced a feature called Skills, and it’s the piece that makes this approach genuinely deployable rather than a one-off demo.
A Skill is, in essence, a self-contained folder that tells Claude how to do one specific thing well. It holds the workflow, the field conventions, the security guardrails, the approval gates, and any reference data the workflow needs. Drop the folder into a Claude installation and the Skill is available as a single command. No prompt engineering, no per-session setup, no lengthy hand-off documents.
This matters in a few specific ways.
Consistency. Every time the Skill runs, it applies the same approval gates, the same field conventions, the same security guardrails. There’s no prompt drift between sessions or between team members. Whoever runs it gets the same workflow with the same controls.
Portability. The whole thing is a folder of markdown files. Send it through Slack, drop it into a customer’s Claude configuration, or share it through a private git repo. The receiving team can use it on day one, with no ramp-up and no setup tax.
Lower onboarding cost for customers. Before Skills, deploying a workflow like this to a new customer meant either embedding the prompts in their environment by hand or asking their team to learn how to brief Claude themselves every session. With Skills, we hand over a folder. Their team types one command and follows the prompts. The “how do I use Claude for this” cognitive cost effectively disappears.
Update once, everyone benefits. When we improve the Skill, whether that’s a new safety guardrail, a new field convention, or support for a new manufacturer, the next user gets the improvement automatically. We don’t have to send a memo, schedule a training session, or chase anyone to upgrade.
For us as a consultancy, Skills also change the shape of the engagement. We build the Skill once, tailor the references to the customer’s NetSuite setup and brand list, and hand it over. The customer owns the workflow from that point. We’re not the bottleneck, and they’re not locked into us for routine catalogue work.
Not autonomous, not no-AI
Here’s the honest truth: people get this wrong in both directions.
One camp wants the AI to be fully autonomous. Press a button, walk away, come back to a populated NetSuite. We think that’s a bad idea for a live catalogue. The cost of one bad record propagating across the system is high, and there’s no good reason to give up the human approval step on something that lands in front of sales, suppliers, and customers.
The other camp is sceptical of AI in the workflow at all. They’ve seen demos that don’t survive contact with their real environment, or worse, they’ve seen tools that promised magic and delivered marketing.
The pattern we’ve settled on sits in between, and it’s where the value actually lives. The Skill does the lookup, the parsing, the inference, the drafting. The human approves every write. The Skill’s job is to compress twenty minutes of dull prep into a thirty-second review with all the information laid out cleanly. The human’s job is the call that matters: does this look right, ship it.
Speed without control isn’t progress. Control without speed is what we already have. The middle path delivers both.
The bit that doesn’t get talked about: security
AI agents that fetch external content are a prompt-injection target. A compromised or impersonated manufacturer datasheet could embed instructions designed to redirect the workflow, set unexpected values, or exfiltrate data.
The agent on its own is not a security boundary. So we don’t rely on the agent.
Three layers we bake into any Skill that writes to production NetSuite:
Trusted-source whitelist. External fetches are restricted to a known list of manufacturer domains. If a URL isn’t on the list, the Skill pauses and asks the user before proceeding. New sources get added explicitly, never silently.
A dedicated review sub-agent. Before any extracted content is used in a write, a separate sub-agent runs in a fresh context, with no conversation history, and reviews the content for injection attempts. It returns a structured verdict. If anything looks off, the user sees the suspicious excerpts verbatim and the workflow pauses.
Surface and confirm. No write fires without explicit human approval of the parsed values. Even after the first two layers pass, the human sees the payload before it lands in NetSuite.
Each layer alone is good. The three together catch almost everything realistic, and the cost is a few seconds of latency. Cheap insurance for a live catalogue.
Because all three layers live inside the Skill, the controls travel with it. When we share the Skill with a customer, the security posture goes too. They don’t have to remember to add the guardrails themselves.
What it actually delivers
Beyond the obvious time saving, the second-order benefits are where this earns its keep.
Consistency: every new record gets the same care, the same fields populated, the same description format. The catalogue stops drifting.
Better content: descriptions parsed from manufacturer datasheets are richer and more accurate than rushed copy-paste. That feeds into better search results, both inside NetSuite and on the storefront.
Capacity unlocked: the team responsible for item creation stops being the bottleneck. Anyone authorised can self-serve onboarding requests, knowing the workflow has built-in safety.
Audit trail: every item creation is a recorded conversation, replayable later. When something needs review, the full reasoning is there.
Onboarding curve: new team members don’t need months to learn the conventions of every field on every brand. The Skill embeds the institutional knowledge so new joiners are productive from week one.
Distribution: the Skill is a portable folder. We can hand it to a customer, a partner, or a new office on a Friday, and they’re running the same workflow on Monday with no environment-specific tweaking.
Who this is for
This works particularly well for businesses that:
- Add new SKUs frequently and feel the bottleneck
- Source from manufacturers, suppliers, or partners that publish proper technical datasheets or product sheets
- Have a catalogue large enough that consistency matters
- Need a clear audit trail for inventory or compliance reasons
- Run on NetSuite and want to keep it that way
That covers distributors, resellers, manufacturers, retailers, e-commerce operators, and anyone else with an inventory of products in NetSuite that’s growing faster than data entry can comfortably keep up with.
If that’s you, this is solvable today. The technology is here, the security patterns are well-understood, and the payoff is measurable from the first month.
The deliverable, when we work with you, is a Claude Skill tailored to your NetSuite setup, your brand list, your field conventions, and your security posture. Your team gets a single command. We get out of the way.
More than this one Skill
Item onboarding is the workflow we’ve walked through here, and it’s a high-impact one in its own right. It’s also one example of a pattern that fits many other corners of NetSuite.
The same approach lands cleanly on a long list of workflows that have lived as manual processes for years because nobody quite had time to automate them properly. Customer onboarding from a CRM enquiry. Sales order intake from emailed POs. Supplier approval routing. Returns processing. Custom report generation. Inventory reconciliations. Anywhere a recurring “someone messages, someone clicks for twenty minutes, the system updates” exists, there’s probably a Skill worth building.
If you can describe the workflow clearly, we can probably package it. And because each Skill is portable and self-contained, you build a library over time. Your team’s institutional knowledge gradually gets encoded as something that can be invoked, rather than something that lives in a single person’s head.
This is also just the start of what’s possible. Claude itself is moving fast: new tools, new Skill capabilities, new integrations land regularly. As an Anthropic partner, we follow those developments closely and bring whatever genuinely fits into the workflows we build with you. The list of “what your team can hand off to an AI assistant” gets longer every quarter.
Get in touch
We’d love to hear from you. Send us an email and we’ll set up a call to talk through where AI-driven workflows could fit in your business, and what the right starting point might look like for your team.