Most teams start with one person on Operations Workbench Starter, free for 30 days. Add LevelOneAI in Claude or ChatGPT, connect your email and add any exports you have, such as open POs or a schedule. Starter shows which changes put orders at risk, with the evidence, and tracks each one to a close. When the whole team needs it, with live ERP data, approvals and write-back, that's the Operations Workbench.

How do Starter, the Workbench and the Operations Engine fit together?

The LevelOneAI Operations Engine does the work: it treats each order, PO or job as one record (Work), compares what changed with what was promised (Context), works out what it affects and who needs to decide (Decision), and assigns, drafts and tracks the next step (Action), until it's resolved. Starter runs the engine on one person's email and the documents they add. A Workbench runs it across a whole team's email, ERP and CRM for one workflow, with live data, approvals and write-back.

Is there a free trial?

Yes. Operations Workbench Starter is free for 30 days, and you don't need a card. After that it's $199/month or $1,990/year (two months free), for 1 user and 1 mailbox. If you don't continue, your account becomes read-only and your data is deleted 30 days later.

What do I need to use Starter?

A paid Claude or ChatGPT plan (the free tiers don't include the access it needs) and a Gmail or Outlook mailbox. For work email, a business or enterprise plan from that provider is recommended. Documents are optional: exports from any ERP or spreadsheet work.

What's the difference between Starter and the Operations Workbench?

Starter works from one person's email and the exports they add, and you set it up yourself in minutes. It's built to show the value on your own work. The Operations Workbench connects live to your ERP and planning data, runs on its own, puts the whole team in one queue, and adds approvals, drafted follow-ups and write-back. Starter fees paid in the 6 months before you buy a Workbench are credited toward its setup.

Which documents should I add to Starter?

Whatever you have, or none. It works on email from the first run, and tells you which document would sharpen the results most. Typical ones are an open-PO export, a production schedule, open sales orders or an AR aging report. It keeps only the fields it needs from them, encrypted, never the files.

It says what it doesn't know. Each fact shows its source and when it arrived. When an email could match more than one order, it asks instead of guessing, and when a document is missing it says "impact not assessed" rather than "no impact." Your team makes the call.

How does billing work?

Starter is billed monthly or annually by card, with a receipt emailed for every payment and invoices on your billing page. Cancel anytime. A Workbench is a one-time setup fee, invoiced on the schedule in your quote, plus an annual platform fee that starts at go-live.

When does my team need a Workbench?

When the work depends on more than one person's inbox: the evidence sits in your ERP, CRM or planning data, several people need to see the same item, and someone has to make sure it closes. A Workbench gives the team one queue, one record per item, approvals and a team view.

What's built, and what do you configure for us?

The Operations Engine, the work queue, the work record, approvals, follow-up tracking and the team view are built. For each Workbench we configure the connections to your systems, the rules that decide what counts as an exception, who owns what, what closes an item, and whether approved changes are written back. Which systems it connects to, and how, is worked out during setup.

Operations Workbench Starter is free for 30 days, then $199/month or $1,990/year. An Operations Workbench is a one-time setup fee, from $20,000 per workflow, plus an annual platform fee that's included in your quote. There's no ongoing services fee: once it's live, your team runs it, and our help is optional. The platform fee covers hosting, AI processing, connector updates and improvements to the Operations Engine. It starts at go-live, and in the first year you can cancel 90 days after go-live if the agreed measures aren't met. The payback calculator compares the setup fee with what the delay costs you each year. The next Workbench reuses the connections you already have, so it costs less to set up. See Pricing for what's included.

How long does setup take?

Starter works the same day. A Workbench typically goes live in 6 to 12 weeks, depending on how many systems it connects and how complex the rules are. We agree the baseline and what success looks like before setup starts.

What happens to our ERP and CRM?

They stay where they are and keep the master data. A Workbench reads from them. Write-back is optional: whether approved updates go back, and to which systems, is decided during setup. Nothing changes in your systems without approval.

What data does LevelOneAI keep?

It's designed to keep only what it needs to run the work, such as references to the source, the facts it extracted (a date, a quantity, a price), status, owners and decisions. For a Workbench, exactly what's stored is agreed with you during setup. Starter never stores email: it keeps short labels, plus the fields it needs from documents you add, encrypted.

What will this ask of our IT team?

Starter needs no IT: you connect your own email in Claude or ChatGPT, and there's a one-page security summary to share with IT. A Workbench needs read-only access to the systems the workflow uses, granted once by whoever administers them, and typically a few hours of IT time during setup. Nothing is installed, and nothing changes in your systems without approval. See Growing Businesses if you don't have an IT department, or Enterprises for how we fit an IT review.

What happens when it isn't sure, or data is missing?

If an email could belong to more than one order or PO, it shows the candidates with their evidence and asks a person to confirm, instead of guessing. If a source is unavailable or a record can't be found, the item is marked incomplete, so your team knows the picture isn't complete. Every fact is labeled confirmed, conflicting, inferred or unknown.

Is this a product or a consulting engagement?

A product. The Operations Engine, the work queue, the work record and approvals are built. Setup configures them to your systems, rules and owners for one workflow. You pay a one-time setup fee and an annual platform fee. There's no ongoing services fee, so your team runs it without us.

Do you have a SOC 2 report?

No SOC 2 report yet. We'll walk your security team through the data flow and complete your security questionnaire on request.

What happens to our data if we stop?

You can export your work history and audit trail. With Starter, your data is deleted 30 days after you cancel or after a trial ends without a paid plan. Your ERP and CRM keep their own data, because they were the record all along.

That's decided with you during scoping, based on your security and IT requirements.

To read the messy information around a piece of work (the email, the attachment, the system field), match it to the right order or PO, compare it with what was promised, and recommend the next step. Every fact is labeled confirmed, conflicting, inferred or unknown, with its source. Your people make the call, and nothing changes in a system without approval.

Why can't we just add an AI agent?

You can, and it will do the steps. What an agent won't do on its own is keep one record of each order across email and your systems, notice a change nobody asked about, know which promise it breaks, and keep the issue open until it's resolved. That's what you'd have to build and maintain around it. LevelOneAI already does that, and works alongside the agents you use. See Why AI Agents Fail under Resources.

How is this different from process mining?

Process mining reads event logs from your ERP or CRM, so it can't see what happens in email, spreadsheets and phone calls, which is where most exceptions start. It also tells you where the process breaks; it doesn't work each exception to a close. A Workbench does.

Why not wait for our ERP or CRM vendor to add AI?

Use it. ERP AI is getting good at work inside the ERP, such as reading a supplier's email and proposing the PO change. A Workbench is for what happens next, across systems: which jobs and customer orders the change puts at risk, what the planner's sheet says, who has to approve, and whether it was actually resolved. If the problem starts and ends inside your ERP, its own AI may be all you need.

When is our ERP's own AI enough?

When the problem starts and ends inside the ERP, updating the record is the whole job, and everything the decision needs is already there. Add a Workbench when the change matters because of what it affects elsewhere (jobs, customer orders, invoices), when the numbers people trust live in a spreadsheet or portal, when resolving it takes several people and an approval, or when someone has to confirm it was actually resolved.

What if our workflow changes?

Rules, owners, thresholds and closing conditions are configuration, so they can be updated as your process changes, without a new project.

What if our team doesn't adopt it?

A Workbench brings the context to each piece of work instead of asking people to go and look for it, and it fits around the systems they already use. We also agree the measures up front, so you can see whether it's working.

All insights

General

How to Integrate AI with Your Workflow Without Adding More Work, for Food Safety and Compliance Teams

By Shanshan "Sasha" Ma

How to integrate AI with your workflow when you work in food safety and compliance: connect the evidence you already have, catch exceptions early, keep people in charge of decisions, and start with one inbox.

"The certificate expired three weeks ago. Why are we only hearing about it now?"

Two weeks before an audit, a QA manager finds out that a supplier's audit certificate has lapsed. The supplier had emailed a renewal delay. The email sat in a buyer's inbox. The project tracker still said "on track," and the document repository still showed the old certificate.

Nobody did anything wrong. The information existed. It just lived in four places, and no one was connecting them.

The problem isn't missing software

Most food safety and compliance teams already have a document repository, a project tool and written procedures. The work still slips, because it runs across all of them:

Evidence sits in the document repository.
Plans sit in the project tool.
Decisions happen in meetings.
Changes arrive by email.
Context lives in someone's notes or memory.

It's easy to call this an adoption problem: people just need to update the system. Sometimes that's true. More often, keeping the system current means someone has to keep translating everyday work into structured records: tasks, owners, dates, decisions, dependencies. That's not why anyone was hired, so it slips.

An AI assistant doesn't fix this on its own. If people have to paste context into a prompt or summarize a meeting so the AI can help, you've added a step, not removed one.

Where AI actually helps

1. Connect the evidence to the work it affects. A missing certificate, an email about a delayed review and a task waiting for approval are often the same issue. AI is useful when it links those pieces to the project, audit or corrective action they belong to, with the source attached. The output isn't another summary. It's a current picture of what changed and what's still open.

2. Catch the exceptions early. Compliance work is easy when everything goes to plan. The cost is in the exceptions: an expired document, an overdue review, two records that disagree, a corrective action assigned but never verified. AI can compare approved sources, flag the conflict, and route it to the right reviewer. It should also keep the evidence and separate a confirmed finding from one that still needs checking.

3. Move the work forward. Knowing a project is behind doesn't fix it. A useful workflow names the blocking dependency and the owner of the next step, prepares the follow-up, and checks later whether the issue closed. Decisions and sign-offs stay with qualified people.

A six-step way to start

Find where people connect the dots by hand. Look for searching across systems, chasing status and rebuilding decision history.
Map the workflow before you automate it. Note where information comes from, which system owns each record, who decides and what goes wrong. Some gaps just need a simple integration or a clearer process, not AI.
Pick one bounded use case. Examples: overdue reviews, audit evidence assembly, or corrective actions through to verification.
Use what people already produce. With proper permissions, use the emails, documents, meeting records and system data you already have. Don't create a new reporting ritual to feed the AI.
Keep evidence and approvals visible. People must see the source, what the AI inferred, what's uncertain and who approved each action. Start read-only with proposed updates, and expand only after it's proven.
Measure outcomes, not output. Count the time spent chasing, how early blockers surface and how reliably actions close, against your current baseline. Don't count summaries generated.

Where LevelOneAI fits

Your compliance systems stay the system of record. We start small, in the place where most changes show up first: the inbox.

Available today: Operations Workbench Starter. Starter runs through your own Claude or ChatGPT on one person's inbox, plus any documents you add, such as a list of supplier certificates with expiry dates. Supplier document requests, review sign-offs, audit prep and corrective-action follow-ups each become one item, with the evidence, what's still uncertain and what happens next. When a commitment changes, like a supplier pushing a renewal date, the item moves up to needs attention. In the opening example, that's the email that sat unseen for three weeks.

It reads mail through your own connector and never stores your email. It keeps short labels, plus the fields it needs from documents you add, encrypted. Free for 30 days. Nobody changes how they work, and nothing is written to your compliance systems.

Next: one workflow, running alongside your PLM or quality management platform. If your specifications and supplier documents already live in a PLM or quality management platform, it stays the system of record. These platforms are built to hold structured product and supplier data. The follow-through around that data still runs through email, meetings, shared drives and the ERP. That's the part an Operations Workbench takes on, one workflow at a time:

A. Supplier compliance evidence. Read supplier certificates, specifications and audit reports. Find evidence that's missing, expired, inconsistent or unverified, tie each issue to a supplier and product, and route the follow-up to its owner. Value: less document chasing and fewer overlooked gaps. We'd start with missing and expired documents, then add consistency checks, such as an allergen statement that doesn't match the product spec.
B. Corrective-action follow-through (CAPA). Turn audit findings, deviations, complaints and investigation notes into actions linked to their evidence. Track owners, due dates, supporting evidence and verification, and reopen anything closed without being verified. Value: fewer stalled investigations and less coordination between QA, operations, suppliers and management.
C. Audit readiness and evidence assembly. Given an audit request, find the relevant records across your PLM or quality platform and shared drives. Show gaps and conflicting versions, and assemble a checklist that links back to each source. Value: less audit prep and fewer last-minute searches.
D. Change-impact coordination. When a specification, supplier document, SOP or regulation changes, identify the procedures, products, training and reviews that may be affected, for a qualified person to assess. Your PLM or quality platform can show which products use a changed ingredient. This extends the check to the SOPs, training and reviews outside it. Value: downstream follow-up stops depending on someone's memory.

How they build on each other. A and B are the natural starting points: bounded, low-risk and easy to measure. C builds on A, because supplier evidence that's already organized becomes audit evidence. D builds on B and the evidence links the others create. All four share one foundation: a unit of work, linked evidence, exceptions, an owner for every next step, and a person approving every change.

Each workflow has a measure agreed up front, such as no expired documents at audit, fewer CAPAs closed without verification, or fewer hours spent preparing for an audit.

Then: across workflows. Once the first workflow has proven itself, two things follow:

A team view. A shared view of stalled reviews, overdue follow-ups and work waiting on someone outside the team.
The Operations Engine. Evidence from email, documents and systems is linked across workflows, so a supplier delay shows up next to the audit and the project it affects.

No compliance authority moves to AI at any stage. You expand only when the previous step has shown measurable value.

The same approach applies in sales. See Your CRM Isn't Going Away.

Try this week

Pick your three most recent corrective actions. Check how many were verified, not just closed, and how long that took.
List every supplier document set to expire in the next 60 days. Then check whether anyone outside QA knows.
Run Workbench Starter on one QA or compliance inbox. See what's waiting on you, and what changed that nobody tracked.

Don't reorganize the work around AI. Integrate AI with your workflow, across the systems you already have.

Talk to LevelOneAI about a first workflow

Talk to us