Scaling AI in Manufacturing: How Harney & Sons Grew From One Pilot to Four Teams
Scaling AI in manufacturing means taking one proven workflow and extending the same data, rules, and controls to the next team without starting over. Harney & Sons Fine Teas did this on SAP Business One with Forge, the AI agent from Third Wave and Structify, and grew from one production pilot to 15 people across four teams.
Most manufacturers get the first win. One workflow goes live, the pilot team trusts the output, and then progress stalls. The second and third workflows need clean shared data, written rules, and access controls the pilot never had to solve. Harney solved all three, and the path is worth studying before you commit budget to your own AI roadmap.
This post draws on the Harney & Sons case study from Structify and our own work with the Harney team. For the step-by-step view of the two workflows Harney showed on stage at SAP Business One Customer Day, read our SAP Business One shop floor data capture recap. Here we cover the layer underneath: the four breaking points Harney cleared on the way from one team to four.
Why AI Pilots in Manufacturing Stall After the First Win
AI pilots in manufacturing stall when the first workflow succeeds on data and rules nobody else shares. The pilot team fixes records by hand and remembers the exceptions. The next team inherits none of this, so every new workflow on SAP Business One starts from zero and adoption stops at one department.
Harney & Sons is a family-owned tea blender in Millerton, New York. The company has blended and sold tea since 1983, carries more than 300 teas and herbal blends, and has worked with Third Wave on SAP Business One since 2022. Behind the heritage brand sits a real production operation, with sachets, filling, blending, and bagging each running on a separate process.
Before Forge, much of Harney’s production ran on paper. Staff filled in forms on the floor, then retyped the details into spreadsheets by hand. The retyping slowed decisions, introduced errors, and left leaders unsure which numbers to trust. If your plant runs the same loop, you know the month-end routine: the report exists, and nobody wants to sign off on the figures.
Harney set one condition from the start. The team wanted clearer data and faster decisions across the company, with no new risk to the systems already running the business. You should not have to trade control for speed to put AI to work. The four breaking points below show how Harney kept both.
Breaking Point One: Production Data Nobody Trusts
Production data loses trust when capture happens on paper and someone retypes the record later. Harney & Sons filled in 20+ paper forms across sachets, filling, blending, and bagging, then retyped handwritten details into Smartsheet. Forge replaced the paper with structured digital forms validating every entry at the point of work.
The cost of the old loop showed up in two places. Retyping created delays, so production data reached decision-makers late. And paper offered almost no validation, so a transposed batch number or a wrong quantity traveled into the spreadsheet untouched. By the time the figure reached SAP Business One, the error sat inside inventory and product cost. We break down the wider cost of this pattern in The Hidden Cost of Spreadsheets in Manufacturing and Distribution.
Validated capture fixes half the problem. The other half is the product record itself. When item names, pack sizes, and specifications differ between the floor, the order desk, and SAP Business One, a clean entry still points at inconsistent data. Harney added a dedicated product information layer, a form of product information management (PIM), to keep product data consistent across the business. Product teams now work from one record instead of several.
The lesson for your plant: scale starts with the first record, not the fiftieth workflow. Fix capture and product data before you ask AI to reason over either. Our ERP data readiness check walks through the signs your data is ready and the signs of a gap.
Breaking Point Two: Answers Locked in Separate Systems
Leaders stop trusting AI answers when data sits in silos and records never line up across sources. Forge gave Harney & Sons a semantic layer, a shared map of what the data means, which links sources and matches records at scale, including more than 100,000 SAP Business One records, through read-only SAP access by default.
Before the semantic layer, Harney’s data sat in disconnected silos. Leaders had no reliable way to analyze across sources or match records at scale without slow manual effort. A question spanning production and sales meant someone pulling exports and matching rows by hand, often for hours.
A semantic layer tells the system what each piece of data means. A product name on a floor form, a line on a customer purchase order, and an item code in SAP Business One often describe the same product in three different ways. Without a link between them, every cross-team question waits on a person. With the link in place, Forge answers across all three sources directly.
For Harney’s leaders, the change is visibility. They now see production, procurement, and sales from one current picture and get direct answers to questions once taking hours of manual work. This is the shift we describe in Stop Waiting on BI Tickets: the answer arrives while the question still matters. Forge, the AI agent for SAP Business One, reads the SAP Business One Service Layer live and reaches adjacent systems through 200+ pre-built connectors, so the same layer extends as your stack grows.
Breaking Point Three: Automation Outrunning the Controls
Automation outruns control when new workflows go live faster than the approvals, permissions, and rules meant to govern them. Harney & Sons closed the gap with three controls in Forge: read-only SAP Business One access by default, narrowly scoped permissions for every automation, and approval before any automation connects to a system or acts.
Structify’s case study names this gap directly. As automation scaled at Harney, the controls lagged behind, and approvals, access, and business rules drifted out of step. Most growing manufacturers see the same drift. Each team adds a connector or a shortcut to solve today’s problem, and six months later nobody holds a full picture of what touches the ERP.
The order workflow shows the fix in practice. Forge ingests inbound purchase orders, classifies each one, and shows a preview for approval before any write to SAP Business One. A person sees what Forge intends to post, and nothing posts until the person approves. Leadership keeps control over who reaches sensitive information, and live connector approvals keep new automation inside the same rules as the old.
Before you add a second or third AI workflow, put these questions to your team and to any vendor:
- Does the agent read SAP Business One through read-only access by default?
- Does each automation run on separate, narrow permissions?
- Does a person approve every new connection before the connection goes live?
- Does every write show a preview before posting?
- Who controls access to sensitive data, and where does the log live?
We cover the wider trust model, from supervised to autonomous, in Agentic AI in ERP: From System of Record to System of Action.
Breaking Point Four: Rules Stored in People’s Heads
Business rules stored in people’s heads break automation the moment a person leaves or a process changes. Harney & Sons wrote each rule down once, in one place Forge follows every time, and captured working know-how straight from Slack, so the context behind each workflow stays with the workflow as the business grows.
Every manufacturer runs on tribal knowledge. The supervisor who knows which customer rounds up, the planner who knows which supplier ships short, the operator who knows the override. None of this lives in the ERP. When an automation skips this knowledge, the output looks right and posts wrong.
Harney’s approach keeps workflows running when forms, vendors, or connected systems change. Consider a customer who changes the layout of a purchase order. In a brittle build, the automation breaks and someone rebuilds the integration. In Harney’s setup, the rules and context carry forward, edits stick, and the workflow keeps producing reliable output.
Now consider a new hire on the blending line. Under the paper process, the new hire learned the rules from whoever ran the shift. With the rules written into the system, every shift follows the same logic from day one, and the know-how captured from Slack stays with the workflow instead of leaving with a person.
This is the difference between automation you rebuild every quarter and automation you rely on. The first costs more every time the business changes. The second holds shape, behaves predictably, and grows with you.
How Adoption Spread From Production to Four Teams
Adoption spread at Harney & Sons because each new team started from the same trusted data and controls. Production began with structured capture at the point of work. Procurement followed with PO-to-sales-order automation, product teams with product information management, and commercial teams with shared data, reaching 15 people across the business.
What began as a production digitization project now runs across the organization. Employees work from the same data instead of across disconnected tools and programs, and more people join as additional areas come online. Teams move faster because nobody starts a new workflow by rebuilding the foundation.
Harney also finished the first phase with a clear, priced plan for the next two automations, product data and order processing. Leadership knows what each one takes and what each one costs before committing. Scaling on a known plan beats scaling on enthusiasm, especially in a food-grade operation where every change touches traceability.
The roadmap keeps building on the same base. Harney is adding two new Forge features, Interfaces and Projects, providing dedicated databases for durable memory and an auto-updating central data warehouse for faster, more accurate analysis at scale. Work continues on secure SAP Business One integration, tighter approval controls, and wider capture of institutional knowledge as usage grows.
A Five-Point Scaling Check for Your SAP Business One Team
Before you scale an AI workflow past the first team on SAP Business One, check five foundations. Each one maps to a breaking point Harney & Sons cleared between the first pilot and four teams, and a gap in any one will slow every workflow you add next:
- Validated capture at the point of work, with no retyping.
- One consistent product record across the floor, order desk, and SAP.
- A semantic layer linking sources and matching records at scale.
- Read-only access, scoped permissions, and approval before any write.
- Business rules written once and kept with each workflow.
Score yourself honestly. If two or more of these are missing, fix the foundation before you buy the next automation. The ERP assessment gives you a structured starting point, and our manufacturing team works through the gaps with you. For a wider view of what AI adds to B1 today, see AI for SAP Business One: What’s Possible Today.
How to Take Your First AI Workflow Past the Pilot
You take your first AI workflow past the pilot on SAP Business One in three steps. Book a Forge demo on your own data, pick one workflow where paper or retyping causes the most errors, and go live with review rules before extending to the next team. Connecting Forge to SAP Business One takes about 30 minutes.
Waiting carries a cost. Every month on paper capture and disconnected tools adds another round of retyping, another report nobody trusts, and another pilot stuck in one department. Meanwhile, the rules in people’s heads leave with every retirement and every resignation.
We understand running a growing manufacturer on systems no longer built for the job. Our team brings 23+ years of SAP Business One experience, 500+ implementations, a 100% go-live record, and 98.7% client retention as an SAP Gold and Master Partner. Forge starts with a free 7-day trial, then $500 per month for three user seats, and Third Wave bills custom connectors and advanced setup on a time and materials basis.
Picture your operation a year from now. Operators capture data once, at the line. Leaders ask a question across production, procurement, and sales and get an answer on numbers they trust. Every new team plugs into the same foundation, and SAP Business One stays the system of record throughout.
Book a demo with our team and we will walk through Forge on your own SAP Business One data, or read more about Forge, the AI agent for SAP Business One. Not ready for a demo? Start with the ERP assessment to see where your data stands.
Frequently Asked Questions
How do you scale AI beyond a pilot in manufacturing?
Scale AI beyond a pilot by fixing the foundation before adding workflows. Validate data at the point of capture, keep one consistent product record, link data sources through a semantic layer, and govern every automation with scoped permissions and approvals. Harney & Sons followed this path on SAP Business One and grew from one team to four.
What is a semantic layer in ERP?
A semantic layer is a shared map of what your business data means. The layer links records across sources, such as floor forms, customer purchase orders, and SAP Business One item codes, so an AI agent answers questions across all of them and matches records at scale without manual effort.
Does Forge write to SAP Business One without approval?
No. At Harney & Sons, Forge connects to SAP Business One through read-only access by default, every automation runs on narrowly scoped permissions, and each automation needs approval before connecting or acting. Inbound purchase orders show a preview for approval before any write to SAP Business One.
How many people at Harney & Sons use Forge?
Fifteen people at Harney & Sons use Forge across production, procurement, product, and commercial teams, with more joining as additional areas come online. Adoption started with production data capture and spread to PO-to-sales-order automation and product information management.
Does Forge use data from systems outside SAP Business One?
Yes. Forge reads the SAP Business One Service Layer live and connects to adjacent systems such as your CRM, SharePoint, databases, PDFs, email, and spreadsheets through 200+ pre-built connectors. Third Wave builds custom connectors where a system needs one.
How much does Forge cost?
Forge starts with a free 7-day trial with no credit card required, then costs $500 per month for three user seats. Third Wave bills custom connectors, data preparation, and dedicated implementation help on a time and materials basis.