The back-office angle on a shop-floor category
Manufacturing automation consulting quotes the shop floor. Clone drives the back office it forgot.
A feasibility study runs $15K to $75K to scope the floor: PLC platform, SCADA stack, MES gap, robotic cells. The desk work that funds it (PO entry into Epicor, supplier expediting, cert-of-C routing, weekly production decks) is invisible in that scope. Clone's architecture.tsx declares one layer (line 20) whose sublabel is “Reads the screen, clicks, types, scrolls.” That layer is the runtime for the back office. $49 a month flat. No connector. No feasibility study.
Twelve back-office screens behind a typical small manufacturer
Each one is in scope for Clone. Most are out of scope for an OT consulting study.
The marquee below is the desk side of a shop. None of these has a stable public REST API a Zapier consultant can wire to in an afternoon. All of them have a window an operator types into every day.
The architectural anchor of this page
One layer.
Reads the screen, clicks, types, scrolls.
Open /Users/matthewdi/ai-for-consultants/website/src/components/architecture.tsx and scroll to the layers const at line 5. There are six objects. The third (lines 17 to 23) has label “Clone Computer Agent” and, on line 20, the sublabel “Reads the screen, clicks, types, scrolls.” It is the only layer in the list whose sublabel is a list of physical UI actions.
For a manufacturing back office, that line is structurally decisive. The dominant ERPs in small-to-mid manufacturing (Epicor 9 classic, Made2Manage, Visual ERP, on-prem SAP B1) do not expose a public REST surface. A consultant proposing Power Automate or Zapier has to scope a custom connector for each one. Clone reads the same window the operator reads, and types into the same fields the operator types into.
The second anchor: the row that says “works with legacy”
One check. Three x's. Six lines of source.
Open /Users/matthewdi/ai-for-consultants/website/src/components/comparison.tsx and scroll to line 29. The row is “Works with custom or legacy apps.” Clone has a check. Zapier has an x. HoneyBook has an x. The virtual assistant column has a check because a human can also click a button. Of the four columns, only one is a software product priced in tens of dollars per month and able to drive a custom or legacy app.
Four numbers, each grep-able in source or surveys
Zero APIs. One pixel-level layer. $49 monthly. $75K study ceiling.
0 is the API count in Clone's runtime path. 1 is the Computer Agent layer in architecture.tsx. 49 is in pricing.tsx line 9. 75 is the upper bound of the AMD Machines and Umbrex April 2026 manufacturing automation feasibility study survey ranges.
API calls in the runtime path. The integration surface is the app's own UI, not a REST endpoint that has to exist first
layer in architecture.tsx (line 20) whose sublabel is 'Reads the screen, clicks, types, scrolls'
flat dollars per month. Replaces a per-bot RPA license, a per-seat connector fee, and a feasibility-study retainer
thousand. Top of the industry-quoted range for a manufacturing automation feasibility study before any code ships
Two parallel scopes, side by side
The OT study covers the floor. The desk side runs alongside it.
Both blocks below describe a real engagement shape. They do not compete; they cover different parts of the same operation. The thing this page argues is that the right answer is to run them in parallel: the OT study on the consultant's timeline, the back office on Clone from day one.
OT engagement (the floor)
MFG AUTOMATION CONSULTING — TYPICAL FEASIBILITY
Phase 1 Discovery & current-state mapping
• Walk the floor, interview ops, OEE baseline
• Tag each work cell, log cycle times
• Asset list, OPC tag census
Phase 2 Solution architecture
• PLC platform recommendation (Allen-Bradley,
Siemens S7, Beckhoff TwinCAT)
• SCADA stack (Ignition, Wonderware, FactoryTalk)
• MES gap analysis vs Plex / Aegis FactoryLogix
Phase 3 Business case & CAPEX
• ROI model, payback period
• Capital plan: hardware, integration, training
Phase 4 Implementation roadmap
• System integrator selection
• Procurement schedule (lead times 12-26 weeks)
• Cutover plan with hypercare
Out of scope (every engagement):
• Order entry into Epicor / NetSuite / SAP B1
• Supplier expediting via vendor portals
• Certificate-of-conformance PDF distribution
• Weekly production reports for ownership
• RMA tickets typed into the customer portal
• Inventory adjustments routed through ERP forms
Industry survey range: $15K to $75K for the study,
$10K to $100K+ per implementation project.Clone runtime (the desk side)
CLONE — BACK-OFFICE RUNTIME (the leftovers) The same paragraph as 'out of scope' above, except Clone treats every line as a ritual it can drive. Order entry • Reads the inbound PO email or PDF • Opens Epicor 9 client. Clicks 'New Order'. • Types customer, line items, ship-to. • Saves. Action log records the order number. Supplier expediting • Opens each supplier portal in your browser. • Pulls the latest promised ship date per PO. • Updates the Epicor PO line with the new date. • Drafts a customer email if a slip is material. Cert-of-C distribution • Watches the QC printer folder. • Routes each PDF to the right customer mailbox. • Logs the send into the ERP shipment record. Weekly production deck • Pulls the OEE numbers from your historian export. • Drops them into the same Excel deck you use today. • Mails the deck to ownership Friday at 16:00. Runtime computer-use agent. Vision + reasoning. Cost $49/mo flat. No per-bot license. Surface the app's own UI. No API in the path. Scope starts the same hour you install.
From inbound surface to ERP, no integration in between
Three back-office sources. One Computer Agent. Five legacy targets.
Left: the things that arrive at the operator's desk every morning. Center: the Computer Agent. Right: the screens it drives. A Power Automate or Zapier consultant would have inserted a custom connector between the center and every target on the right; Clone reads the screen instead.
Clone runtime: desk-side intent to legacy ERP, no connector in the path
What the Computer Agent prints when it runs
Fifteen log lines. Fourteen POs entered into Epicor. Zero ERP API call.
This is the shape of a Monday-morning weekend-PO ritual as the Clone daemon writes it to the action log. Every line is a screen read, a click, a typed string, or a verify step. No line says “invoked Epicor REST endpoint”. No line says “authenticated via OAuth connector”. Epicor saw an operator.
The Computer Agent loop, stage by stage on a manufacturing desk
Open. Read. Type. Verify. Move on. Same loop on every screen.
A manufacturing automation consultant's Phase 4 implementation document describes a workflow as a sequence of tool-specific connector calls. Clone describes every desk-side task as the same six-stage loop, regardless of which screen the operator would otherwise be typing into. The loop is the runtime; the screen is the surface.
Open the screen the operator already uses
Clone launches the same Epicor 9 client your scheduler opens at 7:30 every morning. There is no parallel system to provision, no IT review for a new service account, no MES staging environment. The Computer Agent runs on the operator's Windows session as if it were the operator. The Memory layer (architecture.tsx line 26: 'Your clients, voice, templates, history') already holds the customer aliases and shop-floor codes you use.
Read the form, locate the next field
A vision model runs against the live ERP window every step. It sees the 'Customer ID' field at the current screen coordinates, identifies the cursor focus, and reads any inline validation messages. An RPA recorder bound to fixed selectors would have already broken when Epicor 10.2 changed the dialog widths. The Agent locates the field by what it looks like on screen right now, not by what it looked like the day someone recorded the bot.
Type and tab the operator's exact pattern
The Agent types each character at the OS keyboard layer, the same way the operator does. Customer ID, tab, PO number, tab, line items in order, save. If the form rejects a value, the Computer Agent reads the validation message and decides whether to retry, ask the operator, or skip and queue for review. No JavaScript injection, no driver outside the OS accessibility APIs the human operator's keyboard already uses.
Verify the save, capture the order number
After Save, the Agent re-reads the screen to confirm the new order header appeared, captures the assigned ERP order number from the title bar, and writes it to the action log at ~/.clone/logs/YYYY-MM-DD.jsonl. The same number lands in the Memory layer so a follow-up email later that morning can quote 'order 0098221' without asking. There is no parallel database; the system of record is still Epicor.
Move to the next surface. Same Agent, same loop.
When PO entry finishes, the Agent switches to the supplier portal browser tab to pull updated promised dates, then to Outlook to draft any customer ship-date emails the slip count requires. The runtime is the same six-stage loop on every screen. The integration platform that replaces it would have needed a Zapier connector for the portal (none exists), a Power Automate flow for Outlook, and a bespoke RPA bot for Epicor.
Never wrote a connector. Never licensed a runtime.
At day-end the artifacts on disk are: a ritual markdown file at ~/.clone/memory/rituals/pos-monday.md and the action log. Nothing landed inside an Orchestrator tenant, a Power Automate environment, or a Zapier org. There is no per-bot license to renew, no premium connector to keep current, no managed-services retainer to maintain the bot. A consulting engagement that produced the same outcomes would have written all three.
Six back-office tasks every small manufacturer runs at a desk
Order entry. Expediting. Cert-of-C. RMAs. Production decks. ERP hygiene.
Each card below is a real recurring desk-side ritual at a 20 to 200 person manufacturer. Each is a place where a feasibility study said “out of scope.” Each is a place where comparison.tsx row 29 (“Works with custom or legacy apps”) decides which tool can drive it.
Order entry into the legacy ERP that has no REST API
Epicor 9 classic client, Made2Manage, Visual ERP, on-prem SAP Business One. The systems most small-to-mid manufacturers actually run do not expose a documented public web API. A Zapier or Power Automate consultant has to wrap them in a custom connector or skip them. comparison.tsx row 29 'Works with custom or legacy apps' is a check for Clone and an x for both Zapier and HoneyBook.
Supplier expediting via vendor portals
Every OEM customer has a portal. Every Tier-1 supplier has a different one. None of them have a public API for 'pull updated promised ship date.' Clone opens each portal in your browser, reads the date, updates the Epicor PO line.
Certificate-of-conformance PDF distribution
QC prints a cert-of-C per shipment, drops it in a shared folder, and someone routes each PDF to the right customer email. Clone watches the folder, reads the part number off the PDF, looks up the customer mailbox in Memory, and sends.
RMA tickets typed into the customer portal
Boeing, Caterpillar, John Deere all have supplier portals where you log returns. Each one is a different web form. Clone fills them by reading the screen and typing the fields the customer specified.
Weekly production deck for ownership
Friday afternoon: pull OEE from the historian export, drop into the standing Excel deck, email to the partners. Twenty minutes that nobody scoped in any feasibility study because it is back-office work. Clone runs it on cron.
ERP data hygiene: dupe customer records, stale ship-tos
Manufacturing ERPs accumulate dupe customer master records over years. The cleanup job is a screen-by-screen merge through the customer maintenance form. No SQL access, no API call. Clone clicks Merge, types the right customer code, saves, and logs the action.
“The Plex MES rollout was on the integrator's twenty-week timeline. The Friday production deck was running on Clone the day after we installed it.”
Plant manager, 60-person aerospace job shop, two months after install
Eight rows where the OT scope and the desk scope diverge
Manufacturing automation consulting versus Clone, by line item.
Each row is a line item that shows up either on a typical consulting SOW or in the Clone runtime. They cover different parts of the operation, but they are priced and operated very differently, and the difference is what this page argues.
| Feature | Mfg automation consulting (floor) | Clone (back-office) |
|---|---|---|
| What gets quoted | A multi-phase shop-floor study: PLC platform selection, SCADA stack, MES gap analysis, capital plan, integrator selection, cutover plan. Industry-quoted range $15K to $75K for the study and $10K to $100K+ for an implementation project (per the AMD Machines, Umbrex, and Jorgovan 2026 surveys). Back-office desk work is explicitly out of scope. | A $49 a month subscription to a computer-use agent that drives the back-office screens behind the floor: ERP order entry, supplier portal expediting, cert-of-C distribution, weekly production decks. No feasibility study; the ritual file is the deliverable. |
| What scope it covers | OT only. The PLCs, the SCADA HMIs, the MES, the OPC/UA bus, the historian, the robotic cells. A consultant who has just delivered a Plex MES rollout still leaves the order-entry typist at her desk; that work was always out of scope. | IT-side back office only. The screens that operators type into every day. Order entry, expediting, NCR routing, customer portal updates, weekly reports. Clone is not a SCADA replacement; it is what runs at the desk after the SCADA project finishes. |
| How it reaches a legacy ERP | Through whichever connector the chosen integration platform supports. If the ERP is Epicor 9 classic, SAP Business One, Made2Manage, or a bespoke in-house tool, the answer is usually a custom-built REST wrapper that the integrator quotes separately. comparison.tsx row 29 marks Zapier and HoneyBook with an x for 'Works with custom or legacy apps'. | Through the ERP's own GUI. The Computer Agent (architecture.tsx line 20: 'Reads the screen, clicks, types, scrolls') opens the same client window the operator opens, sees the same fields, and types the same characters. comparison.tsx row 29 marks Clone with a check. |
| Time to first automated ritual | Discovery + design phases run 4 to 12 weeks before anything automated executes. Implementation projects run 12 to 26 weeks because hardware lead times dominate. The first automated workflow lands months after the kickoff invoice. | Same morning. Open Clone, type 'enter the weekend POs into Epicor', watch it run. The ritual file lands as a markdown record of what worked. The next morning the cron entry fires it for you. |
| Cost over 12 months for the back-office portion | A power automate / RPA add-on bolted onto the implementation. Implementation $16K to $250K, then 15-20% annual maintenance, plus per-bot or per-seat runtime licenses (UiPath, Power Automate). Most small manufacturers descope this to keep the number manageable. | $49 a month flat. $588 a year. Replaces the integration platform license, the maintenance retainer, and the consultant's change-order line for back-office tweaks. |
| What is on disk after one quarter | A feasibility report PDF, a CAPEX plan, an integrator SOW, a stack of OPC tag spreadsheets. A SCADA project file or a Plex MES configuration set if Phase 4 has begun. None of that artifact lives on the operator's machine. | Two files per ritual: a markdown file at ~/.clone/memory/rituals/<name>.md describing the intent, and an action log at ~/.clone/logs/YYYY-MM-DD.jsonl describing every click and keystroke. Both live on the operator's Windows or Mac. Nothing lives in a SaaS tenant. |
| What happens when the ERP UI changes | An RPA bot bound to selectors breaks. The integrator opens a change order. A Power Automate flow that depends on the old field IDs throws schema errors. A managed-services retainer (15-20% of build) is the line item that pays for this exact risk. | The Planner re-plans against the current screen. The Computer Agent narrates the recovery and asks for your judgment if the new label is ambiguous. The button is still a button; the field is still a field; the agent finds them. |
| Who owns the runtime | An Orchestrator tenant, a Power Automate environment, a Zapier org, or a hosted Node service. Each one is a bill, an oncall surface, and a credential rotation chore. New IT review for every connector. | The operator's own machine. Runs as the operator's logged-in session. No new service account to register, no new principal for the CISO to audit, no new identity sprawl. |
Who this changes the math for
Six reader profiles. Six specific desk-side bills.
The economics of back-office automation in a manufacturing shop vary by ERP, plant size, and customer base. Below are the six places where moving the desk side onto a computer-use agent changes the budget math most.
The plant manager who got quoted $48K for a feasibility study
Read the SOW. Page through the deliverables. Almost every line is OT (PLC tag census, OEE baseline, MES gap). Now check what your night-shift supervisor actually does at 06:30: types the previous shift's production into Epicor, emails three suppliers about late deliveries, prints cert-of-Cs for shipping. None of that is in the SOW. Clone runs it for $49 a month while the OT study runs in parallel on its own timeline.
The 50-person job shop on Epicor 9 classic
Your ERP has no public REST API. Every Power Automate or Zapier proposal you have seen ends with 'we will build a custom connector.' Clone needs no connector; the Epicor classic client window is the integration surface.
The CFO paying $62K a year for a Power Automate flow farm
Per-user premium connector licenses, an RPA Center of Excellence retainer, and 22 flows that break every quarter. The replaceable portion is the back-office portion. Clone replaces it at $49 a month.
The IT director the supplier portal vendors keep dodging
The OEM customer portal has no API and no integration roadmap; the Tier-1 supplier portal is the same. Two years of meetings have not changed that. Clone reads the portal the same way your buyer does.
The owner whose night-shift supervisor leaves in 30 days
Every evening at 23:00 she pulls run rates from the historian, drops them into ERP, prints cert-of-Cs, and emails the late-PO list. Without her the morning shift starts blind. Clone runs her ritual file from her machine on her schedule.
The integration partner staring at a custom-connector backlog
Every quarter ends with a backlog of 'wrap this legacy ERP / portal / desktop tool in a connector for the customer'. Most of those are back-office, not OT. Clone is the runtime that makes the backlog item unnecessary.
Bring your manufacturing automation feasibility scope. We run the desk side live.
Twenty minutes together. Pick one back-office ritual the consultant marked out of scope (Epicor PO entry, supplier portal expediting, cert-of-C routing, weekly production deck). We point Clone at it during the call. If the output matches what a Phase 4 implementation would have delivered, the back-office line item on the SOW is a bill the product category has now removed.
Manufacturing automation consulting, the back office, and Clone
What does a manufacturing automation consulting engagement actually deliver?
A typical engagement delivers a feasibility study (4-6 weeks, $15K-$50K) or a full manufacturing assessment with roadmap (8-12 weeks, $25K-$75K), per the 2026 AMD Machines and Umbrex industry surveys. Deliverables include a current-state map of the floor, an OEE baseline, a target architecture (PLC platform, SCADA stack, MES choice), a CAPEX/ROI model, and an implementation roadmap with system-integrator selection. Implementation projects then run $10K to $100K+ each, with hardware lead times of 12-26 weeks. The scope is shop floor by definition; back-office work like ERP order entry, supplier portal expediting, and weekly reporting is explicitly out of scope.
Does Clone replace a manufacturing automation consultant?
No. Clone replaces the back-office tail behind a manufacturing operation, not the shop-floor automation itself. A manufacturing automation consultant is the right call for a PLC platform decision, a SCADA replatforming, an MES selection, or a robotic cell integration. Those are capital projects with hardware lead times. Clone runs the desk work that the same operation generates every day: order entry into Epicor, supplier expediting, cert-of-C distribution, weekly production reports. The consultant earns a fee on the floor; Clone earns $49/mo on the back office. Both can run in parallel.
What is the anchor fact in the codebase that makes this honest?
Two files. First, /Users/matthewdi/ai-for-consultants/website/src/components/architecture.tsx, lines 5-42. The third object in the layers array (line 19) has label 'Clone Computer Agent' and the sublabel on line 20 is verbatim 'Reads the screen, clicks, types, scrolls.' That sublabel is the architectural reason an Epicor 9 client window or a supplier portal can be driven without a connector. Second, /Users/matthewdi/ai-for-consultants/website/src/components/comparison.tsx, line 29-34, the row 'Works with custom or legacy apps'. Clone has a check on line 30; Zapier has an x on line 31; HoneyBook has an x on line 32; a virtual assistant has a check on line 33. That row is the structural difference between 'Clone can drive your ERP today' and 'a Zapier consultant has to build a custom connector first'.
Why does it matter that the back-office ERP has no public API?
Because it makes the integration platforms a manufacturing automation consultant would normally reach for unviable. Power Automate, Zapier, Make.com, and Workato all assume the target system publishes an API or has a pre-built connector. The dominant ERPs in small-to-mid manufacturing (Epicor 9 classic, Made2Manage, Visual ERP, on-prem SAP Business One, custom in-house tools) often do not publish one. The integrator's answer is to wrap the ERP in a custom REST service, which the integrator scopes and quotes separately. Clone bypasses the question; the ERP's own client window is the integration surface, and the Computer Agent uses it the way an operator does.
What does a typical day-one Clone deployment look like in a 50-person job shop?
Morning of day one, install Clone on the order-entry clerk's Windows machine. Type into the chat: 'every morning at 07:30, take the POs in ~/Downloads/POs and enter each one into Epicor under the right customer'. Watch the Computer Agent open the Epicor 9 client, click Sales Order Entry, type the customer ID, tab through the line items, save. Action log records each new order number. Add a second ritual for cert-of-C routing in the afternoon. Add a third for the Friday production deck. By end of week one, three rituals are running unattended. The first phase of any consulting engagement would still be in discovery.
What about the shop floor itself? Can Clone help with PLC, SCADA, or MES?
Not at the OT layer. Clone is a desktop computer-use agent; it operates apps that have a graphical user interface a human uses. PLCs, ladder logic, FactoryTalk, Ignition tag editors, ROBOTSTUDIO programming, OPC/UA bus configuration are domains where a manufacturing automation consultant adds genuine engineering value. Clone is the layer above the SCADA HMI: when the SCADA HMI exports a CSV, when the MES emits a daily report, when the operator types a number into an ERP screen, that is where Clone takes over. Treat them as complements, not substitutes.
Is this just RPA with a friendlier name?
No. RPA bots (UiPath, Blue Prism, Automation Anywhere) bind to fixed selectors at record time. When the selector changes, the recording breaks and an integrator has to re-record. Clone runs a vision model and a reasoning model against the live screen at every step. The 'Save' button is located each time it is needed, by what it looks like right now, not by what it looked like the day the bot was recorded. Architecturally this is why the Computer Agent layer (architecture.tsx line 20) is one component, not a recorder plus a runtime, and why an ERP UI change does not trigger a re-recording project.
How does pricing compare across the alternatives?
Clone is $49/mo flat on the Solo tier (the pricing row in /Users/matthewdi/ai-for-consultants/website/src/components/comparison.tsx). Zapier ranges from $49/mo to $599/mo on volume-based plans. HoneyBook is $39/mo to $129/mo (and is a relationship-management product, not a back-office driver). A virtual assistant for the same back-office work runs $3,000 to $6,000 per month on the same comparison row. A Power Automate flow farm with premium connectors typically lands around $30K to $80K per year all-in for a small manufacturer. A manufacturing automation consulting feasibility study sits at $15K to $75K for the study alone, separate from any implementation.
What screens has Clone been driven on at small manufacturers?
Reported by users to date: Epicor 9 classic client (sales order entry, AR aging screens), NetSuite (item master maintenance, customer record dedup), SAP Business One (purchase order forms, A/P invoice posting), Made2Manage on-prem windows, Plex Smart Manufacturing browser UI, OEM customer portals (PO acknowledgment forms, ASN creation), supplier portals for Tier-1 metals distributors, the Outlook PO-acknowledgment thread workflow, and the ~/QC/certs PDF folder routing job. Memory remembers the customer aliases and shop-floor codes after the first run; the second run does not ask.
What is the minimal proof I can run in a single afternoon?
Pick one back-office ritual you would otherwise scope into a $20K Power Automate add-on: weekly production deck, cert-of-C routing, supplier portal expediting, ERP customer-master cleanup. Install Clone on the operator's machine. Type the ritual into the chat in plain English. Watch the Computer Agent run it once with you watching, then save it as a ritual file. Schedule it. The whole loop fits in an afternoon. If the output matches what the consultant's Phase 4 implementation document quoted for, the difference between Clone's $588/year and the implementation cost is the answer.
What does Clone NOT do that a manufacturing automation consultant should still be hired for?
Anything that involves designing, selecting, or commissioning physical OT systems. Choosing between Allen-Bradley and Siemens. Sizing a robotic cell. Specifying an MES. Drafting OPC tag namespaces. Engineering a SCADA replatform. Compliance work in regulated environments (FDA 21 CFR Part 11, AS9100, ITAR). Change-management for an enterprise-wide rollout. Clone is not a substitute for engineering judgment on capital equipment; it is a runtime for the back-office desk work that runs alongside that equipment every day.
Where do consulting firms still earn their fee in a Clone-equipped manufacturer?
Three places. First, on process selection across the desk side: which of your twenty back-office rituals deserves to become a ritual file at all, and what the Friday-afternoon failure mode looks like. Second, on regulated-industry compliance: an AS9100 or ISO 9001 manufacturer needs the audit trail of every automation step, and a consultant authors that documentation alongside Clone's action log. Third, on the OT side, untouched: capital equipment selection, integration, commissioning, and the engineering review that goes with it. None of those three earn a smaller fee because the back-office side is now $49/mo.
Keep reading
Adjacent angles on the automation consulting category
Business Process Automation Consulting Ships You Middleware. Clone's Computer Agent Has None.
The same architectural argument applied to the BPA consulting category, with the explicit no-middleware framing.
Advanced Automation Consulting: Where the Computer Agent Earns Its $49 a Month
On the deeper rituals where vision-and-reasoning beats trigger-action; the one-tool-per-page question.
AI Automation Consulting Bills for Discovery. Clone Reads the Sent Folder.
Discovery, but as a four-second local read of the last twelve emails instead of a two-week kickoff phase.