Organization AI Opportunity Discovery

EveryWorks Supply

Fictional example. EveryWorks Supply is fictional. This page shows a saved Organization AI Opportunity Discovery result generated through the same workflow used by the Workbench. It is read-only. Viewing it does not call a model.

What was submitted

Show

The fictional inputs used to generate this saved result.

Organization name

EveryWorks Supply

Organization type or industry

Regional B2B distributor of janitorial, safety, packaging, and facility maintenance supplies to commercial and institutional accounts

Approximate size (optional)

About 90 employees across three branches, around $32.4M annual revenue

Problems and opportunities

Roughly 55% of our orders are typed into the ERP by hand. About 46% arrive by email — some as a PDF or spreadsheet attachment, some typed into the body of the message, some as "the usual, plus two cases of the blue gloves" — and another 9% come by phone. The portal and EDI orders, about 40% between them, land in the ERP on their own and cause us almost no trouble. A spot check last quarter found errors on about 2% of the orders we key by hand, usually the wrong item where the customer used their own part number, or the wrong ship-to on a multi-site account. Our two shared mailboxes take about 450 emails a day. A large share are the same few requests — where is my order, can I have a copy of that invoice, can you send proof of delivery, I need the safety data sheet for this product. There is no case system, so nothing is assigned and nothing is measured beyond a sample we did by hand: median first response 3.4 hours, with the worst tenth over a full business day. Quotes take a median of 1.5 business days when we want them same day. Every morning somebody rebuilds an open-order exception list in Excel from three ERP exports so we can see which orders are stuck on credit hold, on backorder, or waiting on a price. Supplier price files arrive from about 30 suppliers in about 30 formats and are keyed in by hand. Tracking numbers are pasted from the shipping software back into the ERP. The monthly KPI pack takes the Controller about two days. Several things are also just inconsistent between branches. West came to us in 2021 and still does returns differently.

Primary products or services

We stock and distribute janitorial chemicals and dilution systems, paper and hygiene products with the dispensers that go with them, safety supplies and PPE, packaging and shipping supplies, floor care equipment with its parts and consumables, matting, foodservice disposables, and a small MRO consumables range. About 14,000 SKUs are active, meaning sold at least once in the last twelve months. Roughly 62% of those we stock; the other 38% are special order or direct ship from the supplier. Customers are contract janitorial and facilities-services companies, K-12 districts and municipal buildings, healthcare and long-term care facilities, light manufacturing and warehousing, and property management companies. Almost all of it is account-based repeat purchasing on credit terms against contract pricing, not one-off transactions. Healthcare accounts will not accept a substituted product without agreeing to it first, and several of them require documentation we have to be able to produce on request.

Major processes

Order to cash: an order arrives on one of five channels, is entered or received into the ERP, is checked for price and credit, is picked and shipped or set up as a special order, is invoiced, and is collected. Quote to order for anything not on contract pricing. Purchasing and replenishment: a buyer works from a reorder workbook built out of a daily ERP export, raises purchase orders, chases confirmations, and receives against them. Customer service: order status, proof of delivery, invoice copies, safety data sheets, complaints, and returns, all of it running through two shared mailboxes. Returns and credits, which start with an RMA number, except at West. Month end: invoice runs, statements, and a management KPI pack.

Current technology

A mid-market ERP, about eleven years old and hosted, which is our system of record for customers, items, pricing, orders, purchase orders, inventory and invoices. Its reporting is rigid and mostly ends in a CSV export, and its API is read-mostly and not well documented. A CRM used by the outside sales team, though not consistently, and not linked to the ERP customer master. EDI through a third-party broker for six large accounts. A supplier-provided web portal for customer self-service ordering. Microsoft 365 throughout — Outlook including the shared mailboxes, Excel, SharePoint, Teams. Shipping and freight software at each branch. Handheld scanners for picking at North only; South and West pick from printed tickets. And a lot of spreadsheets: the purchasing reorder workbook, the daily open-order exception list, the supplier price file staging sheets, and the monthly KPI pack. IT is two people who spend most of their time on support and ERP administration. We have no developers and nobody who works with data full time.

Current automation or AI

EDI orders from six accounts post into the ERP automatically, and portal orders do the same. That is the extent of it, and those two channels are the ones that give us the least trouble. There are a few Outlook rules and a scheduled ERP report or two. Nothing else is automated. We do not use AI anywhere today, and we have no policy that says what we would and would not be willing to do with it.

Goals

Quote same day instead of a day and a half. Cut the amount of information we type twice. Reduce the errors that come out of manual order entry, because every one of them costs us a credit, a redelivery, or a customer's patience. See the orders that are stuck without somebody building a spreadsheet at 7am. Grow about 6.5% next year without adding people in inside sales or finance. And get West working the same way as the other two branches.

Constraints or concerns

IT is two people and we have no developers, so anything that needs custom integration work goes to the ERP vendor's professional services and gets expensive quickly. Capital spend over $25,000 needs the President's approval. June through August is our busiest period and the worst time to change anything. Customer pricing is commercially sensitive and we would not want it leaving our systems. There is customer contact information and some personal guarantor information in the ERP. Healthcare accounts impose documentation requirements on us. Safety data sheets have to be current. Pricing exceptions, credit limits and credit holds, credits and refunds above a threshold, purchase commitments, and any promise about a delivery date are all approvals that belong to named people, and that is not something we are looking to change. We are not interested in a system that makes commercial decisions on our behalf or that makes it unclear who is accountable when something goes wrong.

Documents used

EveryWorks-Company-Overview-FY27-Priorities.pdf
EveryWorks-Order-to-Cash-Operations-v3.pdf
Customer-Service-Playbook-2026.pdf
Purchasing_and_Inventory_Process_Notes.pdf
EveryWorks-IT-Systems-and-Data-Readiness-FY26.pdf

Recommendation mix in this saved result: AI 3, Deterministic Automation 2, Process Improvement 1.

Organization context

EveryWorks Supply is a regional B2B distributor of janitorial, safety, packaging, and facility maintenance supplies, serving commercial and institutional accounts. With approximately 90 employees and $32.4M in annual revenue, the company manages around 14,000 active SKUs. Key processes include order-to-cash, quote-to-order, purchasing, customer service, and month-end reporting. The technology stack includes an 11-year-old ERP, a CRM, EDI, a customer portal, and Microsoft 365. IT is a two-person team focused on support and ERP administration, with no dedicated developers. Current automation is limited to EDI and portal orders. Major pain points include manual order entry (55% of orders, 2% error rate), high volume of customer service emails (450/day) with slow response times and no case system, manual creation of operational reports, and inconsistent processes across branches.

Executive summary

EveryWorks Supply faces significant operational challenges stemming from manual data entry, inconsistent processes, and limited system integration. These issues lead to errors, delays, and a heavy reliance on manual workarounds, hindering their ability to grow without increasing headcount and meet customer expectations. This report identifies several opportunities to leverage deterministic automation, process improvements, and targeted AI applications to address these pain points. Key recommendations focus on automating order intake, streamlining customer service, improving operational visibility, and enhancing data management, all while respecting existing approval structures and IT constraints. A phased roadmap is proposed to deliver incremental value and build internal capabilities.

Prioritized opportunities

6 identified, most valuable first

OPP-01

Automate Email Order Entry

Problem or opportunity

Approximately 46% of orders arrive via email in various unstructured formats (PDF, spreadsheet, body text), requiring manual keying into the ERP. A spot check found errors on about 2% of manually keyed orders, leading to credits, redeliveries, or customer dissatisfaction. This is a significant bottleneck and source of errors.

Proposed approach

Implement an AI-powered solution to automatically extract order details (customer, items, quantities, ship-to) from incoming email attachments and body text. The extracted data would be validated against ERP masters and staged for human review before final commitment to the ERP. This would reduce manual effort and errors.

Potential value

Significantly reduce manual order entry time and associated errors, improving order processing speed and accuracy. This directly supports the goal of reducing errors from manual entry and growing without adding headcount in inside sales.

Feasibility

Medium. Requires an AI model capable of handling diverse unstructured data and an integration point with the ERP for staging and committing orders. Initial training data would be available from past email orders.

Estimated effort

Significant initial setup and integration, moderate ongoing monitoring and model refinement.

Human oversight

A human must review and approve all AI-extracted orders before they are committed to the ERP, especially for new or unusual items, or when the AI flags a low confidence score. The Controller retains approval for credit holds, and supervisors for pricing exceptions.

Suggested pilot

Pilot with a specific segment of customers who consistently send orders in a particular format (e.g., PDF attachments or simple body text) to refine the AI model and integration process.

Required data

Customer email content
ERP customer master
ERP item master
ERP pricing matrix

Dependencies

ERP API for staging orders
A clear data classification and external data sharing policy
Dedicated resources for AI model training and integration

Risks and guardrails

AI misinterpretation of order details (e.g., wrong item, quantity)
Data privacy concerns if external AI services are used (customer contact, pricing)
Ensuring the solution respects existing approval flows for pricing exceptions or credit holds.

Evidence (3)

E5 · EveryWorks-Order-to-Cash-Operations-v3.pdf · Page 1

That was never the policy and the error stood for eighteen months. 1.1 Volume this procedure has to carry We take approximately 320 sales orders on an average business day. Average order value in FY26 was approximately $405 across all channels, which is what makes this a volume process rather than a value process: no…

E23 · EveryWorks-Order-to-Cash-Operations-v3.pdf · Page 2

3.1 Customer part numbers Many customers order using their own internal part number rather than ours. Some of those are mapped in the ERP against the customer's account; many are not, and the mapping is maintained item by item as somebody notices the need. Where a customer part number is not mapped, the person…

E10 · EveryWorks-Company-Overview-FY27-Priorities.pdf · Page 2

3.1 Quote same day We tell customers they will have a quote the same day and we frequently do not deliver it. Customer Service holds the measurement and should keep reporting it. The target does not change this year: same day, on every quote that does not need a supplier price back from Purchasing first. 3.2 Reduce…

OPP-02

Automate Daily Open Order Exception List

Deterministic Automation

Problem or opportunity

An open-order exception list, critical for operational visibility, is manually rebuilt every morning in Excel from three separate ERP exports. This process takes a significant part of the morning, is a single point of failure, and prevents real-time visibility of stuck orders.

Proposed approach

Develop an automated script or tool to pull the required data from the ERP (open orders, credit holds, backordered lines), combine them, apply the existing rules (currently in Excel formulas), and generate the exception list. This report can then be automatically distributed to relevant stakeholders.

Potential value

Eliminate a daily manual, time-consuming task, remove a single point of failure, and provide timely, accurate operational visibility into stuck orders. This directly supports the goal of better operational visibility and growing without adding headcount.

Feasibility

High. The rules for combining and filtering data are explicit and do not change, making it suitable for deterministic automation. The ERP's read API is available.

Estimated effort

Low to Moderate initial development, low ongoing maintenance.

Human oversight

A human supervisor should review the automated report for correctness and completeness during the initial pilot phase and periodically thereafter. The Controller retains authority for releasing credit holds.

Suggested pilot

Run the automated report in parallel with the existing manual process for one month, comparing outputs daily to ensure accuracy and build confidence before fully transitioning.

Required data

ERP open orders data
ERP credit hold data
ERP backordered lines data

Dependencies

Reliable read access to the ERP API

Risks and guardrails

Potential for data discrepancies if ERP exports change format without notice
Ensuring the automated rules perfectly replicate the existing Excel logic.

Evidence (3)

E2 · EveryWorks-Order-to-Cash-Operations-v3.pdf · Page 4

The Controller assembles a management pack after each month close from a set of ERP exports; the effort involved is described in the IT systems and data readiness note rather than here. Where an invoice is queried, see the customer service playbook. Where a credit is due, the credit memo approval limits in the company…

E9 · EveryWorks-IT-Systems-and-Data-Readiness-FY26.pdf · Page 4

The daily open-order exception list has the same shape at a smaller scale: three ERP exports combined by hand every morning, described in the order-to-cash procedure. Neither of these is difficult work. Both are rule-based, both follow the same steps every time, and both exist because the ERP's reporting cannot…

E10 · EveryWorks-Company-Overview-FY27-Priorities.pdf · Page 2

3.1 Quote same day We tell customers they will have a quote the same day and we frequently do not deliver it. Customer Service holds the measurement and should keep reporting it. The target does not change this year: same day, on every quote that does not need a supplier price back from Purchasing first. 3.2 Reduce…

OPP-03

Streamline Customer Service Request Handling

Problem or opportunity

The two shared customer service mailboxes receive approximately 450 emails daily. There is no case management system, leading to slow median first response times (3.4 hours), messages being read by multiple people but answered by none, and a lack of metrics on resolution time or customer satisfaction. A large share of requests are repetitive (order status, invoice copies, SDS).

Proposed approach

Implement an AI-powered email classification system to automatically categorize incoming customer service emails (e.g., 'Order Status', 'Invoice Copy', 'SDS Request', 'Return Request'). Key entities (order number, SKU, customer name) can also be extracted. This classification can then be used to route emails to specific folders or queues, and potentially trigger automated responses for simple requests.

Potential value

Significantly reduce manual triage time, improve first response times, provide better visibility into customer service workload, and enable more efficient handling of common requests. This supports the goal of improving customer experience and growing without adding headcount.

Feasibility

Medium. Requires training an AI model on historical email data to accurately classify requests. Integration with Outlook shared mailboxes would be needed. Process changes for staff to utilize the new routing are also necessary.

Estimated effort

Moderate initial setup and training, moderate ongoing monitoring and refinement.

Human oversight

Customer service representatives must review AI classifications and extracted entities, especially during the initial phase. Humans will always handle complex, misclassified, or sensitive requests. Automated responses should be clearly identified as such and allow for human override or escalation. All final customer communication remains human-approved.

Suggested pilot

Pilot the AI classification for the two most common and straightforward request types ('Where is my order' and 'Can I have a copy of that invoice'), routing them to dedicated folders for a small team to manage.

Required data

Historical customer service emails
ERP order history
ERP customer master
Product safety data sheets

Dependencies

A clear data classification and external data sharing policy
Internal agreement on standardized responses for common requests

Risks and guardrails

AI misclassification of complex or ambiguous requests
Risk of impersonal or incorrect automated responses
Data privacy concerns if external AI services are used.

Evidence (4)

E18 · Customer-Service-Playbook-2026.pdf · Page 1

EVERYWORKS SUPPLY Customer Service Playbook 2026 For everyone who answers a customer. Written and maintained by the Customer Service Supervisor, North. Issued January 2026. Replaces the 2024 version. 1. HOW TO USE THIS PLAYBOOK This is not a procedure document. The order-to-cash procedure tells you how an order moves…

E12 · Customer-Service-Playbook-2026.pdf · Page 2

2.2 What we know about response times A sample taken in the third quarter of FY26 put the median time to a first response at 3.4 hours. The worst tenth of messages waited more than one full business day. Those numbers came out of exporting a fortnight of mailbox data and working through it in Excel, and it has not…

E13 · Customer-Service-Playbook-2026.pdf · Page 1

That figure comes from a count somebody did by hand over a week in FY26; nothing counts it for us. 2.1 How work gets picked up There is no case system and no assignment. Everyone with access sees the same mailbox and takes the next message they can deal with. In practice, the people who are quickest at a particular…

E15 · Customer-Service-Playbook-2026.pdf · Page 4

As a result we do not measure: how many requests we receive by category; how long a request takes to resolve as opposed to how long it takes to acknowledge; how many requests are reopened; how many are handled by each person; or whether a customer was satisfied with the outcome. The response-time figures in section…

OPP-04

Standardize and Track Quotes

Process Improvement

Problem or opportunity

The organization's goal is to provide same-day quotes, but the median turnaround is 1.5 business days. Quotes are tracked inconsistently across three different places (ERP, spreadsheets, email sent items), leading to no shared view of outstanding quotes and delays in follow-up.

Proposed approach

Mandate that all quotes be entered and tracked exclusively within the ERP system. Establish a clear, shared view or report of all outstanding quotes, their status, and assigned owner. Implement a standardized follow-up procedure to ensure timely responses and prevent quotes from sitting unaddressed.

Potential value

Achieve the goal of same-day quotes, improve visibility into the sales pipeline, reduce lost opportunities due to delays, and enhance customer satisfaction. This is a direct response to a stated organizational goal.

Feasibility

High. This primarily involves process changes and consistent use of existing ERP functionality, rather than new technology development. Requires strong management buy-in and training.

Estimated effort

Low. Primarily involves process definition, communication, and training.

Human oversight

Sales supervisors must regularly monitor the shared quote report to ensure all quotes are entered, assigned, and progressing according to the new process. They are responsible for addressing any bottlenecks or non-compliance.

Suggested pilot

Implement the new quote tracking process for all inside sales representatives, with weekly reviews by supervisors to ensure adherence and address any challenges.

Required data

ERP quote data
ERP customer master
ERP item master
ERP pricing matrix

Dependencies

Management commitment to enforce the new process
Training for all inside sales and customer service staff

Risks and guardrails

Resistance to change from staff accustomed to personal tracking methods
Incomplete data entry if the process is not consistently followed.

Evidence (2)

E3 · Customer-Service-Playbook-2026.pdf · Page 4

8. QUOTES Our expectation to the customer is a quote the same day. The median in a recent sample was 1.5 business days, and the tail is worse than that when the quote needs a price back from a supplier first. Most of the delay is not the pricing. It is that a quote sits with whoever picked it up, and there is no…

E10 · EveryWorks-Company-Overview-FY27-Priorities.pdf · Page 2

3.1 Quote same day We tell customers they will have a quote the same day and we frequently do not deliver it. Customer Service holds the measurement and should keep reporting it. The target does not change this year: same day, on every quote that does not need a supplier price back from Purchasing first. 3.2 Reduce…

OPP-05

Automate Supplier Price File Processing

Problem or opportunity

Supplier price files arrive in approximately 30 different formats (PDF, Excel, CSV, email body) and on varying schedules. Every file is manually read and keyed into the ERP, with no validation. Missed effective dates lead to sales at incorrect costs and margin erosion.

Proposed approach

Develop a system to ingest supplier price files. For structured formats (CSV, Excel), use deterministic rules to extract data. For unstructured formats (PDF, scanned pages, email body), use AI to extract key fields such as manufacturer part number, EveryWorks SKU, new cost, and effective date. The extracted data would be staged for human review and approval before updating the ERP.

Potential value

Significantly reduce manual data entry for supplier price updates, improve accuracy, ensure timely application of new costs, and prevent margin erosion due to missed effective dates. This supports the goal of reducing information typed twice.

Feasibility

Medium. Handling 30 diverse formats, including unstructured ones, requires a robust AI solution for data extraction. Integration with the ERP for updates would likely require vendor professional services.

Estimated effort

Significant initial setup and integration, moderate ongoing monitoring and model refinement.

Human oversight

A human buyer or purchasing manager must review and approve all proposed price changes extracted by the system before they are committed to the ERP. This includes verifying item matches, costs, and effective dates.

Suggested pilot

Start with 2-3 key suppliers whose price files are either highly structured (e.g., CSV) or consistently unstructured (e.g., a specific PDF format) to build and refine the automation and AI extraction capabilities.

Required data

Supplier price files (various formats)
ERP item master
ERP supplier master

Dependencies

ERP write API access (likely via vendor professional services)
A clear data classification and external data sharing policy
Dedicated resources for AI model training and integration

Risks and guardrails

AI misinterpretation of pricing data (e.g., wrong item, cost, effective date)
Data privacy concerns if external AI services are used (supplier pricing is commercially sensitive)
Ensuring the solution respects existing approval flows for purchase commitments.

Evidence (1)

E6 · Purchasing_and_Inventory_Process_Notes.pdf · Page 3

They arrive in approximately 30 different formats. PDF, Excel, CSV, and in two cases the price change is stated in the body of an email. They arrive on approximately 30 different schedules. Two to four files land in a typical week, with no pattern. Every one of them is read by a person and keyed into the ERP by hand.…

OPP-06

Automate Tracking Number Entry

Deterministic Automation

Problem or opportunity

Tracking numbers are manually copied from the shipping software into the ERP after labels are produced. This is a repetitive manual task that adds to re-keying effort and can delay customer access to tracking information.

Proposed approach

Integrate the shipping software with the ERP to automatically transfer tracking numbers to the corresponding sales order in the ERP. This would eliminate the manual re-keying step.

Potential value

Eliminate manual re-keying, improve data accuracy, and provide customers with faster access to tracking information, enhancing customer service efficiency. This directly supports the goal of reducing information typed twice.

Feasibility

Medium. Feasibility depends on the availability and documentation of APIs for both the shipping software and the ERP. ERP write integrations are noted as potentially expensive via vendor professional services.

Estimated effort

Moderate initial development and integration, low ongoing maintenance.

Human oversight

During the pilot phase, a human should periodically verify that tracking numbers are correctly transferred to the ERP for a sample of orders. Customer service staff will continue to be the first point of contact for tracking queries.

Suggested pilot

Pilot the automation for one branch or for orders shipped via a single, high-volume common carrier, verifying the accuracy of the data transfer.

Required data

Shipping software tracking numbers
ERP sales order numbers

Dependencies

Access to shipping software API
ERP write API access (likely via vendor professional services)

Risks and guardrails

Integration complexity and cost
Potential for incorrect tracking number assignment if order matching logic is flawed.

Evidence (2)

E11 · Customer-Service-Playbook-2026.pdf · Page 2

The work is not deciding what to say; it is finding the specific fact for this customer, on this order, in this account, and then writing it out again. 4. ORDER STATUS AND PROOF OF DELIVERY Take the status from the ERP. Do not take it from the customer's own account of what they were told last week, and do not take it…

E16 · EveryWorks-IT-Systems-and-Data-Readiness-FY26.pdf · Page 3

The CRM and the ERP do not talk. A new customer is created in the ERP by Inside Sales and, if the account is a field-sales account, created again in the CRM by the rep. The two records use different conventions for the same fields and there is no key linking them. The shipping software and the ERP do not talk. Every…

Roadmap

First 30 days

Develop a formal data classification and external data sharing policy.
Implement the new process for standardizing and tracking quotes in the ERP.
Pilot the automated daily open order exception list, running it in parallel with the manual process.

By 90 days

Expand the automated daily open order exception list company-wide.
Begin requirements gathering and vendor evaluation for email order entry automation.
Begin requirements gathering and vendor evaluation for tracking number automation.
Pilot AI-powered customer service email classification for 'Order Status' and 'Invoice Copy' requests.

By 180 days

Expand AI-powered customer service email classification to additional request types.
Pilot email order entry automation for a specific customer segment or order type.
Pilot tracking number automation for one branch or a specific shipping carrier.
Begin requirements gathering and vendor evaluation for supplier price file processing automation.

Prerequisites

Develop a formal data classification and external data sharing policy to govern the use of sensitive customer and pricing data with external tools or services.
Secure read/write access to the ERP API, or budget for the ERP vendor's professional services for necessary integrations.
Address branch divergence, particularly regarding the West branch's unique procedures, to ensure consistent processes can be applied company-wide.
Allocate dedicated internal resources or secure external expertise for the development, integration, and ongoing maintenance of automation and AI solutions, given the limited internal IT capacity.

Cross-cutting risks

IT Capacity: The two-person IT team is primarily focused on support and ERP administration, limiting internal capacity for new development and integration work.
ERP Integration Costs: The ERP's API is read-mostly, and write integrations are noted as expensive, typically requiring the vendor's professional services.
Data Privacy & Security: The organization handles commercially sensitive customer pricing, customer contact information, personal guarantor data, and healthcare-specific documentation. A lack of formal data policy for external services poses a risk.
Change Management: Implementing new processes and technologies across three branches, especially with existing procedural divergence at West, may face resistance and require careful change management.
Single Points of Failure: Several critical operational processes rely on spreadsheets maintained by a single individual, creating a risk of disruption if that person is unavailable.

Next steps

Review this report with key stakeholders (President, VP Operations, Controller, Sales, Customer Service) to align on priorities and strategic direction.
Form a small, cross-functional working group to define detailed requirements and success metrics for the first pilot projects.
Initiate the development of a formal data classification and external data sharing policy to guide future technology decisions.
Engage with the ERP vendor to understand the costs and feasibility of API integrations required for prioritized opportunities.

Where the evidence was thin

Recorded by the model as gaps rather than filled in.

The exact monetary cost of manual order entry errors, including credits, redeliveries, and staff time, is not quantified.
The specific volume of each customer service request type (e.g., 'Where is my order', 'Invoice copy') is not measured, only inferred by supervisors.
The total time spent by staff on re-keying information across different systems is not measured.
Key operational metrics such as picking accuracy, on-time delivery rate, and cost per delivery drop are not currently measured.

How this result was produced

Safe run metadata from the saved generation. Digests, prompts, and provider payloads are not shown.

Provider: Gemini
Model: gemini-2.5-flash
Workflow: business-workflow-v1
Prompt: business-report-v4
Schema: business-report-v1
Retrieval: business-retrieval-v1
Total time: 38.6 s
Generation: 36.2 s
Retrieval time: 1.9 s
Input tokens: 9907
Output tokens: 4026
Thinking tokens: 3749
Total tokens: 17682
Evidence retrieved: 24
Evidence cited: 13

Checks

Citation validationPassed
Evidence groundingPassed
Numeric claimsPassed
No unverified numeric claims detected
Human oversightPassed
Recommendation mixRecorded
AI 3 · Deterministic Automation 2 · Process Improvement 1

Want to try the Workbench?

Access is invite-only while the product is in alpha. Request access if you want to run these workflows on your own material.

Request access How it works