Executive Summary
Kelsan operates in a margin-thin distribution category where profitability depends on the precision
of supplier agreements, contract compliance, and sales velocity — all of which are currently managed
manually at scale. Based on conversations with Ken, we have identified four specific areas
where profit is leaking, each with confirmed precedent: a $100K+ rebate gap Ken found in a
prior analysis, and a $36K+ overcharge on a single SKU — examples Ken identified through his own internal reviews, not outliers.
This plan describes a multi-agent AI sweep that processes Kelsan's ERP exports through
a structured pipeline: data is cleaned, gaps are identified, findings are ranked by dollar value, and
actionable outputs are generated — including drafted vendor emails with attached supporting
data and ranked audit CSVs ready for human review and follow-up. The model
is designed to get Ken from raw ERP data to a prioritized recovery action list within 30 days,
without requiring any system changes to begin.
Pain Points Addressed
4
Each with its own agent sweep
Recovery Potential
$150K+
Conservative first-pass estimate
Outputs Per Sweep
3 types
Ranked CSV · Vendor emails · Audit report
Time to First Deliverable
30 days
From ERP data receipt
The Execution Model
How the Multi-Agent Sweep Works
Each pain point runs through the same core pipeline: raw ERP data goes in, and ranked findings plus
ready-to-send outputs come out. Python and pandas handle the heavy data lifting — no ERP customization,
no IT project, no new software licenses. The preferred setup connects directly to Kelsan's ERP via read-only SQL — flat-file exports are the fallback where direct access isn't available.
Step 1 · Two Options
ERP Data Access
PREFERRED: Direct SQL
Read-only database connection to the ERP. Agent queries on demand — no manual exports, no size limits, schema-mapped for NL queries.
FALLBACK: Flat File
Standard CSV/Excel exports from ERP reports. Simpler to start, no IT config required.
→
Step 2 · Agent
Data Agent
Cleans and structures the raw data. Normalizes supplier names, SKU codes, dates. Builds the master crosswalk between suppliers, products, and contracts. With SQL access, re-queries the ERP directly when it needs more data.
→
Step 3 · Agent
Analysis Agent
Runs the gap analysis for each pain point. Compares invoiced amounts vs. contracted, volume vs. rebate tiers, GPO billing vs. contracted rates. Pandas under the hood.
→
Step 4 · Agent
Audit Agent
Scores confidence on each finding. Flags items for human spot-check before action. Separates definitive variances from items needing Ken's team to verify against source documents.
→
Step 5 · Agent
Output Agent
Generates ranked recovery CSV, drafts vendor dispute emails (one per supplier), formats audit summary. All outputs are human-ready — Ken's team reviews, edits, and sends.
→
Step 6 · Human
Review & Act
Ken's team reviews the ranked list, spot-checks flagged items, edits draft emails as needed, and sends to the appropriate supplier contacts. No black box — every finding is explainable.
Why not just a single AI prompt? The data volume is too large — thousands of PO lines, invoice records, and contract terms. Breaking the work across specialized agents means each one does one thing well: the Data Agent doesn't do analysis, the Analysis Agent doesn't write emails. This produces cleaner, more auditable outputs and makes it easy to re-run any single stage when new data arrives.
Preferred Data Access Approach
Direct Read-Only SQL + Agentic RAG
Rather than waiting for manual exports, the ideal setup connects the agent directly to Kelsan's ERP database via a read-only SQL credential. This unlocks on-demand data access, eliminates the manual pull step, handles datasets of any size, and eventually enables Ken or his team to ask plain-English questions and get answers from their own data in seconds.
Step A · Schema Mapping
One-time setup: the agent is given read-only access to the ERP database and maps the schema — every table, column, and relationship. For a distributor ERP like Epicor P21, this means learning tables like po_hdr, po_line, inv_mast, vendor_mast, customer. The schema map becomes the agent's knowledge base — it knows what data exists and how to join it.
Step B · Agentic RAG Queries
The agent uses its schema knowledge to retrieve exactly what it needs, when it needs it. Instead of loading everything upfront, it pulls targeted data on demand — "give me all PO lines for suppliers with active rebate contracts in Q4 2025." The RAG layer means it can answer follow-up questions by re-querying rather than re-exporting. Data stays in the ERP; the agent reaches in.
Step C · Natural Language Queries
Once the schema is mapped and validated, Ken can ask questions directly in plain English:
"What's our total spend on cleaning chemicals by supplier last 6 months?"
"Which customers haven't reordered in 60 days?"
"Show me all invoices from Supplier X above contracted price."
The agent translates to SQL, runs it, and returns the answer — no analyst, no report request, no wait.
What IT needs to do
Create a read-only database user on the ERP SQL instance. One task, no system changes, no risk to production data.
Implementation path
MCP database connector (read-only) → schema mapping session → validated test queries → production access for analysis runs.
If direct access isn't available
Fall back to flat-file exports for Phase 1. The schema mapping still happens — it just uses the export structure. Direct access can be added in Phase 2.
Opportunity Areas
Four Sweeps, Four Sets of Deliverables
Each sweep targets a specific profit leak. Each one produces its own ranked findings, draft vendor communications, and an audit summary.
How the agents work it
Data Agent
Volume Aggregator
Sums purchase volume by supplier, SKU, and time period. Matches to rebate contract periods.
→
Analysis Agent
Tier Mapper
Applies rebate tier thresholds from each contract. Calculates what should have been earned vs. what was received.
→
Output Agent
Dispute Generator
Writes a supplier-specific dispute email per gap found. Attaches the supporting volume summary as a CSV.
Outputs
📊 Ranked rebate gap CSV (by supplier, by dollar)
✉ Draft dispute email per supplier
📎 Volume summary attachment per email
🔍 Spot-check list (top 10 for Ken to verify)
Forward-looking add-on
- Model shows when each supplier will hit their next rebate tier based on current purchase trajectory
- Enables proactive purchasing decisions to capture tier bumps before period close
How the agents work it
Data Agent
Invoice Matcher
Joins AP invoice lines to PO records. Cross-references contracted prices from the supplier price list.
→
Analysis Agent
Variance Ranker
Calculates variance per SKU per invoice. Aggregates by supplier. Separates recurring overcharges from one-time billing errors.
→
Output Agent
Credit Memo Drafter
Generates a credit request email per supplier with total variance, list of affected SKUs, and the supporting invoice line detail.
Outputs
📊 Overcharge rankings CSV (supplier · SKU · variance · total $)
✉ Credit request email per supplier
📎 Invoice line detail attachment per email
🔍 Confidence score per finding (high / medium / flag)
System feedback loop
- Confirmed overcharges feed back into the ERP price list as corrections
- Future POs auto-flag when invoiced cost deviates from the now-corrected contracted price
- This closes the loop — prevention, not just recovery
How the agents work it
Data Agent
Member Mapper
Maps each customer to their active GPO contract. Pulls contracted price per SKU for each member.
→
Analysis Agent
Billing Auditor
Compares every GPO customer invoice line against the contracted rate. Flags over and under charges separately.
→
Output Agent
Compliance Reporter
Produces audit summary by GPO contract, customer correction notices, and an ERP price-update recommendation file.
Outputs
📊 Compliance summary CSV (by GPO · customer · variance)
✉ Customer correction notice (for overcharges)
📎 ERP price-update file (import-ready for pricing team)
🔍 Contract expiry flag list (renewal outreach queue)
Long-term fix
- Output includes an import-ready price correction file for the ERP — fixes the root cause, not just the symptom
- Contract expiry flags trigger proactive renewal outreach before pricing lapses
How the agents work it
Data Agent
Sales Structurer
Pulls weekly sales by customer, category, and rep. Compares to prior 4-week baseline. Flags trend changes.
→
Analysis Agent
Account Scorer
Scores each account: volume trend, category mix change, spend drop, upsell signal. Ranks by urgency and opportunity size.
→
Output Agent
Brief Writer
Writes a one-page rep briefing: top accounts to call, why, what to say, what's at stake. Delivered by email or as a printable PDF.
Outputs
✉ Weekly briefing email per rep (Monday AM)
📊 Account health CSV (all customers, all reps)
🔍 Anomaly flags (unusual drops or spikes for manager review)
What a rep receives
- Top 5 accounts to call this week with a one-line reason for each
- Any account where spend dropped >20% vs. prior month — with prior spend shown
- Category gaps: accounts buying janitorial supplies but not paper, or paper but not equipment
- New accounts from the prior 30 days — confirm they're getting good service early
Audit & Spot Check Layer
Built-In Confidence Before Anyone Sends an Email
Every sweep includes a structured audit pass before outputs reach Ken's team. No vendor email goes out on a finding the analysis isn't confident about. The goal is to make the outputs trustworthy enough that the only work left is review and send — not investigation.
🎯
Confidence Scoring
Every finding gets a confidence score: High (data clearly matches or mismatches), Medium (likely correct but worth a second look), or Flag (ambiguous — needs Ken's team to pull the source document). Only High and Medium items move forward to email generation.
🔍
Calibration Meeting <90 min
A dedicated meeting — separate from the discovery session — where Keller Creative presents a sample subset of findings and Kelsan's internal experts manually check them against their own records. The purpose is calibration, not approval: confirming the model is reading the ERP data correctly before the full output batch is generated. Any sweep that fails the sample check gets recalibrated and re-run. No vendor emails are generated until this meeting gives the green light.
📋
Audit Trail Output
Every finding includes the source data that generated it: which invoice, which PO, which contract clause. If a supplier disputes the finding, Ken's team has the exact line-item evidence in the attachment. The audit trail also makes it easy to re-run with updated data and track what's been resolved vs. what's still open.
What the Outputs Actually Look Like
Vendor Emails & Data Files
These are the concrete deliverables Ken's team receives from each sweep. Draft emails are generated per supplier with accurate data populated — they review, adjust tone if needed, and send. CSVs are clean and ready for action.
To:
[Supplier Rep Name] — [Supplier Accounts Payable]
Subject:
Rebate Reconciliation — Q4 2025 — [Supplier Name]
Attachment:
Kelsan_Q4_2025_[Supplier]_Volume_Summary.csv
"Hi [Name], I'm following up on our Q4 2025 rebate statement. Per our review of purchase
volume against the terms of our agreement, we believe the correct rebate amount for this
period is $[X], based on total purchases of $[Y] across [Z] SKUs in the [category] tier.
The attached CSV shows our volume summary by SKU and the applicable tier from our contract.
Our records show we received $[received] for this period — a difference of $[gap]. Could
you confirm your calculations or let us know if there's a discrepancy in your records?
Happy to set up a call to walk through the data if helpful."
Generated per supplier · Data auto-populated · Tone editable
To:
[Supplier AR Contact]
Subject:
Pricing Discrepancy — Invoice #[XXXX] — [Supplier Name]
Attachment:
Kelsan_[Supplier]_Overcharge_Detail.csv
"Hi [Name], during a review of recent invoices, we identified a pricing discrepancy on
several SKUs billed above our negotiated rates. The attached detail shows [N] SKUs
across [date range] where invoiced unit costs exceeded contracted pricing.
Total variance identified: $[amount]. The largest item is SKU [X], invoiced at $[billed]
vs. our contracted rate of $[contracted] — a difference of $[diff] per unit across
[qty] units.
Please issue a credit memo for $[total] or let us know if you see the data differently.
We'd like to resolve this before the next invoice cycle."
One email per supplier · All affected SKUs in attachment
Columns in the output file Ken's team receives:
- Supplier Name
- SKU / Product Code
- Product Description
- Contracted Unit Price
- Invoiced Unit Price
- Variance Per Unit ($)
- Total Units Affected
- Total Dollar Variance
- Invoice Number(s)
- Invoice Date(s)
- Confidence Score
- Spot-Check Required?
- Status (Open / Sent / Resolved)
Sorted by total dollar variance · Human-editable status column
Subject:
Your Weekly Account Briefing — Week of [Date]
"Hi [Rep Name] — here's your account summary for this week.
Call These First:
· Acme Facilities — spend down 34% vs. last month ($4,200 → $2,780). Last order was [date].
· Metro Building Services — dropped paper products entirely in last 3 orders. Worth a check-in.
· Lakeview Janitorial — new account, 30 days in. Good time to confirm they have what they need.
Upsell Signals:
· Citywide Cleaning — buying floor care but not dispensers. 8 similar accounts use both.
Full account data attached."
Auto-generated weekly · One per rep · Delivered Monday 7am
Consultative Discovery
What We Need to Learn Before We Can Build
The analysis design depends on where Kelsan's data actually lives. These are the real-world questions — not theoretical, but specific to how a distributor at this scale actually operates. Expect the first session to be half data-mapping, half strategy.
Rebates + Overcharges
- Where are active rebate contracts stored — email, shared drive, a binder?
- Who owns supplier relationships for rebate claims — AP, purchasing, a specific person?
- When a rebate is paid, is there a remittance statement or just a wire?
- Are negotiated supplier prices in the ERP, or in separate spreadsheets?
- When an invoice arrives, does the system auto-match to a PO with a contract price?
- Is there a time limit on clawback — how far back can you dispute?
- Which 5–10 suppliers account for the largest purchase volume?
GPO + Sales Scorecard
- Which GPOs is Kelsan in? How many member customers per GPO?
- Where do GPO price lists live — ERP price levels, spreadsheets, a portal?
- Is GPO pricing applied automatically at order entry, or does a rep select it?
- What ERP system does Kelsan run? (P21, SAP, Eclipse, Infor, other)
- What do reps use today to decide which customers to call this week?
- Is sales data by rep accessible as an ERP report, or does it need IT to pull?
- How do reps prefer to receive information — email, dashboard, printed?
The messy reality: Rebate contracts are probably in someone's email. Negotiated prices are partly in the ERP and partly in a spreadsheet someone maintains. GPO pricing was set up years ago and nobody's sure if it's current. This is normal and expected — the discovery session maps the chaos before the analysis begins. We've seen this at every distributor we've worked with at this scale.
Data Requirements
What We Need From Kelsan's ERP
These are the data sources the agent needs to operate. Direct ERP SQL access is ideal; flat-file exports are the fallback. Either way, no custom development on Kelsan's side.
| Export | Contents | Format | Used For | Priority |
|---|
| Purchase Order History | PO number, date, supplier ID, SKU, qty, unit cost, total, invoice # | CSV / Excel | Rebate reconciliation, overcharge detection | Critical |
| AP Invoice Detail | Invoice date, supplier, line items, unit price billed vs. PO price | CSV / Excel | Overcharge detection | Critical |
| Customer Sales History | Invoice date, customer ID, SKU, qty, unit price, rep ID, order total | CSV / Excel | GPO audit, sales scorecard | Critical |
| Supplier Price List (contracted) | Supplier ID, SKU, negotiated unit cost, effective date, expiry date | CSV / Excel / PDF | Overcharge detection | Critical |
| Rebate Contracts | Supplier, product categories, volume tiers, rebate %, effective period | PDF / Email / Excel | Rebate reconciliation | Critical |
| GPO Member Roster + Price Lists | Customer IDs mapped to GPO, contracted prices per SKU, contract dates | Excel / PDF | GPO audit | High |
| Rep / Territory Map | Customer ID, assigned rep, territory or region | CSV / Excel | Sales intelligence scorecard | High |
| SKU / Product Master | SKU, description, category, manufacturer, unit of measure | CSV / Excel | All analyses — joins and grouping | Helpful |
Engagement Plan
First 30 Days
Structured around getting to the first concrete findings as fast as possible. Two formal meetings bracket the analysis work — discovery to map the data, calibration to validate the findings before any action is taken.
Day 1–3
Discovery
Meeting
<90 min
Stakeholders map data sources, confirm ERP, set timeline
Day 4–7
Data
Collection
ERP exports, rebate contracts, GPO price lists received
Day 8–14
Agent
Sweeps
4 agents run in parallel — findings ranked, confidence scored
Day 15–21
Calibration
Meeting
<90 min
Sample findings checked by internal experts — green light given
Day 22–30
Outputs
+ Action
Ranked CSV delivered, vendor emails reviewed and sent
Meeting 1 — Discovery <90 min
Maps where data lives, who owns it, and what format it comes in. The output of this meeting is a data request list — not an analysis. Brings in purchasing lead, AP contact, and a sales rep.
Meeting 2 — Calibration <90 min
Presents a 15–25 item sample of findings. Kelsan's internal experts check each one against their own records. Gate: if the sample passes, full output is generated and vendor emails go out. If not, sweeps are recalibrated and re-run first.
Day 1–3
Stakeholder Discovery
- Meet Ken + purchasing lead, AP lead, top rep
- Walk all 4 sweeps with discovery questions
- Map where each data source actually lives
- Identify who pulls ERP exports and in what format
- Confirm ERP system and data accessibility
- Agree on data delivery timeline
Day 4–7
Data Collection
- Receive ERP exports: PO history, invoices, sales
- Collect rebate contract documents
- Collect supplier price agreements
- Collect GPO price lists and member roster
- Initial data quality check — flag gaps
- Build master SKU/supplier/customer crosswalk
Day 8–14
Agent Sweeps Run
- Data Agent cleans and structures all exports
- Analysis Agent runs rebate reconciliation
- Analysis Agent runs overcharge detection
- Analysis Agent runs GPO compliance check
- Audit Agent scores confidence on all findings
- Spot-check list generated — top 10 per sweep
Day 15–21
Calibration Meeting <90 min
- Present sample subset of findings to Ken + internal experts
- Kelsan team manually checks samples against their own records
- Confirm the model is reading the data correctly
- Identify any ERP quirks that skewed results
- Adjust confidence thresholds based on expert feedback
- Re-run any sweep segments that need recalibration
- Green-light given before full output is generated
Day 22–30
Delivery + Ongoing
- Ken's team sends validated vendor emails
- Activate weekly sales scorecard (automated)
- Set up monthly rebate + overcharge re-run
- Train team on reviewing and using outputs
- Define Phase 2: ERP price correction imports
- Track resolution status in the recovery CSV
Expected Outcomes
What Kelsan Should Expect to Find
Conservative estimates anchored to Ken's confirmed prior findings, applied at catalog scale. Each number is independently verifiable before any vendor contact is made.
$100K+
Rebate Underpayments
Based on Ken's confirmed prior finding. Multi-supplier analysis typically surfaces significantly more.
$36K+
Per-SKU Overcharge Found
Ken's own finding on a single SKU. Catalog-wide scan expected to surface multiple similar items.
1,000s
GPO Lines Audited
Currently a manual process. AI runs the full compliance audit in minutes, not weeks.
Weekly
Automated Rep Briefings
Every rep gets a prioritized account list Monday morning — no manual prep required.
Future State
Where the Sweep Data Takes You Next
The sweep is the foundation. Once Kelsan has clean, validated cost and contract data — something most
distributors have never had — a second layer of intelligence becomes possible: tools that use that
data in real time to guide quoting decisions, score opportunities, and ensure every proposal goes out
with accurate margin understanding baked in. This section describes what that looks like after Phase 1.
Why the sweep has to come first: Everything in this future state depends on knowing Kelsan's true cost structure — validated rebate rates, confirmed contracted prices, corrected ERP data. Before the sweep, that foundation doesn't exist. After it does, these tools become reliable. Build on bad data and the margin estimates in a quote are just as wrong as they are today.
Phase 2 · Highest Value
Margin-Aware Quoting Engine
When a rep builds a quote or proposal, the system automatically calculates true margin using
current contracted costs, live rebate tier position, and historical deal structures for similar
customers and quantities. The rep sees an estimated margin — not a rough guess — before the quote goes out.
What it knows at quote time
- Current contracted cost from the validated ERP price list (from the overcharge sweep)
- Which rebate tier Kelsan is currently in with each supplier — and if this deal pushes them to the next one
- Historical margin on similar deals: same SKU category, similar quantity, similar customer profile
- If the customer is a GPO member — their contracted price is pre-loaded, no manual lookup
- Close likelihood based on historical win/loss for this deal type and customer segment
From the rep's perspective
"Quote for 500 units of SKU #4821 to Metro Facilities — estimated margin: 18.4%.
If you increase the order to 600 units, Kelsan hits the next rebate tier with Supplier A — margin rises to
21.2%.
Historical close rate for similar deals: 67%."
The complexity is invisible. The rep just sees the number.
Phase 2 · Prospecting Intelligence
Opportunity Scoring + Prospect Intelligence
Connects Kelsan's historical win/loss data with prospect intelligence from a tool like Apollo or
a similar B2B data connector. The system scores inbound leads and existing accounts by estimated
profitability — not just revenue potential, but margin potential — so the team focuses
on the deals most likely to close at the best margin.
The intelligence layer
- Apollo (or similar) provides company size, industry, estimated spend profile for new prospects
- Kelsan's own historical data shows which company profiles convert and at what margin
- Combined: a score for any new prospect — "this type of account historically generates X% margin and closes in Y days"
- Existing customers get scored for expansion opportunity: who's likely to grow spend, who's at churn risk
- Sales team gets a ranked list: work these accounts first, this week, for highest expected return
Integration approach
MCP connector to Apollo
Historical deal data from ERP sweep
Scored prospect briefing per rep
Phase 3 · System Integration
Live Cost & Rebate Backend
Right now, rebate rates and contracted prices get stale the moment a contract renews or a supplier
updates their pricing. This layer keeps the quoting engine's cost data current — automatically.
When a contract is updated, the backend propagates the change to every place that uses that cost:
open quotes, the margin model, the rebate tracker.
How it stays current
- New supplier contracts or updated price lists are ingested automatically (email attachment, portal export, or ERP sync)
- Rebate tier positions are recalculated monthly as new PO data arrives — no manual update required
- If a contract expires without renewal, the system flags it before quotes go out at the wrong margin
- Historical bulk deal returns are tracked over time — the model knows whether "500-unit deals in this category" have been profitable over the past 2 years
- Margin estimates on proposals come with a confidence band, not a single number — "14–18% based on historical variance at this volume"
What this replaces
The spreadsheet someone manually updates when a supplier calls. The margin estimate a rep makes up because they don't know the current rebate position. The quote that goes out at 12% margin because the cost data was 6 months old.
Phase 3 · Strategic
AI Contracting Generation
Once the cost and historical data is clean and current, Kelsan can generate customer-facing proposals
that are grounded in real numbers — not gut feel. The system drafts proposals that include the right
pricing for the customer's profile, recommended bundle structures based on what similar customers buy,
and a projected margin range for Ken's team before anything goes out.
What the proposal generation skill does
- Pulls customer profile: industry, size, current product mix, historical spend, GPO membership
- Recommends a product bundle based on what accounts with similar profiles buy most profitably
- Prices the bundle using live cost data — every line item has a true margin baked in
- Forecasts close likelihood and expected margin range based on historical comparables
- Drafts the proposal document — formatted, professional, ready for Ken's team to review and send
The end state
Ken's team says "build a proposal for Riverside School District." The system pulls their profile,
recommends a cleaning supply bundle based on similar K-12 accounts, prices it at current cost with
live rebate rates applied, shows an estimated margin of
16–19%, flags a
72% historical close rate for
this account type, and outputs a formatted proposal PDF. Done in under a minute.
How the Phases Build on Each Other
Phase 1 — Now
The Sweep
Find the money. Validate the data. Recover the first round of leakage. Establish the clean data foundation everything else runs on.
→
Phase 2 — Next 90 Days
Margin Intelligence
Quoting engine with live cost data. Opportunity scoring connected to prospect intelligence. Reps make better decisions on every deal they touch.
→
Phase 3 — 6 Months+
Automated Intelligence
Live cost + rebate backend stays current automatically. Proposal generation drafts deals in under a minute. The system compounds over time as more deal data accumulates.
Infrastructure Note
Secondary analytics database with stricter schema hierarchy. ERP stays system of record. Agents flag bad data at ingestion — missing prices, unlinked invoices, incomplete contract periods — before they corrupt downstream analysis.
→
Phase 4 — Long Term
Compound Advantage
Every deal Kelsan wins or loses teaches the model. Forecasting accuracy improves. Margin protection becomes structural — not a project, but how the business operates.
Next Steps
How We Start
Two meetings, one data delivery, thirty days to first findings.
-
1
Schedule 90-Minute Discovery Session with Ken + Stakeholders
Bring purchasing lead and AP contact. We map data sources, confirm ERP system, and set a realistic data delivery timeline. This conversation determines whether we start with rebates, overcharges, or both simultaneously.
-
2
Confirm Scope and Which Sweeps Run in Phase 1
Recommend starting with rebate reconciliation and overcharge detection — both use the same PO and invoice data, so they share the data collection effort. GPO and sales scorecard can follow in Phase 2.
-
3
Kelsan Pulls ERP Exports Using Our Data Request Template
We provide a simple one-page data request document — no IT project required. These are standard ERP reports that any purchasing or AP admin can run. We work with whatever format the ERP produces.
-
4
Agent Sweeps Run — First Findings Returned in 10 Business Days
Deliverable: ranked recovery CSV, confidence-scored spot-check list, and a sample subset packaged for the calibration meeting. No vendor emails generated yet — calibration comes first.
-
5
Calibration Meeting — Sample Review with Internal Experts <90 min
Keller Creative presents a representative sample of findings — ideally 15–25 items across each sweep. Ken's team (purchasing, AP, or category experts) manually checks each one against their own records. This session confirms the analysis is reading the data correctly, surfaces any ERP-specific quirks that need adjustment, and gives the green light before the full output batch is generated. Nothing goes to a vendor before this meeting happens.
-
6
Full Output Generated + First Vendor Emails Sent
With calibration confirmed, the Output Agent generates the complete ranked CSV and all vendor draft emails. Ken's team reviews, adjusts tone where needed, and sends. This is where the money recovery actually begins.
-
7
Establish Monthly Cadence + Activate Weekly Sales Scorecard
Once the first sweep is validated, set up monthly re-runs with fresh ERP exports. The sales scorecard activates as soon as the rep/customer data is confirmed. Phase 2 introduces ERP price correction imports to close the loop on future overcharges.