Maple Spark Labs
Method and templates / Worked example: phone, email and paper orders into the system
v1.1Updated 2026-10-04

Worked example: phone, email and paper orders into the system

See the work products from a fictional three-week Sprint about orders that arrive by phone, email and paper, from the first walkthrough to the final go or no-go decision.

Everything on this page is fictional. Pinecrest Supply, its people, its numbers and its customers are made up to show what a Sprint looks like when the workflow is orders arriving by phone, email and paper. No client work is described here.
Download the full PDFHave it done with you

What you leave with

  • 01Fictional · Pinecrest Supply
    1. F01Keying orders from five channels21 h/wk
      Measured
    2. F02Calling back to confirm items5.5 h/wk
      Reported
    3. F03Looking up contract pricing6 h/wk
      Reported
    4. F04Wrong picks from keying errors3.75 h/wk
      Measured
    5. F05Lunch-hour phone orders waitingnot hours
      Observed

    Friction map. A ranked view of where time is being lost, and what evidence supports each number.

    View full artifact
  • 02Fictional · Pinecrest Supply
    1. Email and paper orders36 / 45
      Selected · AI capture with a rep approving
    2. Phone orders27 / 45
    3. Contract pricing31 / 45
    4. Wrong picks29 / 45

    Phone and pricing went to a script, routing and the system's own pricing tables; AI only on email and paper.

    Use-case decision scorecard. The options considered, how they scored, and where process change and automation beat AI.

    View full artifact
  • 03Fictional · Pinecrest Supply
    Order review · sample data
    Original email · order sheet
    Extracted order
    CustomerSample customer 22
    Ship-toOn file
    RequestedThursday
    Lines4
    Line 312 × hand towels
    Pack size: 12 or a 12-pack? Held for a rep.
    Approve and postCorrectHold

    Testable prototype. A realistic review screen for email and paper orders, tested with the reps.

    View full artifact
  • 04Fictional · Pinecrest Supply
    Go · Fund a 6-week pilot on email orders and typed or printed order sheets; handwritten sheets stay with a rep.
    Recommendation
    Process change and simpler automation first (form, pricing tables, confirmation step, phone script). AI capture with human approval on email and paper orders only.
    Controls
    No order posts without a rep approving it. Unknown customers, new ship-to addresses, unmatched products and unusual quantities are always held. Substitutions are suggested, never applied. Customer data stays in Pinecrest's systems. Running cost: cents per order.
    Exit criteria
    Pilot scope: email orders and typed or printed order sheets. Handwritten sheets stay with a rep while those customers move to the form. 96% line accuracy on a 100-order evaluation set. Median rep time under 2 minutes per email order for four weeks. Wrong picks from keying errors under 2 a week. Stop if an unapproved order reaches the warehouse.
    Success measures
    Minutes per order (baseline 7). Share of orders approved without edits. Wrong picks per week (baseline 9). Orders in the system within the hour (baseline not yet measured; capture it in week 1 of the pilot).
    Next decision
    Fund a 6-week pilot on email orders and typed or printed order sheets; handwritten sheets stay with a rep. Owner: operations manager. Revisit phone orders after the pilot, with the baseline the new screen will have produced.

    Opportunity Map and pilot criteria. The recommendation, controls, measures and the exit criteria a pilot must meet.

    View full artifact

The three weeks

  1. Week 1: What we saw

    21 ha week keying orders from five channels, measured
    • Orders arrive by phone, email, paper, text and a web portal almost nobody uses.
    • The reps were fast and careful; the hours were lost to the system around them.
    • Wrong picks from keying errors, measured from credit memos, added rework.

  2. Week 2: Options and decision

    36 / 45AI capture for email and paper orders, with a rep approving
    • Phone orders went to a structured screen and call routing, not AI.
    • Pricing rules moved into the order system's own pricing tables.
    • Confirmation before picking: a free process change, done during the Sprint.

  3. Week 3: Tested and learned

    • Approve a clean email order
      4 of 4
    • Catch a pack-size mix-up (12 vs 12-pack)
      3 of 4 · 1 helped
    • Handle a handwritten sheet with a crossed-out line
      2 of 4 · 2 helped
    • The prototype passed 26 of 30 evaluation cases. The four misses (three handwritten sheets, one two-order PDF) were all held for a rep.
    • The reps wanted flagged fields first, and substitutions only ever suggested.
    • The warehouse asked for one queue, whatever channel the order came from.
The decision

Go: Fund a 6-week pilot on email orders and typed or printed order sheets; handwritten sheets stay with a rep.Stop if an unapproved order reaches the warehouse.

What this example shows

  • Most of the hours were in the system around the reps, not in the reps. Fixing that came first and cost nothing.
  • AI earned its place on one narrow step, email and paper capture, where the format varied too much for rules and a person approved every result.
  • The Sprint left a baseline, a simpler process and one queue someone owns, so the next turn of the loop is cheaper than the first.

The underlying tables

View the underlying table: The setup
Organization Pinecrest Supply (fictional): a 40-person regional distributor of facility and janitorial supplies
Sponsor Operations manager
People interviewed Three customer service reps, the warehouse lead, one outside sales rep
AI tool option B: de-identified notes only. No customer or order data in any AI tool during the Sprint
Long-term goal "An order is in the system within the hour it arrives, entered once, and the reps spend their day on customers instead of re-typing."

Sprint questions:

  1. Can orders that arrive by phone, email and paper get into the order system without being re-typed, and without wrong quantities or wrong ship-to addresses getting through?
  2. How much of the problem is the channel mix, and how much is the system around the reps?
  3. Which part, if any, is an AI problem?
View the underlying table: Week 1

Orders arrive five ways: phone (about 45%), email with the order in the body or an attached PDF (35%), handwritten faxes and paper order sheets from a few long-standing customers (10%), text messages from outside sales reps (7%), and the web portal (3%, built two years ago, barely used).

ID Friction Type Min each Per week Hours per week Evidence
F01 Keying each order into the order system from a phone note, email or paper sheet: customer, ship-to, lines, quantities, requested date Re-key 7 180 21 Measured over one week
F02 Calling or emailing customers back to confirm unclear items, quantities or pack sizes Rework, wait 6 55 5.5 Reported
F03 Looking up contract pricing and substitutions in a spreadsheet kept by one senior rep Search, one-person knowledge 3 120 6 Reported
F04 Warehouse picks a wrong item or quantity because the keyed order was wrong Rework 25 9 3.75 Measured from credit memos
F05 Orders taken by phone during lunch sit on a notepad until the rep is back Wait n/a daily n/a Observed; delay, not hours

What stood out: the reps were fast and careful. The hours were lost to the system around them, not to them. Three channels fed one rep's notepad, one spreadsheet held the pricing rules, and the portal that could have taken half the volume asked customers for twelve fields they didn't know.

View the underlying table: Week 2
ID AI option Simpler automation Process change Weighted score (of 45)
F01 (email and paper) Read email bodies, attached PDFs and scanned order sheets into a review queue; a rep approves each order before it posts A one-page order form, emailed to the top 30 customers, that imports directly Ask the top 30 customers (65% of volume) to use the form or the portal 36 (AI capture into a review queue, with the form and portal ask alongside)
F01 (phone) Transcribe calls and draft the order for the rep to confirm A structured phone script and screen so the rep keys while talking Route lunch-hour calls to a second rep 27 (deferred: the script and routing fix most of it first)
F03 Suggest contract price and substitutions from the pricing rules Move the spreadsheet rules into the order system's pricing tables Document the rules 31 (simpler automation wins outright)
F04 Order confirmation sent to the customer before picking Warehouse picks from the confirmed order only 29 (process change, free, do now)

Decision (sponsor): prototype F01 for email and paper orders only. Phone orders stay with a person, using a new structured screen, and land in the same queue so the warehouse sees one list. Why AI lost on phone and pricing: the phone problem was mostly routing and a screen, and the pricing rules belonged in the system's own pricing tables, where they would be checked by the system rather than guessed by a model. Both changes go ahead during the Sprint; they are free.

View the underlying table: Week 3

Prototype: email orders and scanned order sheets are read by a model with a fixed schema (customer, ship-to, requested date, lines with product, pack size and quantity) and land in a review screen showing the extracted order beside the original. A rep corrects or approves. Anything with an unknown customer, a product that doesn't match the catalogue, an unusual quantity or a new ship-to address is flagged. Sample and synthetic orders only.

Tests with four people:

Done Done with help Not done
Approve a clean email order 4 0 0
Catch a pack-size mix-up (12 vs 12-pack) 3 1 0
Handle a handwritten sheet with a crossed-out line 2 2 0

Patterns: all three reps wanted the flagged fields to be the first thing they see, not a list at the bottom. The senior rep asked that substitutions never be applied automatically, only suggested. The warehouse lead asked for one queue, whatever channel the order came from.

Starter evaluation set: 30 synthetic orders agreed by the senior rep: 14 typical emails, 8 hard (PDF attachments with two orders in one, handwritten sheets, quantities in cases and units mixed), 4 that must be held for a person (unknown customer, new ship-to, discontinued product, quantity ten times the usual), 4 shaped like past credit memos. The prototype passed 26 of 30. The four misses were three handwritten order sheets and one PDF that carried two orders and was read as one. All four were held in the review queue. None would have posted, because nothing posts without a rep's approval. The handwritten sheets set the first improvement and told the sponsor to push those customers toward the form first.

View the underlying table: The Opportunity Map, in brief
Section Pinecrest Supply
01 The work Three reps enter about 180 orders a week from five channels. Done means the right items, quantities and ship-to are in the system the hour the order arrives.
02 Where it hurts 7 minutes each, 180 a week: 21 hours a week keying, plus 9 wrong picks a week. Measured.
03 Future workflow Email and paper orders are captured into a review queue; a rep approves each one. Phone orders are keyed on a structured screen into the same queue. Confirmation goes to the customer before picking.
04 Recommendation Process change and simpler automation first (form, pricing tables, confirmation step, phone script). AI capture with human approval on email and paper orders only.
05 What it needs A shared order inbox the system can read. The order system's import or API. Pricing rules moved into its pricing tables. Owner: operations manager.
06 Controls No order posts without a rep approving it. Unknown customers, new ship-to addresses, unmatched products and unusual quantities are always held. Substitutions are suggested, never applied. Customer data stays in Pinecrest's systems. Running cost: cents per order.
07 Exit criteria Pilot scope: email orders and typed or printed order sheets. Handwritten sheets stay with a rep while those customers move to the form. 96% line accuracy on a 100-order evaluation set. Median rep time under 2 minutes per email order for four weeks. Wrong picks from keying errors under 2 a week. Stop if an unapproved order reaches the warehouse.
08 Success measures Minutes per order (baseline 7). Share of orders approved without edits. Wrong picks per week (baseline 9). Orders in the system within the hour (baseline not yet measured; capture it in week 1 of the pilot).
09 Next decision Fund a 6-week pilot on email orders and typed or printed order sheets; handwritten sheets stay with a rep. Owner: operations manager. Revisit phone orders after the pilot, with the baseline the new screen will have produced.
View the underlying table: The loop after the Sprint

The Sprint was one turn. The Map set up the next ones.

Turn What changes AI involved?
1 (during the Sprint) Order confirmation before picking. Pricing rules into the system. Phone script and screen. Form sent to the top 30 customers. No
2 (pilot) Email and paper orders captured into the review queue. Yes, with a rep approving every order
3 (later, if the numbers say so) Phone orders: with a quarter of measured phone-order data, decide whether transcription into the same queue is worth it. Decided then, not now
Each turn Measure the same four numbers. Simplify before automating. Keep the rep in the design.

Everything, in one document: the full PDF

Use and reuse. © 2026 Maple Spark Labs Inc. You may use and adapt this material within your organization. Keep this notice. Please ask before republishing, redistributing or selling it.

Want this run with you, all the way to production?

Back to Method and templates
Talk through a workflow