Everything in a distributor, in four parts
Shawn’s three groups of functions, with the foundation under all three. Open any function to see its steps, how far AI can take each one, where to start and what’s been said about it.
Every opportunity, by part of the business
Rows are the four parts. Pick the columns: Shawn’s two-by-two, when each one starts at the investment on the dashboard, or how we’d get it. The square on each opportunity switches it in or out of the plan, and everything else follows.
The dashboard
The build order comes first, then the calendar. Drag the investment, tap a dollar mark or use the step buttons, and watch the same work stack side by side, finish sooner and need a bigger core team. The bar stays in view while you scroll.
Scores and weights
Every attribute is scored 1 to 5 against the definitions in How it works. Weights are relative and shared with everyone in the workspace, so a change here moves the matrix and the dashboard for everyone.
The four parts
Three groups of functions on one foundation
Shawn asked for three or four boxes with every function underneath as a sub. These are the four, and every function sits in exactly one.
- Sales. Win and grow the customer: finding new accounts, growing the ones we have, quoting, pricing, selling online, answering product questions and serving customers.
- Sales operations. Get the order to the customer: order entry, sourcing, buying, inventory, the warehouse, delivery, returns and site programs.
- Back office. Run and support the business: paying suppliers, credit and collections, finance, people, marketing, IT, compliance, leadership and acquisitions.
- Foundation. What every function runs on: the data layer, ERP connections, product data and shared memory the integrate level needs, plus AI security and an AI assistant for every employee. Its name is still open.
Work crosses the parts. Quote to cash starts in sales, runs through sales operations and ends in the back office. Procure to pay starts with sourcing and ends with a supplier paid. The Business view shows both.
How far AI goes
Assist, automate, integrate
Every step in every function is rated by how far AI can take it by 2030.
- Assist. People do the step with an AI assistant and shared prompts. It takes assistants for everyone and some training.
- Automate. An AI tool does the step and people check the exceptions. It takes a tool we buy or build.
- Integrate. It reads and writes the ERPs and the data layer, so the step runs end to end. It takes the foundation underneath.
Where a step stops short of integrate, a person is making the call. Capture the reasons there, every time, so the step can move up later. Build in order: prompt before app, app before integration.
What the matrix is for
Value and complexity decide where each item lands
Every opportunity gets two scores: what it’s worth to DSG (value) and how hard it is to do (complexity). Each is a weighted average of five attributes, scored 1 to 5 against fixed definitions, so a 4 means the same thing at Lawson, TestEquity and Gexpro Services.
The two scores place the item on Shawn’s two-by-two. Where it lands decides what we do with it. The build order then decides when, given dependencies and money.
Keep the two scores separate until the picture. A single blended number hides the difference between “easy and worth a little” and “hard and worth a lot,” and those get very different treatment.
The four boxes
Shawn’s four boxes
Value runs left to right. Complexity runs top to bottom, with low complexity at the top, so the top right is the best place to be.
Choose
Worth doing, in a deliberate order. Cheap, quick and good for building comfort with AI. They fill gaps when a team has room and never push out a Do it or a Decide.
Do it
Fund and start. These are the first pilots and the proof points for leadership. RFQ to quote is the model: SalesPatriot is okayed at Gexpro Services, a product exists, and value shows up in months.
No
Not now. Most items here are early rather than wrong. Revisit when a foundation lowers the complexity or the value changes. Merging the three ERPs is the example: Shawn’s call is the data layer instead.
Decide
Leadership decides each one; some we do, some we don’t. The ones we do usually wait on a foundation and run longest, so they need a decision early even when the work starts later.
Items are placed where they score. An item in Decide stays in the build plan unless leadership takes it out. An item in No stays out until someone puts it in.
The attributes
What value and complexity are scored on
Shawn named AI type, current cost, current complexity and FTEs impacted, and he put opportunity into the final priority. The other five are proposed so value and complexity each rest on five measures. Confirm or cut them in the working sessions.
| Attribute | 1 | 3 | 5 | From |
|---|---|---|---|---|
| Value | ||||
| People affected | Under 10 | 50 to 150 | 500 or more | Shawn: FTEs impacted |
| Cost today | Under $250K a year | $1M to $3M | Over $10M | Shawn: current cost |
| Margin and revenue | None, or indirect | $1M to $3M a year | Over $10M | Shawn: opportunity |
| Reach | One team at one company | One company, all sites | All three companies and future acquisitions | Proposed |
| Fit with the long-term concept | Standalone | Runs on the data layer | A core piece of the platform | Proposed |
| Complexity | ||||
| AI type | General LLM, off the shelf | Local LLM: private, on our documents | Data LLM on transaction data across companies | Shawn |
| Process today | One standard process | Different by company or branch | Judgment-heavy, different everywhere | Shawn: current complexity |
| Data readiness | Clean, in one system | In one ERP, needs cleanup | Across ERPs, or in PDFs, email and people’s heads | Proposed |
| Integration | Stands alone | Reads from and writes to one ERP | Writes back to several ERPs or the warehouse system | Proposed |
| Change and adoption | One small internal team | A department at one company | Hundreds of people, or customer-facing, everywhere | Proposed |
Cost today is what the work costs now. What it costs to build, Shawn’s “estimate investment,” lives in the build order, not in the score.
Foundations
How foundation items are scored
Scored on its own, the common data layer lands in No: nobody uses it directly and it’s hard to build. Yet pricing, cross-sell, sourcing, inventory planning and the shared warehouse system all need it.
So a foundation item is placed at the value of the most valuable item in the plan that depends on it, and the chart draws a thin line from where it scored on its own. Without this rule a ranked list puts the data layer near the bottom and everything above it stalls.
Build order
Order of operations, then the calendar
Four rules set the order:
- Committed work goes first: what is in flight or approved, such as the data layer proof of concept and SalesPatriot at Gexpro Services.
- Then Do it, then Decide, then Choose. Within a box, the most value per point of complexity goes first.
- Nothing starts until what it depends on is finished. In Shawn’s words, C must always precede E.
- Each team takes on a set number of changes at once: sales, operations, back office, and IT for the foundation. The setting is at the bottom of the dashboard.
The investment slider sets the annual budget, core team included. More money runs more work side by side, so the curve of value delivered bends up and to the left. The core team gets thicker as it does: two people for architecture and security, plus about one person per active project, split between project management and a business lead from the companies.
Past a point, more money stops shortening the plan. Dependencies and each team’s capacity to absorb change set a floor, and the dashboard shows where that point is.
Rough costs
Orders of magnitude until a partner quotes
Each item’s cost and duration come from its approach and its complexity score:
- Buy, a market product with light setup: $100K plus $75K per complexity point; 2 months plus 1 per point.
- Configure, a partner platform set up for our process: $200K plus $150K per point; 3 months plus 1.25 per point.
- Build, built with a partner on our data: $350K plus $300K per point; 4 months plus 1.5 per point.
- Platform, a shared foundation: $1M plus $500K per point; 8 months plus 2 per point.
Each company beyond the first adds 20% to the cost and a month to the rollout. Replace the estimate with a real number in the item’s detail as quotes come in. Shawn’s test is to check each item with someone who has a strong option in the market, so the numbers are in the ballpark.
Dictating notes
Dictate a note and Claude files it
The Dictate button sits in the corner of every view. Talk or type anything from a conversation, then choose Sort it. Claude splits it into separate points and files each one: a note or decision on a function, an open question, a new idea, something Shawn asked for, or an opportunity going in or out of the plan. You see every point before anything is saved, and you can change where it goes.
To talk instead of type, use your keyboard’s dictation key. Editors and admins can dictate. When Claude isn’t available, File it myself makes the same note and suggests the function from the words in it.
How to run it
From the first pass to leadership
- Raw, starting October 26. Correct the scores, add what’s missing and confirm the attributes. Blank slate first, then map what’s already in flight. Weights come once the list is settled.
- Pressure test with Mike Marks and the HOPS team, right after the raw pass.
- The three companies, in early November. Bring them into one working session rather than sending a spreadsheet; Shawn wants them all involved together, and in a room the answers come back in a week instead of a month. Where two people score an item two or more points apart, talk that item through.
- Leadership by Thanksgiving. Show the matrix and the build order at two or three funding levels, with every item open to the detail underneath.