CG Common Ground | Cannabis Consulting
What we did and what it producedActive

The work, decision by decision

The work, in the order the decisions were made

  1. Research every store around decisions, not departments. We went store by store. The question was not "what does each department report" but "what does someone here decide every day, and what do they need to see to decide it well". Asked that way, the research follows a reorder, a delivery, a stock move or a regulator's request across every team it touches, in every store, which is where the gaps between teams show up. The questions it asks are in Attachment B.
  2. Name the functions the control center has to cover. The control center covers four: supply chain, warehouse, inventory and compliance. Four, because those are the functions whose mistakes land in the regulator's tracking record, and because every everyday decision in the operation needs at least two of them. Each decision has one agent that owns it and others that feed it, and compliance feeds all five.
  3. Give each function its own AI agent. One AI agent per function, built in-house, each watching its own signals on its own and raising what needs a person's attention without deciding for them, rather than one monolithic dashboard that tries to watch everything. An agent with one function is easier to test, easier to trust and easier to replace when the operator's needs change. What each agent watches on its own, and the decision each one hands to a person, is Attachment A.
  4. Put all four on one master control center, across every store. The agents are separate; the screen is not. The point of the control center is that a person looking at a supply problem sees, in the same place, what it means for the warehouse, the stock on hand and the compliance position, store by store. That is the handoff the operator could not see before.
  5. Build it and hand it over. We built the master control center with the four AI agents running on it, tracking every store, and the advisor and engagement lead seat reviewed what each agent surfaced before it went to the operator. It is theirs to run: the screen sits with the operator, the agents raise what needs attention on their own, and the decisions stay with the people who make them.
  6. Measure it against what it was built to do. The control center was designed against three projected outcomes, and the next honest step is to measure them on the operator's own numbers rather than add features to a screen that has not yet been judged by use.

What the control center was built against

What the operator holds today is the store-by-store research and its findings, and a master control center with four operating agents that tracks every store. The control center was built against its own projections, not savings anyone has banked, and they are the first thing to measure on the operator's own data.

What stays, and what the control center adds

Nothing the operator runs was replaced, and nothing should be. The operator's own systems stay the systems of record, including the seed-to-sale tracking system, which no dashboard gets to overrule. The control center adds the one thing none of them provides: a single view across the four functions and every store, with an agent per function raising what needs attention.

The faulty logic this corrects is common in regulated operators and is nobody's fault in particular. Each team picks the tools that make its own function easier, and each tool is right on its own terms. The cost shows up only where the teams meet: a reorder placed without sight of what is aging in the warehouse, a delivery accepted before anyone checks it against the tracking record, a regulator's question answered by three people pulling three reports. It has to change before the operation scales, because every new product, site or supplier multiplies the handoffs, and a gap one experienced person can cover by memory at one size does not survive growth.

What it costs to hold the line, and what to watch

Holding the order has a price. Research across every store before a build feels slow to an operator who already knows what screen it wants. Building from the findings rather than the wish list means saying no to features the operator asks for until the evidence is in. Both costs are worth paying again, because the alternative is a screen that shows the data that was easy to connect instead of the data the decisions need, and an operator who stops trusting it within a month.

What to watch now that the control center is in the operator's hands. Whether it agrees with the tracking record, because a control center that disagrees with the seed-to-sale system is worse than none. Whether each agent's signal has one named owner on the operator's side, because an alert nobody owns is noise. Whether people act on the screen without checking it by hand, which is the only real test of trust. And whether the projections hold once they are measured, because that is the difference between a screen that looked good and a control center that earns its place in the operation.

What it produced

The consulting agreement is active. The research across every store is done, and a master control center with four operating agents for supply chain, warehouse, inventory and compliance tracks every store and is in the operator's hands to run.

A slice of the project list

A few related projects.