VIP Blog | Insights for the Beverage Alcohol Industry

How Distributors Get AI Pilots to Scale

Written by VIP | Nov 16, 2026, 2:20:25 PM


Most distributors that have tried AI can point to a test that worked. Predictive orders for a handful of reps. A forecasting tool in one branch. A routing trial run by your best router. Then the rollout to the rest of the operation slows, and the project stops.



Our first recap from VIP's panel on AI in beverage distribution at the NBWA Annual Convention looked at where AI pays off. This one takes on the harder part: getting from a successful pilot to results across the whole business.

The panel brought together VIP CEO Andrew Criezis, VIP Head of Product Brian Kelly, and Sahil Puri, a partner with McKinsey & Company's QuantumBlack division, with VIP Head of Sales Lucinda Ochwat moderating.



Pilots work. Scaling is where they stop
.

“Pilots are actually more often than not successful. Things just stop going beyond pilots for a lot of companies.”

Sahil Puri, Partner, McKinsey & Company

McKinsey's research shows about 30% of pilots make it to scale, Sahil said. Most pilots are built to win. They run in the easiest part of the business, with people who already want AI, and with leadership paying close attention. That setup wins the pilot, but it doesn't carry over.

Once a tool leaves the pilot, it meets complex processes like order-to-cash, teams that are skeptical, and data spread across systems that don't talk to each other. For a distributor, that means the ERP, supplier portals, spreadsheets, PDFs, and notebooks on desks.

The people closest to the work carry the change

McKinsey's research also shows 47% of managers and individual contributors report AI-related strain, compared with 31% of executives. In distribution, those are the sales managers, reps, merchandisers, and warehouse leads whose day changes most when a pilot becomes standard practice.

One audience question put it plainly. Many field reps have worked the same way for 10 to 15 years, merchandisers may be part-time, and some have never touched an AI tool. How do you bring them along?

  • Start with AI that recommends before AI that decides. Most companies start with AI-assisted work, and once the recommendations prove themselves, often over about 12 months, people become comfortable letting AI act on its own.
  • Make the payoff visible to the people doing the work, in hours saved and time back with customers.

Change the workflow, not just the screen

McKinsey found that about 1 in 4 organizations outside the AI high performers have redesigned their workflows, compared with about 3 in 4 of the high performers. Adding AI to an old process gives you the old process with a new interface.

Predictive ordering shows the difference. When the system builds most of the order, the rep's job shifts from building orders to reviewing them and selling. That shift needs new habits, new measures, and training.

 

Three moves separate pilots that scale from pilots that stall.

1. Run several small pilots, not one perfect one

Companies that scale AI tend to run several pilots at once, in different parts of the business, with different processes, levels of complexity, and types of people. For most distributors, two or three small pilots is a practical start: one in the warehouse, one in ordering, one in the back office.

Keep each one measurable, since fewer people means less friction, and include at least one team of skeptics. What you learn from them is what you will face at scale.

2. Fix data alongside the pilot, not before it

One company told Sahil it needed three years to fix its data before starting with AI. He called that “absolutely not what anyone should be doing.”

Distributors have done this before. When warehouses started validating pallets with scales 10 to 15 years ago, not every product had an accurate weight. That didn't stop anyone from starting. Teams found and fixed bad weights as pallets moved through, until they could trust the system.

Data for AI works the same way: run the first pilots where the data is good enough, and clean up the data the next wave will need while they run.

3. Plan the rollout before the pilot ends

Scaling changes how the work is funded, who owns it, and how people are trained. Settle those before the pilot wraps up:

  • Find the budget for a wider rollout.
  • Identify the owner accountable for the result.
  • Build a training plan for every role the tool touches.

McKinsey's own lesson was that retraining people took more time than anything else.

How VIP helps AI scale

A pilot is easier to scale when the AI is built into the platform you already run, rather than added as one more tool. That's how we build AI at VIP: into the ordering, forecasting, and routing workflows distributors use every day, so a pilot that works has a clearer path to the rest of the operation.

Ready to pick your first workflows? Talk to VIP about where AI fits in your operation.

Three questions to ask your team this week

  • How many AI pilots are we running, and do they span more than one department and type of user?
  • If our best pilot went company-wide next quarter, who would own it, and what would it cost?
  • Which data problem would we fix first if we started now instead of after a cleanup project?