A recall notice and a bid request have one thing in common at an independent distributor: both land on the desk of someone who already has a full job, and both eat the rest of that person's week. The big houses have a team of specialists for each, while an independent will have one person doing these things on the side of their desk.
Over the past year, we at Pepper have been running more of our workstreams on Claude and OpenAI, and we’re seeing massive benefit from handing more and more of the manual stuff to AI. As we talk to our distributor partners, we see use cases like these pop up all the time - so we decided the best thing we could do is start applying some of these third party tools to the issues we hear about most often.
Today we're sharing two skills for both Claude and ChatGPT that we built for independent distributors: one that catches a recall and works it, and one that builds a bid. Both are free, and they don’t require you to have Pepper at all to use. Check them out here or keep reading to learn more about them.
Download now:
Recall Watch and Response
The time between a recall notice landing and the first customer hearing from the distributor is critical. And the clock is usually running before the notice lands: Imagine that the FDA posts a Listeria recall on a bulk cream cheese Tuesday morning. That means the supplier's letter reaches the distributor Wednesday afternoon, and two delivery days might run on that item between when the recall was posted and when it reaches the customer.
When a notice does land, usually a chunk of time is wasted in figuring out exactly which item is being recalled. If the email says "Fully Cooked Diced Chicken Breast, 2/5 lb, item HL-44718, lots 6291A/6291B/6292A, undeclared soy.", a ton of time is taken before even letting customers know:
- It’s likely that nothing in the distributor's system is called that, so Purchasing would have to match it to an item number
- After matching item number, someone in Receiving walks the cooler reading lot codes because they aren't on the log and aren’t found anywhere else
- Sales then has to pull ship history and find everyone who got the item since the recall date because lots aren't captured at pick either.
Nobody can write the customer notice until all these things get done, and reps can't call their customers until the notice is written. That's the reason a Class I recall reaches customers the next morning and shuts operations down for an afternoon.
Recall Watch closes that gap: Simply give Claude or GPT an item master once and set the check to run every morning before the first delivery window. Each run reads the current FDA and USDA recall feeds, matches them against the catalog, and reports one of two things: Either nothing found (most days), or a list of items that appear in a recall with the match confidence stated.
Recall Response does the first hour's work in minutes. Paste in the notice however it arrived (email, PDF, a photo of the letter from the dock) and it hands back:
- The decode: what's recalled, which class, what the supplier is asking versus what the rules expect, and what's still unknown.
- The plan: an action plan matched to the class, the trace-back procedure, and the contact log the supplier and an auditor will ask for.
- The notices: the rep brief with a phone script, then the customer notice, then the company heads-up, in that order so the calls and the email don't disagree.
- Catalog substitutions: the recalled items mapped to the distributor's SKUs, anything else carried from the same plant flagged, and a substitute shortlist so the rep calls with "I've got two subs on tomorrow's truck" instead of nothing.
Before anything goes out, three things Claude/GPT can't see still need a person. Someone walks the cooler and reads the lot codes off the cases. Someone pulls the receiving log for the PO dates in the notice. Someone asks Purchasing whether the substitute is on the shelf. Each of those shows up in the draft as a blank like [Substitute: ___]; the person types in what they found and sends.
What it looks like
Bid Builder
A customer that started buying for one taco shop has now expanded to six, and the owner wants a deal: "If we buy everything from you, what's the price?"
Talking to a DSR about bidding for business recently, he described it concisely: “Bid building is super necessary, and it just sucks”. Pouring over purchase sheets and coming up with volume pricing isn’t anyone’s idea of a fun afternoon, and it’s time that’s better spent out on the road selling.
As a customer grows, locking in more of their business should get easier, not harder.
Bid Builder does the inside-sales work in an afternoon. Hand it an order history export and it hands back:
- The account summary: units, twelve-month spend, volume by category, delivery days, stated for correction instead of asked.
- The interview: the things it asks because no export has them: what the customer buys elsewhere and from whom, what the distributor is trying to win, who else is bidding, which items on the customer's list the distributor doesn't carry, and the margin below which they'd rather not win.
- The trade structure: what to hold, for how long, and what to ask for in return, on one rule: every commitment given needs something asked back. If a floor was named, the structure holds it, even if that means bidding high on purpose.
- The documents: the proposal, a franchise one-pager if the group needs one, and prep for what the incumbent will say in the meeting.
Bid Builder is built for the 5-to-15-unit local group, the account that grew up with the distributor. Formal RFPs from schools, hospitals, government, and national chains are a different motion, and the skill will say so rather than write the wrong document.
What it looks like
Get the skills
Both are zipped skill folders. Unzip, add to Claude or GPT, and the skill triggers the next time someone describes the job. No Pepper account required, and no form to fill out.
The big houses have had people whose whole job is recalls, and people whose whole job is bids, for decades. Independents have had the same person doing both, plus everything else. That's who we built these for.
