Case study 11 · Food Ingredients · AI outbound sales

One salesperson, a new product line, no buyer list — an AI desk that dials and hands over.

A large oilseed processor was moving out of low-margin animal-feed products and into food ingredients — coriander powder, coriander fibre, soya flour and protein. New products, new buyers, and one salesperson who had spent his career selling to feed makers.

We built an outbound desk where the AI does the first mile: a lead list is classified for fit, a voice agent calls, WhatsApp and email follow the call, and anyone who shows interest lands with the salesperson — who is the only one allowed to quote a price.

Client

Specialty-ingredients maker (name withheld)

Industry

Food Ingredients · Oilseed Processing

Location

Maharashtra & Telangana

Workflow

S · Outbound sales & follow-up

58%

Of 348 cold calls answered by a real person — 203 conversations the salesperson did not have to open himself.

~105 / hr

Calls the voice agent places from one console session, every one logged with its outcome.

Human quotes

The bot never states a price. Interest is handed to a person, and the buyer only ever reads human-written words.

What we shipped · 7 pieces

A customer is one person: a grid where every column edits in place and every row carries Call, WhatsApp and Email.

Fit verdicts on scraped lists — 2,244 leads classified in one pass into OK, check and drop — before a single number is dialled.

Voice agents per product line that dial mobiles only, respect do-not-disturb, and record answered, no-answer, interested and not-interested.

WhatsApp that never opens a conversation cold — only after the person has answered once — and email with the salesperson copied on every send.

Workflows that chain the follow-up off the call outcome, run in batches with a live count of what is owed.

A funnel of 32 stages from contact database to order under discussion; every stage is a link to that filtered list.

A human stage per interested product — specification and brochure shared, sample sent, trials, management meeting — owned by the salesperson, never written by the bot.

01 — The challenge

Hundreds of buyers to reach, ten emails a day.

The founder put it plainly: the one salesperson might send ten emails in a day, and the job needed hundreds. Nobody knew who the buyers of the new products were, and a specialist hire per product line did not pass the return test. Lead lists arrived as spreadsheets from expos and LinkedIn, and sat there.

The constraint was as important as the goal: no machine could ever quote a price, and no buyer should ever read something a human had not written.

02 — What we built

The AI opens the door. The human walks through it.

The platform was rebuilt from scratch on 4 September 2026 so that anyone in the company could use it: three screens — dashboard, customers, outreach — over one outreach table that is the whole history of every call, message and email. A lead list is imported, classified for product fit, and dialled by a voice agent at around a hundred calls an hour.

The first real call went to voicemail, and the system marked it not interested. That was the day's most important bug: a voicemail is not a rejection. It was fixed before the next batch, along with the rule that the answered-but-hung-up calls are a pitch problem, not a lead problem — the opening line was losing three in five before the question.

03 — The results

Three hundred conversations opened; the salesperson works the ones that matter.

By 16 September 2026 the contact database held 359 people, 348 had been called, 203 answered and 27 said they were interested — each one now sitting in a per-product stage the salesperson owns. A batch of 272 dials the same week ran end to end in an afternoon. Two days later the database stood at 594.

None of this replaces the salesperson. It replaces the ten emails a day, and it tells him each morning which conversations to have.

Before · After

A specialty-ingredients maker's outbound sales: before vs after the AI outbound desk
Metric	Before	After
Outreach per day	~10 emails, by hand	~105 calls an hour, plus follow-ups
Who the buyers are	Unknown	Lists classified for fit before dialling
Follow-up	Memory and spreadsheets	Chained off the call outcome, every touch logged
Who quotes	The salesperson	Still the salesperson — by design

A new product line and one person to sell it?

We'll build the desk that opens the conversations. Your people close them.

Book a meeting →