The best applied AI work in most companies isn't happening in a lab, a platform team, or a slide deck titled "AI Strategy 2027" — it's happening on the sales floor, built by SDRs and AEs who are measured on outcomes every single week and have no patience for a tool that doesn't move a deal by Friday. Dewi Anggraini, an account executive selling B2B logistics software into mid-market importers across Jakarta and Surabaya, didn't wait for her company's AI rollout. She built her own first, out of a transcription app, a free chatbot tab, and a WhatsApp habit that predates any of it.
Her company's official AI initiative arrived nine months after she'd already solved her own problem. By the time IT sent the onboarding email for the "approved AI notetaker," Dewi had been quietly running her own stack for two quarters — pasting call transcripts into a chatbot to draft follow-up emails, asking it to summarize a rambling forty-minute call into three bullet points she could actually act on, and using it to translate a prospect's Bahasa Indonesia voice notes into English so she could loop in her US-based sales engineer without a delay. None of it was sanctioned. All of it worked.
The Best Applied AI in Your Company Is on the Sales Floor
Strategy teams write the AI roadmap. Reps write the AI workflow that actually survives contact with a real Tuesday. The difference is accountability: a platform team's AI project gets judged in a quarterly review; a rep's improvised prompt gets judged the moment a prospect replies — or doesn't. That immediate, weekly feedback loop is what makes frontline sellers the most ruthlessly pragmatic AI users in almost any organization. They don't adopt a tool because it's impressive. They adopt it because it got them to the next call five minutes faster, and they drop it just as fast if it doesn't.
This is not a small or isolated phenomenon. Across the revenue teams we talk to in India, Southeast Asia and the Gulf, the pattern repeats: a handful of reps — usually the top quartile by quota attainment — are running informal AI workflows that never touched a procurement form. They stitch together a transcription tool, a general chatbot, a scheduling assistant and a personal note-taking habit into something that functions like a lightweight sales system, built entirely out of tools that were never meant to work together.
"Nobody trained me on any of this. I just got tired of writing the same follow-up email at 9pm after a call that went well, so I started pasting my notes into a chatbot and editing what it gave me. It took me a week to trust it. It took me a day to realize I'd never go back."— Dewi Anggraini, Account Executive
Why Reps Out-Innovate the AI Task Force
A centralized AI task force optimizes for governance, procurement approval and a defensible rollout plan — all reasonable goals that also guarantee the project moves slower than a rep's actual quota clock. Reps don't have that luxury, and it turns out that constraint is a feature. An SDR with forty calls a week and no budget line for enterprise software has exactly one incentive: does this thing save me time on this specific, ugly, real task in front of me right now. That is a far sharper filter than anything a steering committee produces in a slide about "AI use cases."
The task force also tends to start from the tool and search for a use case. The rep starts from the pain — a call that needs summarizing before it's forgotten, a lead that needs qualifying before the SDR's next dial, a quote that needs redrafting in the buyer's language before the moment passes — and grabs whatever's on hand to solve it. That's applied AI in the literal sense: applied to a specific, felt problem, by the person who owns the outcome, under a deadline that can't be extended by a steering committee meeting.
What the Reps Actually Built (and What Broke)
Dewi's stack, like most rep-built workflows, was resourceful and genuinely fragile at the same time. It solved real problems. It also broke in ways that only show up once you look at the whole pipeline instead of one rep's afternoon:
- Transcript-to-follow-up drafting — worked well for tone and speed, but nothing wrote the summary back into the CRM, so her manager's pipeline view stayed exactly as stale as before.
- Voice-note translation for a cross-border deal team — solved a real language gap, but lived entirely in Dewi's personal chat history, invisible to anyone who inherited the account if she left.
- A personal prompt for qualifying inbound leads — genuinely sharper than the old scoring spreadsheet, but every other AE on the team was scoring leads a different way, because each had built their own version alone.
- Copy-paste between three separate tools — every workflow step required Dewi to manually move information from one tab to the next, which meant the "automation" still cost her real minutes and real chances to paste the wrong thing into the wrong deal.
None of that makes the workaround a failure. It makes it exactly what shadow IT has always been: proof of an unmet need, built by the person closest to the pain, with none of the guardrails a company would want at scale. The busywork got shorter. The system of record didn't get any more honest. And the moment Dewi went on leave for two weeks, her whole workflow — and the pipeline hygiene riding on top of it — went on leave with her.
From Shadow AI to a System Everyone Can Trust
The fix isn't to shut the workaround down — reps who've been quietly told to stop using the tool that actually works for them just go back to doing it without telling anyone, which is worse. The fix is to notice what they built, understand why it worked, and rebuild the same value into something that writes back to the record instead of living in a personal chat history. Dewi's translation habit, her drafting shortcut, her qualifying instinct — each one is a real, validated workflow. What's missing is the plumbing that makes it durable, shareable, and visible to a manager without asking Dewi to become a systems administrator on the side.
That's the actual design brief for enterprise AI in sales: don't replace what reps already invented, formalize it. Take the transcript-to-summary habit and wire it to update the deal automatically. Take the lead-qualifying instinct and turn it into a consistent score every AE inherits instead of reinvents. Take the translation step and make it part of the pipeline instead of a favor Dewi does from her own phone. piRevenue was built on exactly that premise — every one of Dewi's improvised habits is a feature elsewhere in the platform, minus the copy-paste and minus the risk that it walks out the door with her.
See what a formalized rep workflow looks like
Where the Human Still Closes
Formalizing the workaround doesn't mean handing the deal to the agent. The point of noticing what Dewi built isn't to automate Dewi — it's to give every AE on her team the version of Dewi's Tuesday that used to only exist for the top quartile clever enough to build it themselves. Agents can draft the follow-up, translate the voice note, flag the quiet deal, and update the record the moment a call ends. None of that touches the read on a hesitant buyer, the decision to hold firm on price, or the sentence that actually closes. That's still Dewi's, every time, reviewed and sent in her own voice.
What changes is who gets access to the workflow the best reps were already running in private. The lesson from the sales floor isn't that reps need less oversight — it's that they've been doing the R&D for free, under real pressure, for longer than most AI committees have existed. A company that notices, learns from it, and builds the missing plumbing ends up with every rep operating at the level its best rep already reached alone.
Stop feeding the CRM. Watch it feed you.
