BWC AI Workshop
The full day
8 June 2026 · Mumbai
Session 1 Foundations · Session 2 Hands-on · Session 3 Deep-dive
Press F for fullscreen. Arrow keys or click to move. S for speaker notes.
BWC AI Workshop
Foundations and Magic Prompts
Session 1 · 8 June 2026 · Mumbai
By end of day today, you run AI yourself. Not watch. Run.
Before we explain anything
Look what AI can do today.
No magic tricks. Ordinary things, done in seconds. The kind you will do yourself by this afternoon.
You can do this today
Ask your numbers a question.
In plain English. Get the answer in seconds. No analyst, no waiting.
You can do this today
Turn reviews into a fix list.
Paste a week of customer reviews. Get the top complaints, ranked, in a minute.
You can do this today
A poster in a minute.
Describe it. Get designs, on brand, in any language. No agency, no wait.
You can do this today
Photograph anything.
A menu, an invoice, a competitor's board. AI reads it and acts on it.
You can do this today
Messy sheet to clean report.
Drop a messy export. Get a tidy summary and a chart back.
And our industry is moving
Pizza Hut. Wendy's.
Now take drive-thru orders with an AI voice. Hundreds of restaurants, this year.
Yum! Brands with Nvidia, and Wendy's with Google. 2025.
Companies run on it · Design
Heinz asked AI to draw "ketchup."
Every image came back looking like a Heinz bottle. They made that the campaign.
Heinz "A.I. Ketchup".
Companies run on it · Video
Training and ad videos with no camera.
An AI presenter, any script, 160+ languages. Used by 9 in 10 of the Fortune 100.
Synthesia. Customers include Zoom, Heineken, SAP.
Companies run on it · Building software
20 million developers build with an AI partner.
Software that took quarters now ships in weeks. 9 in 10 of the Fortune 100 use it.
GitHub Copilot.
Companies run on it · Virtual influencers
2.5 million people follow her. She does not exist.
Lil Miquela is a fully AI character. She releases music and has fronted campaigns for brands like Calvin Klein.
Instagram @lilmiquela. Spain's @fit_aitana is a newer, photo-real one that fooled people into thinking she was a real model.
Quick game
Real or AI?
Three rounds. Look, listen, watch.
Hands up each time you think it is AI.
Round 1 · Real or AI?
Did a designer make this?
AI. No designer, no stock library. Swap in your own AI food shot for an even bigger gasp.
Round 2 · Real or AI? Press play.
Is this a real customer call?
AI. Voice, tone, timing. It took the whole call on its own.
Round 3 · Real or AI? Watch.
Is this interviewer real?
Opens in a new tab (needs internet). A live AI interview for a real Belgian Waffle store-manager role.
AI. No human ran this interview. It asked, listened, and followed up on its own.
Now picture it inside BWC
One eye-opener for every function.
"Which outlet is losing money right now?" The fix, before chai.
Know exactly what each outlet needs tomorrow.
A full festival campaign in an afternoon. No agency.
50,000 customers, 50,000 personal messages.
Every competitor price change on your desk each morning.
Close the month in a day.
1,000 applications screened by tonight, fairly.
Describe the app. Watch it get built.
Not just the global giants
Watch. Right now.
Blank page to a board-ready BWC brief. 90 seconds.
By end of day, that is you. But first, how does this actually work?
Part 1 · How it works
So what is this thing?
Five minutes. No math beyond a menu board.
The idea
A prediction machine
It is not a database that looks up stored answers.
It is very advanced autocomplete that read most of the internet and learned the patterns of language.
So what: brilliant at language and drafting. Not a calculator. Not a fact vault.
What is inside it
The trained brain that predicts.
Everything it studied, once.
Your instruction to it.
The facts you hand it right now.
It reads and writes in small pieces, fast.
How it acts, and reuses know-how.
The honest bit
Three limits to respect
| The limit | In plain words | What you do |
|---|---|---|
| Hallucination | It can be confidently wrong | Ask for sources. Check what matters |
| Recency | Its memory has a cut-off date | Give it today's data yourself |
| Context window | Short-term memory has a size | Feed it the relevant slice |
Respect these three, and it is a force multiplier. Ignore them, it bites.
Part 2 · The one skill
Toy or tool?
The difference is the prompt.
Four levers
Who should the AI be?
What must it know?
What shape is the answer?
What must it not do?
Most people pull zero of these. Then they judge the tool.
Prompt engineering
The lazy prompt
Vague. Generic. Confident about nothing.
Prompt engineering
The strong prompt
Same model. Thirty seconds more input.
Same model.
The model did not get smarter. You got clearer.
Quick vote: which output would you actually send? It is unanimous.
Part 2 · Level up
Six techniques the pros use
Show two or three of what good looks like, it copies the pattern.
"Here are 3 captions we love. Write 5 more."
For anything tricky, reasoning beats a snap answer.
"Think it through step by step, then answer."
Catches the confident-but-wrong.
"Now review your answer against the data, flag any error."
Big task? Tell it the stages.
"Do it in steps: first pull, then summarise, then draft."
Keep nudging, it remembers the thread.
"Shorter. Warmer. Add a number."
Say exactly what shape you want back.
"As a table." "Five bullets." "A WhatsApp message."
Part 3 · The shortcut
Magic Prompts
Stop writing prompts from scratch.
Two kinds. Two frames.
Configure an assistant once. It behaves that way forever. Like hiring and training a teammate.
A perfect one-time brief. Like a spec so tight a vendor cannot get it wrong.
Each is fill-in-the-blanks. You never face a blank page again.
The System frame
Seven sections
Push-back mandate: you tell it to challenge you when you are wrong. That one line changes everything.
The Delegation frame
Eight blocks
Phased steps with check-points are the difference between handing off a job and babysitting it.
Now, live
A Magic Prompt for BWC.
Describe the outcome in one breath. The tool writes the prompt.
This exact tool is in your take-home. The blank page is retired.
Part 3 · The prompt library
Wrong prompt. Right prompt.
For every function.
For each one: read the lazy prompt, you fix it out loud, then we reveal the strong version. Yours is in the take-home.
Prompt library · Operations
OPS Operations
Lazy
Strong
Prompt library · Supply Chain
SC Supply Chain
Lazy
Strong
Prompt library · Marketing
MKT Marketing
Lazy
Strong
Prompt library · CRM
CRM Customer / Loyalty
Lazy
Strong
Prompt library · Sales
SALES Sales
Lazy
Strong
Prompt library · Finance
FIN Finance
Lazy
Strong
Prompt library · HR
HR Human Resources
Lazy
Strong
Prompt library · Tech
TECH Technology
Lazy
Strong
Challenge
Prompt Duel.
Two volunteers. Same task: a Monday offer for a slow outlet.
Sixty seconds to write the prompt.
The room votes on the better answer.
Winner names who goes next.
Pause · Your turn
Now you try it.
Ten minutes. I will walk the room and put a few of yours on the big screen.
Part 4 · The most useful thing
Your AI Executive Assistant
It runs your morning, and automates the routine your team does by hand.
While you sleep
It reads everything, and briefs you.
Slack, WhatsApp, email, your dashboards and reports. By 7am, one summary of what matters.
Then it acts
It drafts, flags, and schedules.
Drafts your replies. Flags the few things only you can decide. Books the meetings on your calendar.
This is not someday.
Aakash runs exactly this, every morning.
You describe your routine once. Then it is yours, and your team's.
Part 5 · Your playbook
One by one,
in your function.
The wow you saw, turned into a Monday playbook.
OPS Operations
Pain Mornings start by hand-compiling sales and wastage across outlets.
AI A daily store scorecard, written overnight, flagging the outliers.
Win Managers walk in to answers, not spreadsheets.
SC Supply Chain
Pain Fast-movers run out while perishables get over-ordered and binned.
AI Demand-sensing par levels from sales, day, weather, and events.
Win Less wastage, fewer stockouts on the items that drive footfall.
MKT Marketing
Pain A festival or daypart offer takes days to brief, write, and localise.
AI Brief, offers, and channel-ready copy per city, in minutes, on brand.
Win Campaign turnaround drops from days to an afternoon.
CRM Customer / Loyalty
Pain One blast goes to everyone. Regulars and lapsers get the same message.
AI Segment the base, tune a message to each, surface who will churn.
Win Higher repeat orders. Spend aimed where it pays back.
SALES Sales
Pain Aggregator performance and new leads tracked by hand, slowly.
AI A daily Swiggy and Zomato digest, plus auto-drafted lead follow-ups.
Win Faster lead response and a clear read every morning.
FIN Finance
Pain Outlet P&L consolidation and variance write-ups eat the first week.
AI Ask the numbers in English. Get the variance and a draft commentary.
Win Month-end shrinks. Finance explains, instead of assembles.
HR Human Resources
Pain High-volume hiring, repetitive JDs, exit reasons never themed.
AI Draft JDs and screening, and theme months of exits into the top five.
Win Faster hiring, and attrition you can finally see.
TECH Technology
Pain A backlog of report and data requests, and no time to document.
AI Let teams ask in plain English, auto-route tickets, generate the docs.
Win The backlog drains. Teams self-serve.
Before you go · Your toolkit
Five tools. That is all you need.
Pick one daily driver. Add a specialist only when the job needs it.
Every one has a free or low-cost tier. No big budget needed.
Take a photo. This is yours.
Your AI cheat sheet
Role + Context + Output + Guardrail.
"You are [role]. Here is [data]. Give me [output]. Do not invent numbers, flag what you are unsure of."
Do not write the prompt. Describe the outcome, let the tool write it for you.
Chat to think. Cowork to do it with you. Code to build. Chrome to act on the web.
It can be wrong. It misses last week. Feed it the slice. Verify what matters.
Your function's exact prompt is in the take-home. Snap this now, use it Monday.
You saw the world.
Now you have the playbook.
A worksheet for your function, a one-page cheat sheet, and a take-home that holds all of this.
That is Session 1
Lunch. Back at 14:45.
Analytics, CRM, and Marketing leads: bring a laptop. We put hands on the keys.
BWC AI Workshop
Hands-on with your data
Session 2 · 8 June 2026 · Mumbai
You brought a laptop. Good. This session, you drive.
This session, you drive.
Live data. A dashboard you can talk to. A competitor scrape. Segments. A scheduled robot.
Four surfaces. One assistant.
Talk to it, use Chat. Build with it, use Code. Send it onto the web, use Chrome. Work a task together, use Cowork.
Before we build · three steps
Set up Claude Code
1. Install. One line in your terminal.
2. Sign in. Type claude, log in through the browser.
3. Check. claude --version shows a number. Ready.
Set up in advance? Skip straight to the build. Stuck? Follow along at claude.ai.
Build 1 of 5
Build it with me. Step by step.
Laptops open. No watching. Every step is on the screen. We finish with a real, working tool.
Setup · Follow along
Open Claude Code.
The AI that actually does things: reads your files, writes, posts, schedules.
Drop today's sample file, sales.csv, into the folder.
Step 1 of 6 Type this
See what it can see.
It can read the files in your folder. Just ask.
Step 2 of 6 Type this
Read the data.
No formulas. Ask for a plain-English summary.
Step 3 of 6 Type this
Ask a real question.
The kind you would ask an analyst. It reasons over the numbers.
Step 4 of 6 Type this
Get the deliverable.
Tell it the shape and the guardrail. It writes what you would send.
Step 5 of 6 Type this
Make it act.
It does not just talk. It posts, sends, files.
Step 6 of 6 Type this
Make it repeat.
Build once. Schedule it. Forget it.
Six steps. One tool.
You just built a briefing robot.
Now swap sales.csv for your real data, and it is yours.
Your turn now: rebuild it on your own export. The same six steps work for any report. Sheet: demos/session_2_demo_1.md
Your shot
Stump the Bot.
Throw your hardest BWC question at it, live.
Nails it? Admit you are impressed.
Flops? You just found a limit worth knowing.
Tableau · Step by step
Pull your Tableau data. Get insights. Auto-email them.
Five steps. By the end, a clean insight email lands in inboxes every Monday morning without you.
Step 1 of 5 Connect to Tableau
Sign in with your token.
One line. Claude handles the REST API call. You do not touch any JSON.
Step 2 of 5 Pull this week's data
Name the workbook and the view.
It pulls the latest data, outlet by outlet, ready to read.
Step 3 of 5 Generate insights
Ask for the story, not the numbers.
Tell it who reads this and what they need. It writes the five things that matter.
Step 4 of 5 Format and send the email
Name the recipients. It sends.
Formatted subject, clean HTML, the right people. One instruction.
Step 5 of 5 Schedule it forever
Once a week. Without you.
Set the schedule. The email lands in inboxes every Monday morning. You do nothing.
Your turn
Now build yours.
Ten minutes. I will demo a variation on screen while you work.
Build 2 of 5
Ask your BI model in plain English. Get a chart.
A "semantic model," in plain words
The tidy, agreed layer your BI tool sits on: sales, orders, average order value, by outlet, by day.
Until now, only analysts could query it.
Today you ask it a question in English, and it answers.
The exact connector pattern proven at Cars24. Your BI stack, same idea. Pattern, not data.
Hands-on Ask and chart
No one waited on the analytics team. You asked, you got a picture.
Handout: demos/session_2_demo_2.md
Pause · Your turn
Practice, then we go again.
Ten minutes. Raise a hand if stuck. I will demo the next build while you work.
Build 3 of 5
Connectors. Giving AI hands.
What is a connector (MCP)?
A universal adapter that lets Claude safely use an outside tool: a browser, a database, a file store.
it can only talk
it can act
An advisor who tells you what to do, versus a colleague who just does it.
Hands-on Watch a competitor, live
Competitor pricing intelligence, on demand, without a research vendor.
Handout: demos/session_2_demo_3.md
Build 4 of 5
Customers into segments. Segments into messages.
CRM, deeper · Real examples
Companies already run CRM on AI
"Deep Brew" picks the next offer for 30M+ rewards members.
AI recommendations and Beauty Insider personalise every touch.
Recommendations drive about a third of sales. Personalised email at scale.
Personalised playlists, and the Wrapped campaign everyone shares.
AI personalises the feed and the offer, per user, per craving.
Personalised rows keep you watching. Retention by recommendation.
Same playbook for your customers: know them, segment them, speak to each one.
Real example · QSR
Starbucks runs loyalty on AI.
"Deep Brew" sends each of 30 million-plus members a different next-best offer. Real time, per person, on what they actually buy.
Personalised offers, more visits, more spend. The loyalty engine is AI.
CRM, deeper
From customer data to the right message
Understand the purchase pattern, then run the campaign that fits it.
Hands-on Customer 360 to the right campaign
One blast becomes the right message to the right regular.
Sample data and handout: demos/session_2_demo_4.md
Build 5 of 5
Research that builds itself every morning.
Hands-on Hire a robot researcher
Build the brief once. Then schedule it. It runs while you sleep.
Handout: demos/session_2_demo_5.md
Speed round
Sixty seconds.
Shout out a task from your function AI could take off your plate.
Keep them coming. No bad answers.
We build the most-requested one, live.
Do not build what you can buy
Economical tools for each job.
Two or three options each, with real prices. For when a ready-made tool beats building.
Real brands. Real tools. India.
Indian D2C brands already run on AI.
These are not pilot projects. These are live, this year.
For CRM and WhatsApp
Tools Indian D2C brands use today
Broadcasts, chatbot, no code.
From Rs 1,500/mo
Used by 35,000+ businesses in India
Sell and support over chat.
From ~Rs 1,200/mo
Backed by Jio Haptik
Visual chatbot builder.
Pay-as-you-go, Rs 999
Cheapest legitimate entry
For marketing and content
Create faster. Distribute smarter.
Posters, reels, brand kit, AI edits.
Free, Pro ~Rs 1,200/mo
The easiest start
Social posts, video, voiceover. Indian team.
From ~$32/mo
Best for social volume
Copy, campaigns, briefs, ideas.
~$20/mo
Cheapest creative brain
For voice bots
Phone agents. Orders, feedback, support.
Built in India. Open-source base.
~Rs 2.5/min
Cheapest, India-first
Advanced voice agent builder.
~Rs 4/min
Most flexible
Reliable phone agents, fast setup.
~Rs 6/min
Easiest to start
Pay per minute. A customer feedback call costs two to six rupees. You pay only for what you use.
Demo · Step 1 A voice bot in 5 minutes
Describe your bot.
Plain English. No call-flow charts, no code.
Demo · Step 2 A voice bot in 5 minutes
Give it a voice and a number.
Pick an Indian-English voice. Attach a phone number. Done.
Demo · Step 3 A voice bot in 5 minutes
It calls, and reports back.
It makes the calls, listens, and hands you a summary of each one.
One more multiplier
Skills
Teach Claude once. It just knows, forever.
What is a skill?
A saved playbook Claude loads by itself when a task fits. You write the expertise down once.
Like hiring a specialist and writing their playbook, so every Claude follows it the same way.
Aakash already runs many: a prompt writer, an inbox chief-of-staff, a self-writing weekly brief, and packs that hold an entire business playbook.
What are they for?
The same quality output, every time.
Your outlets, metrics, and rules, baked in.
Everyone's Claude follows the same playbook.
One person writes it, the whole team gains.
How to add one
Easy path: describe the repeatable task and ask Claude to turn it into a skill. It builds it for you.
Under the hood: a folder with one file, SKILL.md. A name, a "when to use it" line, and the steps. Claude runs it automatically when a task matches, or you type its name.
Now anyone at BWC types "morning brief" and gets the same standard output. That is a skill.
You touched all of it.
Live data. A talkable dashboard. A scrape. Segments. A scheduled robot.
Now take one of these and make it something BWC runs every week.
BWC AI Workshop · Session 3
The agent that owns your Monday morning
Deep-dive with Ankit · 8 June 2026
Assumed focus: the weekly state-of-business. We retarget in two minutes if Ankit names a different pain.
The pain
Every week, someone assembles it by hand
Pull from a few dashboards. Line up this week against last. Write the story. Format it. Send it.
High-value, repetitive, and it eats a day.
The target
One agent. Every Monday at 8am. No one assembling.
Reads the week's numbers. Computes the changes that matter. Writes the narrative. Delivers it. Runs on a schedule, untouched.
The same pattern Aakash runs at Cars24. Tonight, the BWC version.
How it is built, in plain shape
Sales, aggregator, supply, loyalty, ops. Week-on-week. Narrative. Slack plus a one-page brief. Every Monday.
Let us build it. Live.
What we wire, step by step
Full build sheet: demos/session_3_demo_1.md
Every Monday
What lands
The three moves that matter, with why.
The one risk to watch this week.
The one decision to make.
A low-confidence flag on anything stale.
Same shape every week. Leaders learn to read it in a minute.
Handoff and cadence
Owner. Ankit owns the agent.
Cadence. It runs Monday 8am. Review the first few runs, then trust it.
Extend. Add a source or metric by editing one instruction, not rebuilding.
Guardrail. It flags low-confidence numbers rather than papering over them.
Next 30 days, Ankit.
trust the brief
add a source
a second agent
hand one to your team
The goal is not one agent. It is a team that builds them.