Skip to content
hiromi

For enterprises

AI across your company, one division at a time.

Find where AI fits in each division, start with quick wins, and scale what works, with your teams on board and the impact measured at every step.

Plan an AI assessmentSee use cases by division
enterprise AI ideas mapped and triaged
~200
enterprise AI ideas mapped and triaged
divisions covered by one AI roadmap
50+
divisions covered by one AI roadmap
AI solutions already built
21
AI solutions already built

01AI maturity

Where is your company today?

Most companies sit at level 1 or 2. Knowing your level tells you the right next step.

  1. 01 / 04

    Exploring

    You are here if

    • A few people use ChatGPT on their own
    • No shared rules for company data
    • Nobody tracks what AI delivers

    Next step: Assess every division and pick the first quick wins.

  2. 02 / 04

    Experimenting

    You are here if

    • One or two pilots, run by enthusiasts
    • Results are anecdotes, not numbers
    • Each team picks its own tools

    Next step: Turn the best pilots into measured solutions.

  3. 03 / 04

    Scaling

    You are here if

    • Several AI systems in daily use
    • Clear rules for data and approvals
    • Impact is measured per use case

    Next step: Roll out across divisions with champions and a catalogue.

  4. 04 / 04

    Embedded

    You are here if

    • AI sits inside core processes
    • Every division has an AI owner
    • Leadership reviews impact and cost monthly

    Next step: Keep improving as better models arrive.

02Use cases

What AI can do in your division

Common AI use cases we see in enterprise backlogs. Pick a division to see what fits, and how we would deliver it.

Agent Skill
Live in days, inside tools you already use
Solution
One of our proven solutions, adapted
Custom
Built for your process

Less retyping and matching, faster closing and fewer payment errors.

6 use cases

  • Invoice packages

    Invoice, tax invoice, proof and purchase order matched into one package per invoice.

    Solution

    How it works

    1. 1Billing files for the cycle are collected
    2. 2AI finds each invoice's documents and amounts
    3. 3Billing approves every package

    What we measure

    Hours per billing cycle, and packages held back for errors

    How we deliver it

    Solution · Live in 4–8 weeks · From Rp 35 million

    Discuss this use caseSee the full solution

  • Payment checks

    Payment documents checked against contract, order and budget before anyone pays.

    Solution

    How it works

    1. 1Payment letter, contract, order and budget are uploaded
    2. 2AI reads each one and compares the values
    3. 3Cost control decides on anything flagged

    What we measure

    Payments stopped before an error, and review time per payment

    How we deliver it

    Solution · Live in 4–8 weeks · From Rp 35 million

    Discuss this use caseSee the full solution

  • Expense capture

    Receipt photos turned into expense records, with proof attached.

    SolutionWatch it work

    How it works

    1. 1Staff photograph a receipt
    2. 2AI fills in date, vendor, total and items
    3. 3Admins verify and export

    What we measure

    Time per expense, and receipts without proof

    How we deliver it

    Solution · Live in 4–8 weeks · From Rp 35 million

    Discuss this use caseSee the full solution

    ▶ Watch it work

    Input

    struk_toko_sumber_rejeki.jpg

    • TOKO SUMBER REJEKI
    • Jl. Pasar Baru 12, Bekasi
    • 11/10/2026 09:14
    • Beras 25kg ........ 320.000
    • Minyak 5L ......... 92.000
    • Telur 2kg ......... 58.000
    • Gula 3kg .......... 51.000
    • TOTAL ............. 521.000

    Step 0 of 9

    What the AI produces

    Tanggal
    Vendor
    Rincian
    Total
    Kategori

    Automatic checks

    • Total = jumlah semua item
    • Foto struk terlampir sebagai bukti
    • Tidak ada struk ganda bulan ini
    Result—

    Fictional sample data. Your own documents work the same way.

  • Bank, receivable and payable reconciliation

    Transactions matched automatically, with only the exceptions left for a person.

    CustomWatch it work

    How it works

    1. 1Bank statements and ledger are exported
    2. 2Transactions are matched automatically
    3. 3Finance clears only the exceptions

    What we measure

    Share matched automatically, and days to close

    How we deliver it

    Custom · Scoped after a free call · Quoted per scope

    Discuss this use case

    ▶ Watch it work

    Input

    bank_statement_oct.csv · ledger_oct.xlsx

    • 02 Oct Transfer in Rp 12,500,000 PT Mitra Jaya
    • 03 Oct Transfer in Rp 4,750,000 CV Sinar Abadi
    • 05 Oct Transfer out Rp 2,100,000 Vendor payment
    • 07 Oct Transfer in Rp 8,000,000 no reference

    Step 0 of 8

    What the AI produces

    Matched
    PT Mitra Jaya
    CV Sinar Abadi
    Vendor payment

    Automatic checks

    • Amounts match the invoices
    • No duplicate payments
    • Every transfer has a reference
    Result—

    Fictional sample data. Your own documents work the same way.

  • Monthly financial report

    A first draft of the monthly report and its commentary, from your ledger.

    Agent Skill

    How it works

    1. 1Ledger and last month's report go in
    2. 2AI drafts the figures and commentary
    3. 3Finance reviews and signs off

    What we measure

    Hours to a first draft

    How we deliver it

    Agent Skill · Live in 1–2 weeks · From Rp 7.5 million

    Discuss this use case

  • Group cash report

    Cash positions from every entity consolidated into one daily view.

    Custom

    How it works

    1. 1Balances come in from every entity
    2. 2They are consolidated and checked against yesterday
    3. 3Treasury sees one daily view

    What we measure

    Time to a group cash position

    How we deliver it

    Custom · Scoped after a free call · Quoted per scope

    Discuss this use case

03Automations

Automations that connect your systems

The hard part of automation is not the happy path. It's what happens when a payment doesn't match or a system is down. Each of these handles that, and hands exceptions to a person.

  • Payment confirmation from WhatsApp

    A customer sends a transfer screenshot. The payment is checked against your bank, the order is marked paid and a receipt goes back, with no admin in between.

    WhatsAppAIBankOrdersAccounting

    Run log

    Step 0 of 6

    1. 10:02:01WhatsAppTransfer screenshot received for order #2207
    2. 10:02:04AIRead: Rp 1,250,000 from BCA, sender Dewi Anggraini
    3. 10:02:09BankMatched to an incoming transfer of Rp 1,250,000
    4. 10:02:10OrdersOrder #2207 marked paid and sent to packing
    5. 10:02:11WhatsAppReceipt and delivery estimate sent to the customer
    6. 10:02:12AccountingPayment recorded against invoice INV-2207
    Result—

    Simulated run with fictional data.

    Starts when

    A customer sends proof of transfer on WhatsApp

    If something goes wrong

    • An order is never marked paid without a matching bank transfer
    • No match within 2 hours: Finance gets the screenshot and the order

    What we measure

    Minutes per confirmation, and orders marked paid by mistake

    CustomScoped after a free call · Quoted per scope
    Discuss this automation

  • Marketplace orders to stock, courier and accounting

    Orders from every marketplace and your website update one stock count, create the shipping label and land in your books.

    ShopeeStockWarehouseCourierAccounting

    Run log

    Step 0 of 6

    1. 13:40:02ShopeeOrder received: 2× Kopi Arabika 250g
    2. 13:40:03StockStock reserved, 14 left, synced to Tokopedia and website
    3. 13:40:05WarehouseAdded to today's picking list
    4. 13:40:07CourierShipping label and tracking number created
    5. 13:40:08AccountingSales invoice and marketplace fee recorded
    6. 13:40:08StockBelow safety stock of 15: reorder suggested
    Result—

    Simulated run with fictional data.

    Starts when

    A new order arrives on Tokopedia, Shopee or your website

    If something goes wrong

    • A failed sync is retried, then the listing is paused so you never oversell
    • Stock below the safety level alerts purchasing

    What we measure

    Overselling incidents, and admin hours per day

    CustomScoped after a free call · Quoted per scope
    Discuss this automation

  • Invoice reminders and collections

    Reminders go out politely and on time, get firmer as days pass and stop the moment the customer pays.

    AccountingEmailWhatsAppBank

    Run log

    Step 0 of 5

    1. Day −3AccountingINV-0912 for PT Mitra Jaya, Rp 12,500,000, due Friday
    2. Day −3EmailFriendly reminder with the payment link sent
    3. Day +3WhatsAppSecond reminder, account manager copied
    4. Day +5BankPayment received and matched to INV-0912
    5. Day +5AccountingInvoice closed, reminders stopped
    Result—

    Simulated run with fictional data.

    Starts when

    An invoice is 3 days from its due date

    If something goes wrong

    • A disputed invoice pauses its reminders
    • Key accounts get a call from a person instead of a message

    What we measure

    Days sales outstanding, and overdue amount

    Agent SkillLive in 1–2 weeks · From Rp 7.5 million
    Discuss this automation

  • Morning business report

    Sales, cash and stock from different systems in one short summary on WhatsApp, before the day starts.

    POSBankStockAIWhatsApp

    Run log

    Step 0 of 5

    1. 07:00:00POSYesterday's sales: Rp 18.4 million, 312 transactions
    2. 07:00:03BankCash balance across 3 accounts: Rp 246 million
    3. 07:00:05Stock6 items below minimum stock
    4. 07:00:09AISummary written: sales down 12% at the Bekasi branch
    5. 07:00:10WhatsAppReport sent to the owners' group
    Result—

    Simulated run with fictional data.

    Starts when

    Every day at 07:00

    If something goes wrong

    • If a source is missing, the report says so instead of showing a wrong number
    • Every figure links back to its system

    What we measure

    Hours spent compiling reports each week

    Agent SkillLive in 1–2 weeks · From Rp 7.5 million
    Discuss this automation

  • Purchase and expense approvals

    Requests reach the right approver by amount, get approved in one tap and chase themselves when they stall.

    FormRulesChatERP

    Run log

    Step 0 of 6

    1. 09:15FormRequest: 4 laptops, Rp 48 million, quote attached
    2. 09:15RulesAbove Rp 25 million: manager, then director
    3. 09:40ChatManager approved in one tap
    4. Next day 09:40ChatDirector reminded after 24 hours
    5. Next day 11:02ChatDirector approved
    6. Next day 11:02ERPPurchase order PO-5531 created and logged
    Result—

    Simulated run with fictional data.

    Starts when

    Someone submits a purchase request

    If something goes wrong

    • Nobody can approve their own request
    • Over budget: Finance must sign before it continues

    What we measure

    Time from request to approval

    CustomScoped after a free call · Quoted per scope
    Discuss this automation

  • Joiner and leaver accounts

    New staff get every account they need on day one, and leavers lose access on their last day.

    HRWorkspaceAppsITChat

    Run log

    Step 0 of 5

    1. Day −5HRNew joiner: Rina, Marketing, starts Monday
    2. Day −5WorkspaceEmail account and calendar created
    3. Day −5AppsAdded to marketing groups and tools by role
    4. Day −4ITLaptop request opened, admin rights need IT approval
    5. Day 1ChatWelcome message with the first-week checklist
    Result—

    Simulated run with fictional data.

    Starts when

    HR records a new joiner or a leaver

    If something goes wrong

    • Leavers lose every access on their last day, with a checklist
    • Every change is logged for audit

    What we measure

    Time to full access, and accounts left open after exit

    CustomScoped after a free call · Quoted per scope
    Discuss this automation

  • Leads routed and followed up

    Every lead from ads, forms or WhatsApp lands in the CRM once, goes to the right rep and is never forgotten.

    WebsiteAICRMWhatsApp

    Run log

    Step 0 of 5

    1. 16:20:01WebsiteForm: catering for 300 staff, Cikarang
    2. 16:20:04AICompany found, fit score 82, estimated value Rp 90 million/month
    3. 16:20:05CRMNo duplicate, assigned to the East Bekasi rep
    4. 16:20:06WhatsAppWelcome message sent, rep notified
    5. 18:20:06CRMNot contacted in 2 hours: sales manager reminded
    Result—

    Simulated run with fictional data.

    Starts when

    A new lead comes in from an ad, form or WhatsApp

    If something goes wrong

    • Duplicates are merged, so two reps never chase one lead
    • Untouched leads escalate to the sales manager

    What we measure

    Response time to new leads, and lead-to-deal rate

    Agent SkillLive in 1–2 weeks · From Rp 7.5 million
    Discuss this automation

  • Email inbox to the right system

    Attachments sent to your shared inboxes are sorted, read and entered, and only the unclear ones reach a person.

    EmailAIERPReview

    Run log

    Step 0 of 5

    1. 08:05:00Email12 new emails with attachments
    2. 08:05:20AISorted: 7 invoices, 4 orders, 1 contract
    3. 08:05:48AIKey fields read from each document
    4. 08:06:02ERP11 documents entered and filed
    5. 08:06:02Review1 invoice with an unreadable total sent to a person
    Result—

    Simulated run with fictional data.

    Starts when

    An email with an attachment reaches invoices@ or orders@

    If something goes wrong

    • Anything the AI is unsure about goes to a review queue, never straight into your books
    • Original emails and files are kept and linked

    What we measure

    Share of documents handled automatically

    Agent SkillLive in 1–2 weeks · From Rp 7.5 million
    Discuss this automation

04Roadmap

From first assessment to AI in every division

Each stage ends with something live and measured, so you decide the next step on evidence.

  1. 1

    Weeks 1–4

    Assess & quick wins

    We interview every division, score your AI maturity and rank use cases by impact and effort. The first Agent Skills go live.

    You get

    • Maturity score per division
    • Ranked use-case roadmap
    • 2–3 Agent Skills live
  2. 2

    Months 2–3

    Pilot

    The top two or three use cases run as pilots on your real data, measured against today's baseline.

    You get

    • Measured pilots
    • Go / no-go per use case
    • Cost per task
  3. 3

    Months 4–9

    Scale

    Proven pilots roll out across divisions, with champions, training and shared rules for data and approvals.

    You get

    • Solutions live in several divisions
    • AI catalogue and request flow
    • Trained champions
  4. 4

    Month 10 onwards

    Embed

    AI becomes part of how every division works. We run, monitor and upgrade it, and leadership reviews the impact monthly.

    You get

    • Monthly impact and cost review
    • Model upgrades
    • New use cases each quarter

05Change management

Technology is the easy part. People make it stick.

AI fails when teams don't trust it or don't know it exists. We plan for adoption from day one.

  • 01

    Champions in every division

    One owner per division who knows the work, tests new AI first and brings their team along.

  • 02

    Training by role

    Short, practical sessions for leaders, daily users and the people who will build with AI.

  • 03

    One AI catalogue

    Every AI system in one place: what it does, who uses it and how to request a new one.

  • 04

    Clear guardrails

    Rules for company data, human approval wherever money or customers are involved, and a cost limit per system.

We already run this for an enterprise group

Nearly 200 AI ideas from more than 50 divisions, mapped by process and triaged into: build it, use an agent, or use a tool you already have. Leadership sees coverage, progress and impact on one page.

enterprise AI ideas mapped and triaged
~200
enterprise AI ideas mapped and triaged
divisions covered by one AI roadmap
50+
divisions covered by one AI roadmap

06Measuring impact

If we can't measure it, we don't scale it

Every use case gets a baseline before we start and the same measurement after, so leadership sees real numbers, not anecdotes.

  • Hours saved

    Staff time per month, compared with the baseline.

  • Cycle time

    From request to done, such as claim to decision.

  • Error rate

    Mistakes caught or avoided, checked against the source.

  • Cost per task

    AI and running cost for every document or request.

  • Adoption

    Who actually uses it, every week.

How we measure

  1. 1

    Baseline

    Measure today's process before any AI.

  2. 2

    Pilot

    Same measures on real data with real users.

  3. 3

    Live

    Tracked monthly on an impact dashboard.

Example: hours spent per month on one process
  • 100Baseline
  • 55Pilot
  • 35Month 3

Illustration of the measurement, not a client result.

Start with an AI assessment

In 2–4 weeks you get a maturity score per division, a ranked roadmap and your first Agent Skills live.

Assessment from Rp 75 million

Plan an AI assessmentBook a free 30-minute call