AutoCount + AI: the layer the AutoCount Plugin Portal will not write

AutoCount has a Plugin Portal. It is well-built, it is documented, and the developers who ship through it are mostly the same developers who would build a custom add-on for a Penang dealer ten years ago. The Portal is a good marketplace for accounting-shaped extensions — extra reports, OCR for invoices, industry-specific chart-of-accounts templates. It is not the AI layer. Here is what the AI layer looks like, and why the difference matters.

What "AI on AutoCount" actually means

AutoCount is a suite. AutoCount Accounting is the ledger. AutoCount Stock is the inventory. AutoCount Payroll is the HR. The Plugin Portal is where third-party developers add modules that sit next to those — a GST helper, a manufacturing add-on, a CRM that knows what the AutoCount debtor master says. Every one of those is a feature, in the traditional sense: a button, a screen, a report the user has to open.

The AI layer is a different category. It is not a button. It is an assistant that reads the whole AutoCount database, answers questions in plain language, and writes the answers into the channels your customers already use — WhatsApp, email, Messenger. The Plugin Portal will not ship it because the AI is not a feature. The AI is the front door to every feature.

Three concrete use cases

1 · The Penang dealer who loses 3 hours a day to "any update?"

David runs an AutoCount dealership in Penang. He has 80 active B2B customers on his AutoCount AR. Every one of them has a WhatsApp. Every one of them, at some point in the working day, sends "any update?" — about an invoice, a delivery, a back-ordered part. David's two admin staff split the WhatsApp inbox between them. They spend three hours a day answering questions the AutoCount AR could answer in three seconds, if the AR could see the WhatsApp.

With the AI layer, the WhatsApp inbox is the assistant's inbox. A customer sends "any update on my order PO-2025-08812?" The AI reads the AutoCount purchase order, sees the goods-received note from last Thursday, sees the invoice raised on Friday, and replies: "Hi Encik Lim, your order PO-2025-08812 was fully received on Thursday and invoiced as INV-2025-09234 for RM8,420. Payment terms are 30 days. Let us know if you need a copy of the invoice." David's staff see the conversation, the customer is happy, the three hours a day are back.

2 · The chain of three retail shops in KL

Linda runs a chain of three retail shops in KL, all on the same AutoCount Stock. Every evening, the manager at each shop closes the till, exports the day's sales to a CSV, and emails it to head office. Head office imports the CSVs into AutoCount, reconciles the bank, and the owner gets a sales report the next morning. The whole ritual takes 90 minutes. By the time the report is in Linda's hand, the day is over.

The AI layer reads AutoCount Stock in real time. At 9pm, the AI writes a one-paragraph summary of the day's sales across all three shops — total revenue, top SKU, slow-moving stock, the one item that sold out at the Bangsar shop and needs replenishing from the Subang shop. The summary lands in Linda's WhatsApp at 9:15pm. She reads it in bed. The next morning, the manager at Bangsar has a list of three things to do. The 90 minutes of CSV import is gone.

3 · The logistics company in Shah Alam

Razif runs a logistics company in Shah Alam. He has 240 active customers on his AutoCount, and a fleet of 18 trucks. Every customer expects a delivery status update. Every delivery exception — a failed delivery, a wrong address, a customer not home — is a phone call from the customer that Razif's office has to answer by looking up the delivery in AutoCount, finding the driver's last location, and calling the driver back. The cycle takes 15 minutes per exception. There are 8 exceptions a day.

The AI layer reads AutoCount + the delivery system. When the driver marks a delivery as failed, the AI writes a WhatsApp to the customer: "Hi Puan Tan, our driver tried to deliver your order at 2:30pm but the address was closed. We will retry tomorrow between 9am and 12pm. Reply here if you need a different time." The customer replies, the AI updates the AutoCount delivery note, the next-day retry is scheduled. Razif's office stops being the call-back centre.

Why the AutoCount Plugin Portal does not (and will not) ship the AI

The Plugin Portal is a marketplace for AutoCount-shaped add-ons. The auto-summarise-the-day feature is a feature of the AI, not a feature of AutoCount. The WhatsApp-but-smarter feature is a feature of the channel, not a feature of the plugin. The Portal will keep shipping features that sit inside the AutoCount UI. The AI is the layer that lives above the UI. The two are complementary — your AutoCount add-ons still work, your AI still reads them — and the gap is the layer your business actually wants next.

“My customers do not log into AutoCount. They log into WhatsApp. The AI is the bridge that lets me serve the customer on the channel they actually use, without the staff having to copy numbers out of the system I already paid for.”

— David, who runs an AutoCount dealership in Penang

What a normal Tuesday looks like with the AI on

It is 7:42am on a Tuesday. The owner is at the counter, the staff are opening the shop, the WhatsApp is already pinging with overnight messages from customers. Before the AI, the morning was the same every day: open the ledger, check the bank balance, open WhatsApp, start answering. After the AI, the morning is one prompt and a coffee.

The owner opens the minidesk chat on the laptop. The AI reads the ledger, the bank feeds, the open WhatsApp threads, and replies in 4 seconds with a three-point summary: who paid overnight, who is past due, which customer follow-ups are queued for approval. The owner taps approve on the queued messages, the WhatsApp is up to date, the ledger is clean, the bank feed is reconciled. The morning is done by 7:46am.

That is the shape of a normal day. The bookkeeper opens her ledger dashboard once, around 10am, to do the human-shaped work — the reconciliation that needs judgement, the journal that needs a note, the month-end that needs a sign-off. The AI handles the rest: the WhatsApp, the reminders, the daily summary, the "any update?" reply. The owner handles the strategy. The bookkeeper handles the judgement. The AI handles the repetition. Everyone's day gets a little longer, and the customer gets a faster answer.

What it costs, in real money and in real time

The AI plan is a flat monthly fee — the number is on the pricing page, the quote is the quote, no per-message markup, no per-prompt surcharge, no enterprise tier. The translator for desktop ledgers is included. The OAuth for cloud ledgers is included. The WhatsApp Business API access is included. The Meta conversation fees are the same fees every platform pays, and we do not add a per-message markup on top. The only other line item is the AI usage, which is metered in the same units every modern AI tool uses, and which your bookkeeper will not notice unless she is doing something unusual.

The time saving is harder to put a number on, because the time is not a single block — it is a thousand two-minute savings across a month. The Monday morning phone tag that used to take three hours. The Friday afternoon invoice chase that used to take ninety minutes. The "any update?" reply that used to take fifteen minutes per customer. The end-of-day reconciliation that used to take an hour. The end-of-month report that used to take a day. The AI takes each one of those from minutes to seconds, and the sum is the difference between a business that is drowning in WhatsApp and a business that is using WhatsApp to grow.

What to do if you are already on a BSP

If you are already on a Business Solution Provider — WATI, SleekFlow, AiSensy, Twilio, MessageBird, or any of the long tail — the good news is the WhatsApp number and the API credentials are already in your business name, and Meta has already verified you. The migration is the part you are weighing, not the setup.

The practical path is: keep the BSP contract running for one billing cycle while you set up minidesk in parallel. Connect your ledger, install the translator or click the OAuth, and let the AI read your data. Authorise the Meta Embedded Signup for your existing WhatsApp number through the minidesk console — the number is yours, the verification is yours, the migration is a settings change at Meta, not a new application. Test the AI on a small set of conversations — say, the top 10 overdue customers — and compare the messages the AI drafts to the messages your BSP has been sending. When you are comfortable, switch the routing over and cancel the BSP at the next billing date.

The two parts that take the longest are the parallel-run (1–2 weeks of comparing the AI's messages to the BSP's) and the BSP cancellation (some BSPs require 30 days' notice, read the contract before you sign anything with us). The Meta-side change is a settings switch, not a re-verification. The number stays the same. The conversations stay the same. The customer does not notice the migration — and the AI is reading the same data your BSP never had access to.

Common mistakes when you set this up yourself

Three mistakes we see most often, all worth naming. (1) Connecting a personal WhatsApp number instead of a business number — Meta does not allow a personal number on the Business API, and the porting process is a separate project. (2) Trying to set this up before the chart of accounts is clean — the AI reads what is in the ledger, and a chart of accounts full of one-off accounts makes every answer longer than it needs to be. (3) Skipping the test phase — the first WhatsApp the AI sends should go to your own number, so you can see the layout, the language, the speed, before it goes to a customer. We have made all three of these mistakes ourselves; that is why the install wizard walks you through each one.

How to add the AI to your AutoCount

The install is the same shape as the SQL Account install, with one difference: AutoCount's underlying database is Firebird or SQL Server depending on which edition you run, and the translator handles both. Twenty minutes from signup to the first answer. The dealer-written AutoCount add-ons you already have keep working — the AI reads the AutoCount database, not the add-on layer, so there is no version conflict, no API rate limit, no migration.

1. 1Sign up at minidesk.co and pick AutoCount as your connector (the on-prem or cloud edition, your choice).
2. 2Install the translator on the PC that runs AutoCount. The translator speaks Firebird and SQL Server, and reads the same database the desktop client reads.
3. 3Connect your business WhatsApp number. minidesk holds the Meta verification, so the typical 2–4 week wait is gone.
4. 4Ask the AI a question in plain language. The first answer takes a minute because the translator is mapping your chart of accounts. After that, every answer is under three seconds.
5. 5Hand the AI a workflow ("text every customer whose delivery failed today") and it does it, shows you the queue, and asks before anything goes out.

AutoCount is the software your bookkeeper chose. The AI is the layer your customers will thank you for. The two do not have to be the same vendor to be the same system.

→ See the AutoCount integration: /integrations/autocount