2026-11-17 · 6 min read
What changes when you run your ERP by talking to it
Most ERP demos show you a dashboard. Ours shows you a conversation. Here's what changed after we stopped clicking through menus and started typing sentences instead.
Every ERP demo looks the same: a sidebar, a dashboard, a form for every record. The pitch is always “everything in one place.” What it rarely mentions is how much of your day still goes to finding the right screen, filling the right form, and clicking through to the one number you actually needed.
We build ERPs for a living, so we had the same screens. Then we tried something different on our own system: instead of opening the dashboard, we started asking it questions. “What is pending today?” “Make a quotation for this client.” “Did we get paid on that invoice?” The ERP answers from live data, and the screen stays closed unless we want it open.
A working day, in practice
None of this is a concept demo. It is how the founder runs Tikram day to day, from ChatGPT, Claude, and Gemini, rarely opening the ERP interface at all.
“What is pending today?” pulls the company’s operating map and returns only the work that is actually unblocked right now, based on stored status and dependencies, not a guess about what probably matters.
“Make a quotation: setup fee plus a monthly plan” finds the client, builds the quotation with one-time and recurring items, lets the ERP compute every total, and hands back a PDF link ready to send.
“Record a payment against that invoice” checks the payment history for a duplicate first, shows a preview of exactly what will be recorded, and only writes it to the books after an explicit yes.
“Send them the invoice on WhatsApp” generates the document, shows the exact message, recipient, and attachment, and sends it through the WhatsApp Business API only once that preview is approved.
“What did this customer say yesterday?” reads back the actual WhatsApp conversation, because the chat lives on the same record as the quote and the invoice, not in a phone nobody else can search.
“How did last week’s campaign do?” reports delivery numbers for a WhatsApp campaign and lists exactly who did not receive, read, or reply, so follow-up goes to the right ten people instead of the whole list.
Why this is not reckless
Letting an AI assistant touch customer records and money sounds like the kind of thing that goes wrong quietly. The reason it has not is that reading and drafting are free, but acting on money, customers, or data is gated, by design, not by prompt.
Every connection signs in as a real ERP user, so the assistant can only see and do what that person’s role already allows in the dashboard. Nothing is shared across accounts, and nothing runs as a generic service identity.
Payments, WhatsApp sends, and permanent deletions all require a preview first: the exact numbers, the exact message, the exact recipient, shown before anything happens, then a deliberate approval. Generating a document is not the same as sending it. Nothing reaches a customer unless that is asked for separately.
Every one of those actions is logged the same way a change made in the dashboard would be, so it can be reviewed later, by anyone with the access to do so. Deactivate a person in the ERP and their AI access stops immediately, because every request is re-checked against the live account, not a token issued once and trusted forever.
Three assistants, one ERP
The same ERP connects to ChatGPT, Claude, and Gemini, through the Model Context Protocol, the open standard that lets an AI assistant call into a piece of software the way a person would use its API. That matters less for the novelty and more for what it avoids: picking one AI provider and building the business around it. Whichever assistant a team already prefers, on a laptop or a phone, is the one that runs the ERP.
It also means the work that used to require opening a tool now happens wherever the conversation already is. A founder checking what is pending before a call, a salesperson confirming whether a client paid, a manager asking what a campaign actually delivered: none of it needs a tab.
What this is not
It is not a replacement for the ERP interface. The dashboard still exists, still works, and some people will always prefer clicking through it, especially for anything visual: a finance report, a chat inbox, a mind map of the business. The conversation is an additional way in, not the only one.
It is also not a reason to skip the decisions a human should make. The assistant will tell you an invoice is overdue; it will not decide to write it off. It will draft the quotation; it will not decide the discount. That line is deliberate, and it is where a Tikram ERP keeps the people in charge of the business actually in charge of it.
If your ERP still needs a tab open all day
Most businesses we meet are not short on software. They are short on a version of it that fits how they actually spend a day: in WhatsApp, in email, now increasingly in an AI assistant. An ERP that answers a plain question instead of waiting to be clicked through is a smaller change than it sounds, and it is one we can show on your own business before you commit to anything.