Replacing Excel spreadsheets with web applications
Why businesses replace spreadsheet processes with web applications, updated for the AI era - and what the data those spreadsheets hold is now worth.
This article first appeared in 2018 and has been updated for 2026.
It usually starts innocently enough - someone needs to track jobs, or quotes, or stock, they open Excel, and an hour later the business has a new tool. Spreadsheets must rank highly on any list of the most important inventions of the computer age - since VisiCalc appeared on the Apple II in 1979 they have made business possible, and they remain the fastest way to get from "we need to keep track of this" to something that works.
The trouble comes later, when the business has quietly come to depend on it. The spreadsheet lives on one person's machine, copies circulate by email with filenames like final_v2_ACTUAL.xlsx, there are no permissions and no record of who changed what, and the formulas encode business rules that are written down nowhere else. By the time the person who built it moves on, a critical process is running on a file nobody fully understands.
Google Sheets took a lot of that pain away, to be fair. One copy, everyone in it at once, a version history to unpick the disasters - and for plenty of processes that really is enough. We run a good chunk of Si Novi on Sheets to this day. But a shared spreadsheet is still a spreadsheet: anyone can sort a column and quietly scramble the lot, the permissions are all-or-nothing, the business rules are still buried in formulas, and it still doesn't talk to anything else the business runs on.
The web application answer
Back in 2018, when I first wrote this article, businesses were coming to us with spreadsheets they had outgrown and asking for web applications to replace them - a steady stream of the same story in different industries. The case for it hasn't aged:
- Users log in, so you control who sees what, give different people different permission levels, and keep a record of every change.
- The process connects to the rest of the business - accounting, CRM, your suppliers' systems - through APIs rather than copy and paste.
- Data entry is structured and validated - proper fields with rules, rather than freeform cells a stray keystroke can quietly corrupt.
- The system can grow with you - processes inevitably change, and an application can be improved iteration by iteration as they do.
Building isn't always the answer, and in 2026 the honest first question is whether an off-the-shelf product already fits. Low-code tools like Airtable and the small monthly-subscription SaaS products that now exist for every common process cover a lot of ground. The custom build conversation usually starts with someone telling us "we tried X, but it doesn't quite do what we need" - the process is genuinely theirs, the integrations need to run deep, or the off-the-shelf data model doesn't match how the business actually works.
We eat our own dog food on this - we recently replaced the Google Sheet that had served as Si Novi's CRM for years with a web-based CRM we built ourselves, for exactly the reasons on that list.
These projects also come with a head start, because the specification already exists. The spreadsheet is the prototype: its columns are the data model, its formulas are the business rules, and building the application is a matter of reverse engineering something the business has already proven it needs. The formulas are the easy half, mind - the hand-edited cells, the overrides and the workarounds that never made it into a formula are where the careful work is, because they're the rules nobody wrote down.
How AI changes the case for replacing spreadsheets
That answer hasn't been retired - businesses still outgrow spreadsheets, and a web application is still what they graduate to. What's changed is everything around it. The infrastructure is better and cheaper: serverless and managed cloud services mean a small internal application no longer needs a server to run on or anyone to patch it, and it can cost pennies a month to operate. Running these systems stopped being the expensive part some time ago - the real investment is the build itself, and sizing that honestly against what the process is worth is where any project should start. I've written more about that shift in digital transformation in the AI era.
The bigger change is what the data itself is now worth, because years of quotes, jobs, prices, customers and decisions - whether they made it into a web application or are still sitting in Excel - are exactly the context a large language model needs to be useful. A spreadsheet a business has maintained for a decade isn't just a process waiting to be replaced; it's a record of how that business actually works, and AI systems can now put that record to work. The fair question that raises is where the data goes - whose model reads it, and what leaves the building. These systems can and should be built so the models run under your own accounts, with nothing retained or used to train anyone else's product, and I'd treat that as a requirement rather than a nice-to-have.
Where AI fits in a business process
For a typical SME process in 2026, these are the patterns I'd be weighing up:
- Asking questions in plain English. "Which customers haven't ordered since March?" answered directly from the data, with no report-building or formula-writing in between.
- Getting documents into the system. Invoices, purchase orders and delivery notes read by a model and turned into structured records, replacing the retyping that eats admin hours.
- Reconciliation and checking. Comparing records across systems - orders against invoices, bank lines against ledgers - and flagging the handful that don't line up, so a person reviews the exceptions and a sample of the matches rather than every row.
- Agents for multi-step admin. The chase-up flows - request the missing timesheet, update the record, notify the manager - carried out by an agent, with a person approving anything customer-facing or financial.
- Capturing the knowledge nobody wrote down. An agent working through the documents and files a business has accumulated and assembling them into a knowledge base that can be searched and questioned - including, done openly and with them involved, capturing what someone working their notice knows before it leaves with them.
Si Novi's own bookkeeping already runs this way: scripts handle the deterministic work of pulling bills, invoices and bank transactions from Xero, an AI agent does the judgement work - matching a foreign-currency bill to the sterling amount that actually hit the bank, grouping invoices into the project they belong to, spotting the reconciliation that doesn't add up - and a human reviews before anything is written back.
None of this needs a business to be especially "AI-ready". The requirement is the one it always was: data that is structured and accessible, which is precisely what replacing a spreadsheet with a proper application gives you. There's more on where these tools genuinely earn their keep in AI automation for business.
Choosing where to start
If a spreadsheet is creaking somewhere in your business, the place to start is the process itself: what it does, who touches it, and what the data it holds would make possible once it lives somewhere better than a workbook. The technology choices - buy, build, AI in the loop, or some of each - follow from that understanding, though rarely without trade-offs worth arguing through properly. It's a conversation I've been having with businesses since 2018, and I'm always happy to have it - get in touch.