Skip to main content

Answer2 min read

Do I need AI or a spreadsheet?

If the work is arithmetic on data you already have in one place, a spreadsheet. If the work is reading, deciding, or moving information between systems, that is where AI and automation pay, and a spreadsheet was only ever the place the re-typing landed. Most businesses need the spreadsheet to keep doing its job and something else to stop feeding it by hand.

The test

Ask one question about the task: where does the information come from before it reaches the sheet?

  • If someone types it in from an email, a PDF, a call, or another system, the spreadsheet is not the problem. The re-typing is. That is an automation, and sometimes an AI step to read the source.
  • If the information is already there and the job is to add it up, compare it, or chart it, the spreadsheet is the right tool and will stay the right tool.
  • If the job is to decide something about each row (is this lead worth a call, is this contract clause acceptable, which of these invoices to chase first), that is judgment, and judgment is where an AI step earns its keep.

What the re-typing costs

The numbers here are other people's, and they are consistent.

  • Sales reps spend 17% of their week on CRM entry and about 30% of their time actually selling (Salesforce and Forrester, a survey of 3,031 reps).
  • Workers spend 1.8 hours a day, about 9.3 hours a week, looking for information they already have somewhere (McKinsey).
  • Processing one invoice by hand costs $9.40 on average against $2.78 at the firms that automated it; 9.2 days to 3.1 (Ardent Partners, 212 firms).
  • CRM data decays at roughly 34% a year when nobody feeds it, and nearly half of companies say more than a tenth of revenue is lost to bad data (Fullcast).

None of those is a spreadsheet problem. Every one of them is a "the data arrives somewhere else and a person carries it" problem.

Where AI is the wrong answer

Three cases we see often.

  1. The rule is simple and never changes. "Thirty days late, send the reminder" wants an automation, not a model. A model adds cost and a small chance of being wrong to a job a rule does perfectly.
  2. The data is small and lives in one place. Twelve clients in a sheet you update monthly do not need a system. They need the sheet.
  3. The demo is the product. A chatbot that answers questions about your own documents demos well and, on independent tests, resolves around 38% of real support tickets in year one against the 67 to 76% vendors quote. It can still be worth it. It is not magic.

The usual shape of the answer

Keep the spreadsheet. Stop feeding it by hand. The systems we build most often are exactly that: the eight recipes are all "something arrives, something reads it, the right system is updated," and the spreadsheet, where it survives, becomes a report rather than a job.