All writing

Why your AI isn't paying off yet

Guide4 min read

Why your AI isn't paying off yet

Guide · November 20, 2026 · 4 min read

Most small and mid-sized businesses now pay for AI somewhere: a writing assistant, an AI add-on in the CRM, a chatbot on the website, a tool someone in sales found and expensed. Far fewer can say what any of it has returned.

They are not alone. In a 2026 study of more than 1,600 small and mid-sized businesses, carried out by IDC for SAS, nearly seven in ten were still at the experimental stage with AI, and fragmented data and tools were among the main reasons.

The problem is rarely the AI. It is that each tool can see only a slice of the business, and nothing connects the slices.

How it happens

Nobody plans it. Each product you already pay for adds an AI feature, and each new tool arrives with its own. The CRM's assistant knows your leads but not your invoices. The support chatbot knows your help pages but not what is in stock. The writing assistant knows nothing about your business at all.

So people fill the gaps by hand. They copy a customer's details from one system into another, paste figures into a spreadsheet for the AI to read, and check its answers against the system that actually holds the truth. The AI was meant to save that time.

The signs you are here

  1. Staff copy the same information between tools every day
  2. One customer exists in four systems, with four slightly different addresses
  3. The AI answers confidently and wrongly, because it cannot see the rest of the business
  4. You pay for AI features in several products that overlap
  5. Nobody can say what each tool costs in a year, or what it returns
  6. The reports the owners rely on are still assembled by hand

Why another tool will not fix it

AI is only as useful as what it can see. A model that can read your customers, jobs, orders and invoices together can answer real questions and do real work. The same model looking at one system is guessing at the rest.

Each new tool adds another island. It may be excellent on its own, and it still leaves someone carrying information between it and everything else.

Start with one foundation

The businesses getting value from AI have one place where the core records live: customers, jobs, orders, invoices, stock, whatever the business actually runs on. Each record lives once, named the way the business names it, with clear rules on who can see and change it.

That does not mean replacing everything. The accounting package can stay, and so can the tools people genuinely like. They connect to the foundation instead of each keeping their own copy.

Connect before you add

Before buying anything new, connect what you keep. Decide which system is the source of truth for each kind of record, and let the others read from it rather than be re-typed into.

This step is unglamorous, and it is often where the largest saving is. Removing the copying and re-keying frees hours every week before any AI is involved.

Then put AI where the work is

With the foundation in place, choose one process that costs the most hours: the inbox that has become a job, the quotes that take a day, the follow-ups nobody has time for. Put AI there, able to see across the business.

That is where an agent earns its keep: reading what comes in, checking it against your real records, doing the work in your systems, and leaving anything that matters for a person to approve. Measure the hours, errors and response times before and after. If the numbers do not move, change course early.

Audit what you pay for

An afternoon with a spreadsheet tells you most of what you need. List every tool, then answer four questions for each.

QuestionWhat it tells you
What does it cost in a year, with every seat?The true price, not the monthly headline
What information does it hold?Where your data is, and where it is duplicated
What does it connect to?Which tools are islands
Would anyone notice if it went?What to keep, what to connect, what to cancel

A ninety-day plan

  1. Weeks one and two. Audit every tool: cost, data, connections, and who relies on it.
  2. Weeks three and four. Choose the core records and decide where each will live.
  3. Month two. Connect the systems you are keeping, and stop the re-typing.
  4. Month three. Put AI on one process, with a person approving what matters, and measure it.

The businesses that will get the most from AI this year are not the ones with the most tools. They are the ones whose tools can see the whole business.

Start with one problem, or bring us the whole operation.