AI-first company

Data and tools · Explore

Start with ready-made AI in software you already use, before building your own

Your office, accounting, support and sales software probably has AI features now. Try those on one task, with a few licences and a measured result, before you pay anyone to build something custom.

By
Last reviewed
Corrections
Report an error
When Do first. Before you start with AI, or as you start.
Effort Low (days, one person)
Evidence Common practice

Widely recommended for firms without technical staff. One measured pilot, two randomised evaluations of an office assistant and one small-company case show what it looks like. No study compares buying with building.

Fits
Industries All industries
In the guide Micro and small businessesTools and systems by size

What it is

Start with the AI already inside software you pay for. There are three ways to get AI into a company, from cheapest to dearest:

  1. Switch on what you have. AI features inside software you already pay for: email and documents, accounting, customer support, sales records.
  2. Buy a ready-made tool for one task, such as a chatbot for customer questions.
  3. Build your own, or pay someone to.

Start at 1. Move to 2 when 1 does not cover the task. Consider 3 only when a bought tool has shown the task is worth it.

What the evidence shows

One measured pilot, two randomised evaluations and one small-company case are the basis for this advice. No study has compared buying with building.

  • A trial inside existing software is quick to run and to judge. A UK government department gave 1,000 staff an AI assistant inside the office software they already used. In three months the department knew which tasks gained and which lost. Summarizing reports took 13 minutes with the assistant and 42 without, with better results. Spreadsheet analysis was slower and worse.
  • The pilot also showed the limits early. 72% of users were satisfied, but the department found no evidence of higher productivity. The department learned that from 1,000 licences and three months.
  • Two randomised evaluations show what an office assistant changes. In a field experiment with 7,137 workers at 66 firms, workers randomly given Microsoft 365 Copilot spent about two hours less on email a week, and the kind of work they did stayed the same. Three of the four authors worked for Microsoft. HMRC, the UK tax authority, gave 3,000 licences to staff chosen at random. 83% used them. After the evaluators cut the self-reported figures by about 20% for bias, staff reported saving about 60 minutes a week. Among staff who did not use the tool, 46% cited security and data concerns.
  • A typical small firm pays little. Among small US firms in Chase’s bank records that paid for AI, the median spend was about $28 a month in 2025, according to the JPMorganChase Institute.
  • A small firm got a result with a bought tool. An online shop in Singapore bought a ready-made chatbot that works in WhatsApp. Its founder estimates the team spends 30% less time answering customer questions.
  • Building does not remove the risks. A large US law firm built its own AI tool. The tool made up eight court cases in one filing.

How to do it

  1. List the software you pay for. For each product, check whether it now has AI features, whether your plan includes them, and which are switched on by default.
  2. Pick one task that is frequent and takes real time: summarizing long documents, drafting routine replies, writing up meeting notes. See start with one costly process.
  3. Take a few licences for a short period. Five to ten people for one to three months is enough at most firms. Mix keen and sceptical staff.
  4. Check the data terms before anyone starts. See decide what data may go into which AI tools.
  5. Time the task with and without. Have someone who knows the work judge the quality of both. Do not rely on whether people like the tool.
  6. Decide at the end date. Keep the feature for the tasks where it clearly helped. Drop it for the rest.
  7. Move to a separate tool only for a task the built-in feature cannot do. Then ask the vendor for proof.
  8. Consider building only when a bought tool has shown the task is worth it, no product fits, and you can name who will maintain what you build.

What to measure

  • Time per task with and without the feature, on the same kind of work.
  • Quality, scored by someone who does not know which version used AI.
  • The share of licence holders still using it after the first month.
  • Licence cost per month against the time saved.

Where it fails

  • Licences for everyone on day one. You pay for seats before you know which tasks gain.
  • Judging by satisfaction. In the UK pilot most users were satisfied, and the timed tasks told a more mixed story.
  • Assuming it helps every task. The same assistant was three times faster at summarizing and slower at spreadsheet work.
  • The supplier sets the price. Google put its Gemini AI into all Workspace business plans and raised Business Standard by $2 per user per month. Cursor, a coding tool, changed its plan to usage-based charges and later refunded surprise charges. Australia’s competition regulator alleges that Microsoft raised its Personal and Family plans by 45% and 29% when it added Copilot, and hid the cheaper plan without it. Check the price at renewal, and whether a plan without AI still exists.
  • AI you did not switch on. Small firms told a Federal Reserve survey that some AI was “forced upon us” by software they already use. One firm said its bookkeeping software’s AI feature could not be disabled. These are firms’ own words, from 2024 survey answers.
  • The vendor does not carry the risk. Buying ready-made does not move responsibility to the vendor, and your insurance may exclude AI losses. See know the rules that apply to your AI use.
  • Skipping the training. A feature nobody was shown how to use will sit unused. See train staff on the tools you chose.

By company size

  • Micro and small: steps 1 to 6 with two or three people. Do not build. If you sell on a platform, the AI that matters most may be the platform’s. In randomised experiments on a large online retail platform, AI the platform built into its own tools raised sales by up to 16.3%, and the sellers did nothing. Two of the authors consulted for the platform and one works for it.
  • Medium: a pilot group of five to ten, a named owner, and a decision date.
  • Mid-market: the same pilot per department. Building becomes an option for a task that is central to the business and proven with a bought tool first.

Company cases

Companies that did this

Sources

  1. Dillon, Jaffe, Immorlica and Stanton (2025). Shifting Work Patterns with Generative AI. NBER Working Paper 33795; arXiv 2504.11436, version 4 (November 2025). What this study did and found
  2. Wheat, Mac and Passalacqua (JPMorganChase Institute) (2026). Understanding the use of AI among small businesses. JPMorganChase Institute, published 14 April 2026. What this study did and found
  3. Fang, Yuan, Zhang, Donati and Sarvary (2025). Generative AI and Sales Productivity: Field Experiments in Online Retail. arXiv 2510.12049, version 6 (29 June 2026). What this study did and found
  4. UK Government Digital Service (2 June 2025). Microsoft 365 Copilot experiment: cross-government findings report. gov.uk
  5. UK Department for Business and Trade (August 2025). The evaluation of the M365 Copilot pilot in the Department for Business and Trade. assets.publishing.service.gov.uk
  6. HM Revenue & Customs (2026). Evaluating the impact of Microsoft Copilot in HMRC (Phase III evaluation report). gov.uk
  7. Sanchez-Moyano, Holmes and Simms (2026). AI Adoption Among Small Businesses: Qualitative Insights from the Small Business Credit Survey. Federal Reserve Bank of San Francisco, Community Development Research Brief 2026-01. frbsf.org
  8. Google Workspace blog. Announcement that Gemini is included in all Workspace business plans, with the new prices. workspace.google.com
  9. Cursor (June 2025). Company blog post on its June 2025 pricing change. cursor.com
  10. Australian Competition and Consumer Commission. Microsoft in court for allegedly misleading millions of Australians over Microsoft 365 subscriptions. Media release; allegations. accc.gov.au

Claims on this topic, checked against these studies: