01 the business
Pest control company
40 staff, considering a new region
The owner wants to know whether to open in a neighbouring region. The research falls to whoever has a spare afternoon, and turns into a pile of bookmarks. The decision ends up being made on instinct.
02 before
The problem
When an owner wants to test a new service, region or price point, the research falls to whoever has a spare afternoon. It turns into a pile of bookmarked links, and the decision often gets made on gut feel because the scan never quite gets finished.
- Research is unfinished and scattered.
- Claims cannot be traced to sources.
- Decisions rest on anecdotes.
03 the build
What the AI does
- Step 1: Takes your question, such as a new suburb, a new product line or a competitor's price change, and agrees the scope with you first.
- Step 2: Searches public sources such as company websites, review sites, job ads, government statistics and industry news.
- Step 3: Pulls search demand and competitor visibility from Semrush or Ahrefs where you have an account.
- Step 4: Writes a short report on who is already there, what they charge, what their customers complain about and what is still unknown.
- Step 5: Cites every claim with a link and marks anything it could not verify.
paused: waiting for a person to approve
How it was put together
- Agreed the question and scope with the owner first.
- Searched public sources: competitor sites, reviews, job ads, government statistics and news.
- Pulled search demand and competitor visibility from SEO tools.
- Wrote a short report with every claim cited and unverified points marked.
04 human in the loop
Where a person stays in charge
The leader who asked reviews the report and decides what to act on. Anything that costs money or contacts outsiders, such as test ads or supplier calls, needs separate sign-off.
- Only public sources are used.
- Anything that costs money or contacts outsiders needs separate sign-off.
- Unverified claims are clearly marked.
05 rollout
From kickoff to live in 1–2 weeks per scan
- Day 1Agree the question, scope and decision it supports.
- Days 2–5Run the scan and draft the report.
- Week 2Review with the owner and decide next steps.
06 after
What changes
Leaders get a consistent, sourced first look at an opportunity before committing time or money. Ideas are dropped or pursued on evidence, and each scan is saved for the next time the question comes up.
Assume a scan would take a manager 16 hours to do properly by hand, against 2 hours to review a sourced draft.
16 − 2 = 14 hours saved per scan, which is often the difference between doing the research and skipping it.
What to measure
- Time from question to report.
- Share of claims with a source.
- Decisions made with a written scan.
- Follow-up questions answered.
07 evidence
What the research says
In a field experiment with 758 BCG consultants, those using GPT-4 on tasks within its abilities finished 12.2% more tasks, 25.1% faster and at over 40% higher quality, but did worse on a task outside its abilities.
A US federal judge fined two lawyers and their firm $5,000 for filing a brief that cited cases ChatGPT had invented.
ABA Journal, Lawyers who defended ChatGPT's fake cases must pay $5K, judge says (opens in a new tab) · 2023
08 lessons
Mistakes to avoid
- Vague questions. Scope first.
- Uncited claims. Every fact needs a link.
- Treating desk research as proof. Test small before committing.
09 questions
Common questions
- Can AI research be trusted?
- Only with sources. A study of BCG consultants found AI helped on tasks within its abilities but made people worse on a task outside them, so every claim is cited and checked.
- Can it scan international markets?
- Yes, using public sources in each country.