The same people, with the same AI tool, got much better results on some tasks and worse results on another. The difference was not the tool. It was whether the work suited it.
The finding
Consultants at BCG worked on realistic tasks, some with AI and some without. On tasks that AI handles well, those using it finished 12.2% more tasks, worked 25.1% faster and produced work of more than 40% higher quality.
On a task outside AI’s strengths, the picture flipped. Consultants using AI were 19 percentage points less likely to reach the correct answer.
Both results come from skilled professionals. The lesson is not that AI is good or bad. It is that results depend on matching the tool to the task, and on knowing where that line sits.
What it means for small teams
For a small business, the message is simple: AI is not a switch you flip for everything. It is excellent at some jobs and unreliable at others. The skill is knowing which is which.
That is why generic tools often disappoint. They are set up for an average case, not for your work. An agent that drafts invoices from your own price list is one thing. An agent guessing at a question it was never given sources for is another.
Good setup means giving agents the right knowledge, clear boundaries and a point where a person takes over. It is not glamorous work, but it is the difference between hours saved and mistakes made.
What it means for larger companies
At scale, the risk is uneven quality you cannot see. Some teams get large gains. Others quietly get worse answers and trust them anyway, because the output looks confident.
The fix is design, not enthusiasm:
- Map tasks to what AI does well before rolling it out.
- Ground agents in company data, not general knowledge.
- Require human approval where mistakes are costly.
- Log every action, so errors can be found and fixed.
- Teach people to recognise when an answer needs checking.
How we apply it
Engineers design every system. In the Discover step, we study your processes, tools and data, and sort the work into three groups: what agents can own, what they can draft for your approval, and what stays with your people.
Agents answer from sources. We ground them in your knowledge with Agentic RAG, Cache-Augmented Generation and knowledge graphs, so they work from your documents rather than guesswork.
People stay in the loop. Important actions need your approval, and every action is logged. When a task falls outside what an agent does well, it goes to a person.
We keep measuring. Your ROI dashboard tracks hours saved, cost per task and accuracy every week, and we adjust the setup as your business changes. We’re on hand 7 days a week.
Tell us how your business runs. We’ll show you where AI saves the most time.