think assets.

Design the work, not just the prompt.

How evidence, AI and professional judgement are put together decides what a firm gets from AI.

By Think Assets

Start with a decision worth improving

Professional work is more than a question and an answer. People gather evidence, weigh competing accounts, decide what matters and explain the conclusion. Adding AI changes that arrangement, whether or not anyone designs the change. The design question is what the AI should contribute, how its contribution reaches the expert, and what the expert needs in order to use it well.

Match what AI can do to the work

Begin with one bounded task: an earnings adjustment in due diligence, or matching a company's spending to the tax relief it may qualify for. Set out the evidence, the judgements that carry consequences, and what preparation and review cost today. Then look for where AI adds something: reading more widely, comparing sources, exploring scenarios, or giving a clearer account of what is uncertain. A longer list of discrepancies is not the only improvement on offer.

Set out the way of working

A way of working connects the sources, the reading, the reasoning, the checks and the decisions people take. It might combine a search that knows which version of a document it is reading, a structured comparison, sums done outside the AI, and an AI that drafts alternative explanations. It also says what happens when the evidence is incomplete, or when two parts of the work disagree. The right combination follows the task, not the fashion.

Make the expert's part deliberate

In one task the expert should form a first view before seeing what the AI recommends. In another, reliable preparation shown straight away is the better design. Decide what the expert sees, when they see it, what they can challenge, and how their reasons reach the record that goes on file. A signature at the end does not show that the expert's judgement was put to good use.

Test the gain and the cost

Compare the new way of working with a fair version of the present one. Include routine cases where the AI's help is useful, and hard ones where it should be challenged. Look at the quality of the finished work, confidence in particular statements, what people relied on in practice, and the whole effort of review. An extra check can earn its place on an ambiguity that matters and add little on a routine reconciliation. The test shows where the benefit justifies the cost.

End with a decision the firm can act on

A useful recommendation says what to delegate, what stays with expert judgement, what evidence supports the choice, and what would prompt another look. The firm ends up holding a way of working its senior people chose, written down, which it can put in place with its own people or its technology supplier.

This is how Think Assets works, one expert workflow at a time, in four steps: Choose where AI matters, Design the way of working, Show what it is worth, and Keep it right. The tax worked example shows the split on a relief claim: what the AI does, what the two do together, and what the expert decides.

Start here

Get the pack for your field.

Finance, tax or real estate, or tell us your field. Ten pages. Pages 2 to 4 take five minutes.

Get the pack for your field

Thirty minutes with our founder, not a salesperson. Bring one workflow, with no client details, and we will sketch where AI could matter in it.

Arrange the 30 minutes