Tax law moves daily, new legislation, rulings, treaties and circulars. Their senior advisers spent hours in legal databases, cross-checking sources and working through long judgments before they could write a single opinion. It was expensive, it ate the partners' time, and clients who needed an answer before closing a deal were left waiting days.
Building an opinion meant reading, thousands of pages of regulation and hunting by hand for the relevant precedents, before an adviser could even start to reason. The knowledge existed; the time to assemble it was the bottleneck.
Cross-border work multiplied it. Lining up different regimes at once, say Israeli corporate tax against the cross-border reporting rules that apply in the United States, meant holding several bodies of law in view for a single transaction, and getting any of them wrong was not an option.
So opinions became the constraint. A client structuring a complex deal waited days for an answer, which slowed the deal down, at exactly the moments when a fast, defensible view was worth the most.
Ground the research in the firm's own knowledge, cite everything, and keep the opinion a human's.
We built a research copilot over a closed, private retrieval layer holding all the relevant legislation, court rulings, advance rulings and the firm's own library of past opinions, so it reasons from the firm's authoritative material, not the open internet.
An adviser can type a question like the tax consequences of a dividend from a UK subsidiary to an Israeli parent under the current treaty, and get a reasoned answer in seconds, complete with precise citations and links to the specific rulings behind it.
Guardrails keep the copilot to its sources: every answer carries its references, so an adviser can verify each point against the ruling it came from rather than trust an unsupported statement.
The copilot helps build the initial skeleton of an opinion or an advance-ruling request from the research, in precise professional wording, which the senior partner then refines, rather than starting from a blank page.
Because it scans the most up-to-date material, the fear of missing an obscure judgment or a last-minute regulatory update goes away, the copilot is always working from the current state of the law.
The senior partner reviews, refines and signs every opinion. The copilot compresses the research and the first draft; the professional judgement and the accountability stay with the person whose name is on it.
The research collapses; the judgement stays with the partner.
The time to research even the most complex opinions fell more than 50%, which let the firm give clients a fast, competitive answer where it used to take days, and gave the advisers full confidence they had not missed a recent ruling or update.
The partners' expensive time shifted from basic information-gathering to strategic planning and real tax-risk work for clients, which improved the firm's profitability while raising the quality of the advice it delivered.
A general chatbot is a liability in tax: it invents, and it cannot be current. The value here is a copilot grounded only in the firm's authoritative sources, citing every claim so it can be checked, and a partner who signs the opinion. It compresses the research without ever owning the judgement. It is the same grounded-retrieval discipline behind our contact-centre assist work, aimed at professional tax research rather than a support desk.
Agents dug through five systems mid-call while customers waited. A grounded on-screen assistant surfaced the answer in context, cutting in-call search 42% and resolution time 30%, and lifting CSAT from 56% to about 77%.
Read the case → Reporting automation · AccountingAn accounting firm assembled every client's monthly report by hand from email and WhatsApp intake. AI built each report from the scattered sources with a human sign-off, turning days of assembly into minutes across dozens of clients.
Read the case →
Reef TRH