One of the most common questions Accupe users ask in the first week is straightforward: which AI mode should I use, and when? Fast, Planning, and Ultra-Detailed are not arbitrary tiers - they are calibrated to three genuinely different categories of professional task. Using the wrong mode wastes time at one end and produces shallow answers at the other.
This is a practical guide to choosing correctly, with examples a UK accountant will recognise.
Fast Mode: lookup, extraction, and "tell me the number"
Fast Mode is the workhorse. It is calibrated for questions where the answer is genuinely in the document, the question is unambiguous, and you simply need it surfaced. Asking "What is the registered office address on this engagement letter?" or "What was the closing inventory balance in the 2024 accounts?" is a Fast Mode question. The model reads, extracts, and cites - usually in under three seconds. A senior associate using Fast Mode for routine lookups will save twenty to thirty hours a month.
- Pulling specific numbers out of a set of management accounts
- Finding the termination clause in a 40-page contract
- Confirming a director's appointment date from a filed CS01
- Checking whether a particular VAT period is in scope of an uploaded return
- Extracting bank account details from a signed mandate
Planning Mode: when the answer requires reasoning across the document
Planning Mode is for questions that require the AI to think in steps. Not just "what is the figure?" but "given the figures, what is the implication?" The model produces a visible chain of reasoning - read this, then this, therefore that - which a senior reviewer can audit before relying on the conclusion. This matters because professional advice requires the reviewer to understand the working, not just the answer.
A typical Planning Mode question: "Looking at the cash-flow statement and the loan covenants in the attached facility agreement, is the company at risk of breaching its DSCR covenant in Q4?" The model will trace the EBITDA line, locate the covenant definition, compute the ratio, and explain the buffer. The reviewer reads the working, satisfies themselves, and signs off.
Ultra-Detailed Mode: advisory depth and nuanced interpretation
Ultra-Detailed Mode is the heaviest of the three. It is intended for advisory work where nuance matters and where the client is paying for a considered opinion, not a quick answer. The model spends meaningfully longer thinking, draws on more of the document, surfaces edge cases, and structures its response with the kind of caveats a senior advisor would write themselves. Ultra-Detailed is the right mode for preparing a partner brief before a strategic client meeting, for drafting the technical reasoning behind an R&D claim, or for analysing the tax implications of a proposed group restructure.
A worked example: the same question in three modes
Take a single, real question: "What are the main risks for the client in the attached shareholders' agreement?" In Fast Mode, the model returns a bulleted list of clauses that contain the word "risk" or "liability" with citations. Useful for a quick pre-read, but not deep. In Planning Mode, the model identifies the clauses, then walks through why each one is a risk in the context of the agreement's structure - drag-along, tag-along, deadlock, pre-emption. The reviewer sees the logic. In Ultra-Detailed Mode, the model produces a structured memo: ranked risks, likelihood, mitigation strategies, and references to common market practice - ready for a partner to refine and send to the client.
When to escalate up the modes
A reliable habit: start in Fast, escalate to Planning if the answer is incomplete or surprising, and reserve Ultra-Detailed for partner-level deliverables. Trying to start in Ultra-Detailed for a routine lookup is wasteful - both of the model's time and the user's attention. Trying to use Fast for a partner brief produces something that reads thin and may miss exactly the nuance that justified the fee.
Pairing modes with docs-only safety
All three modes can be combined with docs-only mode, where the model is constrained to answer only from the documents the firm has uploaded. For client-facing or regulated work, this combination is the default. Planning Mode with docs-only is the sweet spot for most senior associate work: deep enough to be useful, constrained enough to be trustworthy.
Mode selection becomes muscle memory
New users typically take about a fortnight to develop instinct on which mode fits which task. Firms that introduce Accupe with a short internal "AI modes" cheat sheet on the wall report the learning curve flattens within a week. Once it does, the AI becomes an extension of the team rather than a separate tool to be remembered.
What changes for the firm
Firms that use the three modes deliberately report two outcomes. First, time spent on lookup-style tasks collapses by 60-80 percent, because Fast Mode genuinely replaces the act of skimming a document. Second, the quality of partner deliverables rises, because Ultra-Detailed produces a stronger starting draft than a junior associate would. The senior time saved at the bottom funds more senior thinking at the top.
Closing
Three modes is not a marketing tier; it is an acknowledgment that professional questions have wildly different depths. Accupe gives the firm the right tool for each depth and trusts the user to choose. The practice-management layer surfaces the work; the firm still owns the judgment.