Ad Hoc Review
An ad hoc review is what an analyst does when a single point has them stuck: they turn to whichever peer is free, put a fragment of their work in front of that person and ask one bounded question. BABOK counts it among its seven review types and defines it in one sentence: the business analyst seeks informal review or assistance from a peer (§10.37.3.2). Of the seven it is the only one that runs with no planned session and with no prior reading by the reviewer, and those two absences settle everything else: it costs minutes, it happens every day and the corrected work product is the only trace it leaves. It is a collaboration reflex: it unblocks an author, while a work product's evidentiary weight is built by the reviews that log what they find.
Goal
The ad hoc review puts a fragment of work in front of a second reader who has never seen the draft. It is the cheapest way to find what an author left implicit. An author cannot see the gaps in their own text: tacit knowledge silently fills whatever was left out, and the author re-reads what they meant to write. It costs a few minutes, which is why it is practised in every team, whether or not a method made room for it.
BABOK itself draws the distinction the technique lives on (§10.37.2). For a completed work product, the review usually aims to remove defects or inform the reviewers about the content. For work in process, the review may be conducted to resolve an issue or question. The draft is the ad hoc review's ground. BABOK lists seven possible objectives for a review (§10.37.3.1). The ad hoc review serves two of them: to answer a question, resolve an issue or explore alternatives, which is its own territory, and to remove defects, in the narrow sense of the single defect the author could no longer see. The other five, conformance to specifications or standards, completeness and correctness, consensus on an approach or solution, educating reviewers about the work product, measuring work product quality, all presuppose preparation, a written record or a team. They belong to the reviews that have those things.
The only deliverable is the corrected work product. An ad hoc review leaves no defect log, no minutes, no follow-up action, and nothing afterwards establishes that it took place. Wiegers puts it bluntly: its effect rarely reaches beyond the immediate problem it solves.
Usage
When to use it
- Work in process: the draft need not be finished; the question is bounded.
- The author is stuck on an ambiguity: they no longer see it; another reader sees it in five minutes.
- One answerable question: "does this rule cover this case?" rather than "is this specification any good?"
- A competent peer is available: they know the domain and answer without rebuilding the context.
- No record is required: the finding is fixed on the spot, with no tracking and no report.
- Ahead of an expensive review: one glance puts the document in a fit state to be inspected.
- A drafting judgement call: two defensible wordings and the author wants a second opinion.
When not to use it
- A record is required (audit, quality gate, approval): an inspection (10.37a) logs the defects and verifies the rework.
- A complete work product needing systematic examination: a desk check (10.37e) brings in a reviewer who has read it beforehand.
- The question spans several perspectives (legal, security, business): a pass around (10.37f) collects independent opinions.
Description
BABOK gives this technique no steps. What separates a useful ad hoc review from a polite nod comes down to four choices, which the author makes before and during the exchange.
What decides the outcome
Bound the question before asking it
"Can you look at my specification?" invites a courtesy answer. "Does this cancellation rule cover a slot booked less than twenty-four hours in advance?" invites an answer. The technique's commonest failure sits right there: the author asks for approval while believing they are asking for an opinion, and approval is what they get.
Show a fragment that fits on one screen
The peer meets the document at the moment it is put in front of them, and they will get from it only what a first reading gives. Beyond one screen they will skim, and a skim that produces a nod costs more than no review at all, because the author now believes the text has been checked.
Choose a peer of equal rank
BABOK notes that a supervisor or manager sometimes takes part in a review for their expertise and that the moderator then takes care to preserve the candour of the other participants (§10.37.3.3). An ad hoc review runs between two people, with no moderator and no scribe: there is nobody there to take the precaution BABOK describes. A peer of equal rank says "this sentence reads two ways"; a superior reading a staff member's work is appraising a staff member, whatever their intention.
Say what becomes of the finding
If it is fixed on the spot, the exchange is over. If it has to be arbitrated by a third party, tracked or traced to closure, it has left the ad hoc review and belongs to item tracking (10.26). The author says it out loud before leaving: "that one I fix right now" or "that one I turn into an item". That sentence is what stops a finding from getting lost between two desks.
Where the ad hoc review sits
BABOK lists its seven review types in descending order of formality and calls the first three formal techniques, the next four informal techniques (§10.37.3.2). Two binary questions cut across that ordering. Is the session planned? Does the reviewer examine the work product alone before speaking? They sort the seven into four groups, and the ad hoc review is the only one answering no to both. Wiegers marks those two activities for six of the seven types (Table 3-1); the single issue review, absent from his spectrum, takes its two yeses from BABOK itself, which describes it as a careful individual examination ahead of a joint session. The double no delivers both at once: the ad hoc review's speed and the reach of what it attests to, which stops at the problem it has just solved.
The field's two reference standards go still further down the formality scale without ever meeting the ad hoc review: IEEE 1028-2008 recognises five types of review and audit, ISO/IEC 20246:2017 ten, down to the buddy check and the author's own author check. Neither standard names it. Wiegers, who does name it, leaves it out of the table that sets review objectives against review methods: it has no column there at all (Table 3-3). The technique exists, every team practises it and its value lies in what it unblocks. A work product's evidentiary weight is built elsewhere, with an inspection (10.37a).
The words "ad hoc" also designate something else entirely, and a reader coming from testing or software quality will hit the collision: ISO/IEC 20246 gives the name ad hoc to an individual reading technique, a reading with no checklist and no assigned perspective, used inside a formal review, inspection included. That is a way of reading; BABOK's ad hoc review is a type of review.
Four of the seven reviews are informal, and those are the ones people confuse. The line separating an informal walkthrough (10.37d) from an ad hoc review comes down to one point: was the session convened?
| Review | Who reviews | Planned | Individual preparation | Session | What is recorded |
|---|---|---|---|---|---|
| Informal walkthrough (10.37d) | The author's invitees | Yes | No | Yes | Scribe's notes |
| Desk check (10.37e) | One reviewer | No | Yes | Possibly | Written comments |
| Pass around (10.37f) | Several reviewers | No | Yes | Rarely | Collated comments |
| Ad hoc review (10.37g) | One peer | No | No | Impromptu | Nothing |
The traps
Taking it for quality assurance
"John had a look at it" settles into the project's memory in place of a review. BABOK describes the limit: informal reviews by one or two reviewers are practical in terms of the effort required, but they provide less assurance of removing all significant defects than using a larger team and more formal process (§10.37.4.2). Nothing was prepared, nothing was recorded, nothing was verified: the evidence that an independent review took place is built with a defect log and a verification of the rework, which is to say with an inspection (10.37a). This is the trap that does real damage, because it costs the team a review it believes it has held.
The nod
The peer settles for being helpful rather than being right, and that is the direct price of an exchange with no preparation, no checklist and no defined role. The remedy lies in the phrasing: ask the peer what they read in the sentence, instead of asking them whether they agree.
Running it constantly
An analyst who calls for twenty ad hoc reviews a week has outsourced their own thinking. The cost is counted in the peer's concentration, broken several times a day, and somebody else pays it.
Sliding into a desk check without noticing
The moment the author sends the document instead of showing it and the peer sets aside time to read it, a desk check (10.37e) is under way, with none of a desk check's discipline. Either own the desk check or keep the exchange short and synchronous. A distributed team keeps the technique: a screen-shared call, held on the spot, is an ad hoc review.
The author's own summary
The author reports what the peer said, in their own words and from memory. That summary becomes the review's only written trace, and it is the worst witness that review could have, because it is produced by the person whose work was under examination.
AI considerations
Of all the reviews, the ad hoc review is the one whose economics rest entirely on a peer's time, and a large language model is the closest thing there is to a permanently available peer whose interruption costs nobody any time. A bounded question put to a model about a fragment of a draft is an ad hoc review in every respect except the identity of the reviewer. The model is strong exactly where the author is blind: it reads the sentence as written. Ambiguity, a term used without definition, a case a rule leaves unanswered, a pronoun with two antecedents, these are its ordinary findings.
The organisational context
A model lacks the organisational context. It does not know that the counters close at 16:30, that the twenty-four-hour deadline is carried over from an existing regulation and that the question was already settled in March. Part of a peer's value is that they know the project. A model's answer is fluent and context-free, and that fluency is easily mistaken for knowledge of the context.
The right question
A model will not tell you the question is the wrong question. Where a peer asks why that rule is being specified, a model answers what it was asked.
The nature of the review
A model does not change the nature of the review. "I had the AI check it" sounds like diligence and produces what the technique produces: a corrected text and nothing that attests to it. No independent reviewer was involved, and the trap becomes harder to spot.
Data protection
A draft specification carries customer data, internal identifiers and unpublished business rules. Pasting it into a public model is a question of the revised Federal Act on Data Protection (revFADP) before it is a question of productivity, and the organisation's rule (which model, which data, hosted where) is settled before the exchange.
Examples
A second reader finds the case a rule does not cover and its author no longer sees. The author is stuck on a cancellation deadline, walks across the office and puts three lines in front of a colleague who has never read them.
| Rule, before | "Users may cancel their appointment up to 24 hours before the scheduled time. After that, the fee remains payable." |
|---|---|
| The colleague's question, at the desk, five minutes | "And if the appointment is booked in the morning for the afternoon? It would be non-cancellable from the moment it is made." |
| Rule, after | "Users may cancel their appointment up to 24 hours before the scheduled time. An appointment booked less than 24 hours in advance may be cancelled up to two hours after it was made and no later than the scheduled time. After those deadlines, the fee remains payable." |
The colleague was meeting the text for the first time, so she read what was written. The rule counted the deadline from the scheduled time and took no account of when the booking was made. She looked for the slot that makes the rule unworkable. Left outside the rule, the case would have surfaced at the counter, with a user contesting the fee. The author settled it on the spot: the correction was his alone to make. He rewrote the rule during the exchange, without turning it into an item.
There is no defect log, no minutes, no follow-up and no evidence that this exchange took place. The document history carries an edit by the author, under his own name, as if he had re-read his own text alone. Three weeks later nothing says that the rule was reviewed, by whom or on what question. The rule changed. That is the whole of what the technique produces.
Visualizations
The formality spectrum is a spatial object: seven types placed on an axis, two marks per cell, one cell alone in answering no to both. Rendered as a table it would lose what it teaches, namely the isolation of cell (g) and the reason for it. So it is drawn.
The comparison of the four informal reviews falls into rows and columns that the reader scans line by line: that is a table. The two binary questions are not enough: the desk check and the pass around answer them the same way. Three further columns separate them: who reviews, whether a session is held and what is recorded. The choice between the seven reviews (10.37) turns on what the review is for and on how much assurance it must give.
Cost
| Phase | Level | Justification |
|---|---|---|
| Preparation | Low | The author picks a peer and walks across the office. Of the seven review types it is the only one that runs with no preparation at all, on the author's side or the reviewer's. |
| Execution | Low | A few minutes, two people, one bounded question. The real expense is the peer's concentration, broken. It multiplies as soon as the practice becomes a habit. |
| Documentation | Low | Nothing is recorded. The saving on documentation and the absence of evidentiary weight are one fact seen from two sides: this "low" is paid for in assurance. A review that has to attest to something pays for its documentation. |
Tooling
The technique needs no tool, and reaching for one is generally the sign that you have left it. On site, you need the work product on a screen and a chair. Remotely, you need a call with a screen share.
A document sent in a message with "can you have a look?" is a desk check (10.37e) stripped of a desk check's discipline, and the author will wait a day to receive "looks fine to me". You move from one technique to the other without anyone deciding it.
The useful boundary is this: the moment the exchange gets recorded, in a comment thread, a review request or a ticket, the finding has to be tracked to closure and it has entered item tracking (10.26). Review platforms, pull request comments and a document's revision threads are the tools of the documented reviews, and listing them among the ad hoc review's tools would amount to changing the technique without saying so.
Sources
- IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.37 Reviews.
- Karl E. Wiegers, Peer Reviews in Software: A Practical Guide, Addison-Wesley, 2002, chapter 3, "Peer Review Formality Spectrum" (Table 3-1, activities by review type; Table 3-3, objectives and methods).
- ISO/IEC, ISO/IEC 20246:2017, Software and systems engineering: Work product reviews.
- IEEE, IEEE Std 1028-2008, IEEE Standard for Software Reviews and Audits.