If you tracked a typical RIBA Stage 5 against the time recorded, four activities would account for most of the spend. Site instructions. RFIs. Submittals. Variations. None of them are glamorous, and none of them are optional. Together they form the administrative spine of a project on site, and they quietly absorb far more senior time than the fee assumed.
The pattern repeats across firms of every size. A 5,000 square metre education project might generate two hundred RFIs, a hundred and fifty submittals, and dozens of instructions over a build programme. Each one is small. Each one needs a competent person to read it, check it against the contract documents, and respond. Multiply small by hundreds and you get the real cost of Stage 5.
What makes this hard to manage is that the CA fee is usually fixed at appointment, long before anyone knows how chatty the contractor will be or how many gaps the tender package will expose. The fee is a guess. The workload is a fact. When the two diverge, the practice absorbs the difference, and it absorbs it in the most expensive currency it has, which is senior time.
Site instructions rarely arrive in a neat batch. They trickle in, one at a time, often in response to a condition nobody could have drawn in advance. A clash between services and structure. A material that's been discontinued. Ground conditions that don't match the survey. Each instruction is a small decision, but issuing it properly means checking the drawings, confirming the cost implication with the QS, and recording it so the change is traceable later.
The drip is the problem. Because instructions come in singly, they never feel big enough to systematise, so most practices handle them ad hoc. The result is a scattered trail across email, marked-up PDFs, and someone's notebook. When a dispute surfaces eighteen months on, reconstructing that trail can take a day of someone's life. Good record-keeping upfront is cheaper than archaeology later.
An RFI looks like a question, but most of the cost sits in the answer. A contractor asks which fire-rated board applies at a particular junction. To respond, the architect has to locate the relevant clause, confirm it against the fire strategy, check the junction on the drawings, and make sure the answer doesn't contradict something agreed three months earlier. The typing takes two minutes. The finding takes thirty.
That gap between asking and answering is where RFI time disappears. UK practices commonly take anywhere from five to fifteen working days to close an RFI, and a large slice of that is simply hunting for the right information across documents that don't talk to each other. When a specification is complete and well structured, the answer is usually one lookup away. When it's vague or contradicts the schedules, every RFI becomes a small investigation.
This is the link practices underrate. RFI volume and RFI response time both track back to the quality of the documents issued at tender. Tighten the specification and you don't just answer RFIs faster, you get fewer of them in the first place.
Submittals are the quiet backlog of Stage 5. Product data sheets, shop drawings, samples, method statements, all arriving for review on the contractor's timeline rather than yours. Each one has to be checked against what was specified. Does the proposed product meet the performance criteria. Does it match the fire rating, the U-value, the acoustic requirement. Is it an or-equal substitution that needs a closer look.
The trap is that submittals pile up. A contractor pushing the programme will send a dozen in a week, and if the review turns around slowly, the whole job feels the drag. Most of the review is comparison work, lining up what was submitted against what was required. That comparison is exactly the kind of task that eats an afternoon and produces nothing a client would recognise as architecture.
There's a hidden risk in the backlog too. When reviews are rushed to clear a pile, the checking gets shallow, and a substitution that doesn't quite meet the acoustic or fire requirement slips through. That isn't caught until it's installed, when it becomes a defect rather than a comment. Slow submittal review and weak submittal review are two sides of the same understaffed coin.
Variations are where the money lives, which makes the paper trail around them unforgiving. A change gets instructed, priced, agreed, and recorded, and every link in that chain has to be documented or the firm carries the risk. Miss a confirmation and you've either done unpaid work or created an argument for the final account. The admin isn't busywork. It's the difference between getting paid and not.
What makes variations expensive in time terms is the coordination. A single change can ripple through the drawings, the specification, the schedule, and the cost plan. Keeping all four aligned, and being able to show they're aligned, is genuine work. It's also work that compounds, because an uncoordinated change made in month three becomes the RFI nobody can answer in month nine.
Look at those four activities together and a pattern emerges. The expensive part is almost never the decision. It's the retrieval, the comparison, and the cross-referencing that surround the decision. Finding the clause. Checking the submittal against the spec. Confirming the variation doesn't contradict something agreed earlier. This is the connective tissue of CA workflow, and it's where senior people lose hours they should be spending on the things only they can do.
That distinction matters because it tells you what's automatable and what isn't. The judgement is yours and always will be. Whether an or-equal product is acceptable, how to resolve a clash, what a variation should cost, these need an architect. But the work of surfacing the right information, flagging where a submittal diverges from the spec, and keeping the documents coordinated is mechanical. It follows rules. And anything that follows rules can be handled by software built for the job.
This is the part of the problem AI tools are genuinely good at. Avoice ingests a firm's existing project documentation, the specifications, schedules, drawings, and material libraries, and turns them into structured, searchable data rather than a folder of disconnected files. When an RFI lands, the relevant clause is one query away instead of thirty minutes of hunting. The architect still makes the call. The looking is done for them.
Be honest about the line. A tool that promised to answer RFIs for you would be selling a fantasy, because half of those answers depend on context only the design team holds. The realistic gain is narrower and more useful. Automate the retrieval. Automate the comparison. Automate the consistency checking that currently relies on someone remembering what was agreed in a meeting last spring.
Specification quality sits underneath all of this. Because Avoice generates specifications classified under Uniclass, CAWS, and NATSPEC standards and grounds them in a firm's own data rather than generic clause libraries, the documents that drive Stage 5 are coherent from the start. It can flag where a specification and a schedule disagree before the package goes out, which is the cheapest possible moment to catch it. Fewer contradictions at tender means fewer RFIs on site, and fewer RFIs means the whole CA workflow gets lighter.
The same logic applied to hand-drafting before CAD, and to 2D drawing before BIM. The profession kept the judgement and handed the repetition to the machine. Construction administration efficiency is the next place that trade makes sense, because the repetition is enormous and the judgement is untouched by it.
None of this means surrendering oversight. The architect stays in the loop on every instruction, every submittal review, every variation. What changes is what they're spending the loop on. Instead of forty minutes finding a clause and five minutes deciding, the ratio flips. The decision gets the attention. The lookup gets handled.
For a practice running on thin CA fees, that shift is the whole game. RIBA Stage 5 admin will never disappear, and it shouldn't, because the record it produces is what protects the firm. But the hours spent producing that record can fall a long way without any loss of rigour. Tools like Avoice are built around exactly that boundary, automating the documentation work while keeping the architect's judgement where it belongs.
The firms that work this out first will price Stage 5 differently. They'll take on construction administration knowing the admin won't eat the margin, because the parts that used to eat it are being handled. If you want to see how that plays out on a real project, Avoice offers demos tailored to the way your practice actually runs.
