How a Virtual Assistant Handles Data Entry and Admin for UK Businesses
It is Tuesday the 4th of August 2026, and somewhere in Leeds a founder is sitting in front of a spreadsheet called Q1-expenses-v3-FINAL-actual.xlsx with three days to go until her first Making Tax Digital quarterly update is due. The bank feed has pulled in 411 transactions. Roughly 60 of them are miscategorised. Two supplier invoices are sitting as photographs in a WhatsApp thread. The CRM she paid £89 a month for has four versions of the same customer, because three different people typed the company name three different ways and the fourth was created by a web form with a trailing space.
None of this is difficult. That is precisely what makes it unbearable. Every single item on that screen is a task a competent person could clear in four minutes, and there are ninety of them, and she is the only person who can do them because she is the only person who knows which supplier is which.
This is the shape of the problem in UK small business right now. Not strategy. Not funding. Not talent in the sense recruiters mean it. It is the slow accumulation of small, unglamorous, entirely necessary keystrokes that nobody has been assigned, so they default to whoever cares most — which is almost always the owner, working at 21:40 on a weekday at roughly the least economically productive rate imaginable.
What follows is an honest look at what data entry and administrative support actually involves for a British business in 2026, why the automation story has turned out to be more complicated than the vendors promised, and what changes when a trained person owns the work instead of the founder squeezing it in around everything else.
The Eleven-Hour Week Nobody Put in the Budget
Start with the numbers, because they are worse than most owners assume.
American Express’s annual SME Business Barometer, published in July 2026 and covering 1,000 owners of UK micro, small and medium-sized firms, found that respondents estimated spending an average of 11 hours a week on administrative or finance-related tasks — roughly six working days a month — while sales and business development activity got just over half that, at 3.6 days a month. The same research found that more than half said paperwork gets in the way of running their business, over a third named their own lack of capacity as the single biggest barrier to growth, and one in five was working 60 hours a week or more.
The average British small business owner spends nearly twice as long on admin as on winning work — not by choice, but because admin comes with a deadline attached and business development does not.
The pattern holds among employees too. Fyxer’s Admin Burden Index, which surveyed 5,000 UK and US office workers across November and December 2025, found office workers spend 5.6 hours a week — close to a full working day — on admin they believe could be handled some other way, putting the combined UK and US productivity cost at roughly $954 billion a year, or about $17,000 per employee. The report’s framing is blunt about why this persists: administrative work underpins finance, HR, healthcare, logistics and professional services, yet it is routinely treated as too fragmented, too people-dependent or too small to be worth fixing.
Too small to be worth fixing, repeated 250 times a year, is the whole problem.
Eleven hours a week on admin. Six working days a month. And 3.6 days on the thing that actually brings money in. The average UK small business is running its calendar upside down.
Business insurer Superscript found something similar among 500 British SME owners. A third said they simply do not have enough time to complete the admin required during working hours — data entry, bookkeeping, communications, customer relations — and while employers said they would ideally free up as much as four hours a day, in practice they manage around 90 minutes a week.
The gap between four hours a day and 90 minutes a week is not a discipline gap. It is a staffing gap that has been misdiagnosed as a discipline gap for about a decade.
What “Data Entry and Admin” Actually Covers in Britain in 2026
Here is where UK businesses have a specific problem that generic outsourcing advice tends to miss. The British administrative load has grown considerably heavier in the last eighteen months, and almost all of the growth is compliance-shaped and non-negotiable.
Consider what has landed since April 2025.
Making Tax Digital for Income Tax went live on 6 April 2026. Sole traders and landlords with qualifying income above £50,000 can no longer file a single annual Self Assessment return; they must keep digital records through the year and submit quarterly income and expense summaries through approved software. HMRC estimates around 780,000 people are in scope from this first phase alone, with a further 970,000 joining in April 2027 when the threshold drops to £30,000. HMRC has been writing to a broader group of roughly 864,000 individuals, and a sole trader who is also a landlord has to file two updates each quarter — one for the trade, one for the property business. The first deadline is 7 August 2026. That is this Friday.
Companies House identity verification is running on a hard clock. The rollout launched on 18 November 2025 with a twelve-month transition period ending 17 November 2026, linking every UK company officer to a verified personal code, with deadlines triggered by confirmation statement dates or month of birth. The consequences are not administrative tuts. Unverified directors cannot file documents at Companies House, cannot be appointed to new director roles, and risk criminal penalties for acting as a director without verification — and an accountant cannot file the confirmation statement until the personal code is supplied, which blocks routine compliance filings. On top of that, paper and PDF web forms for statutory accounts are being phased out in favour of software-only filing, and the abridged accounts option for small companies is being removed.
Data accuracy is a legal obligation, not a nice-to-have. The ICO is unambiguous that the accuracy principle sits at the heart of UK GDPR: the Data Protection Act 2018 defines “inaccurate” as incorrect or misleading as to any matter of fact, and a record saying someone still lives in London when they have moved to Manchester is plainly inaccurate. The ICO also warns that failing to comply with the principles exposes an organisation to substantial fines, since infringements of the basic processing principles sit in the highest tier of administrative penalties. And from June 2026, every organisation processing personal data must have a clear internal process for handling data protection complaints, regardless of size or sector.
So when a UK business says “data entry and admin,” what it actually means in 2026 is: quarterly HMRC submissions built on in-year records that have to be right; a compliance calendar with criminal exposure attached; a customer database that is subject to a statutory accuracy duty; supplier reconciliation; and a complaints process that now has to exist on paper.
That is not a job for a spare afternoon. It is a role.
The Error Maths Almost Nobody Runs
The instinctive response is that admin is cheap, because the person doing it is either the owner (nominally free) or someone junior. Both assumptions collapse once you cost the mistakes.
The research on manual keying accuracy has been consistent across four decades. The most-cited benchmark, from 1980s and 1990s studies of skilled operators, put error rates at 0.5–1% under controlled conditions with verification procedures — the best achievable floor. Real-world rates run higher: a 2015 study of clinical data entry found 3.7%, financial data entry research reported 2.5% in structured numeric fields and up to 4.8% in free-text descriptive fields, and government processing audits have documented 2–5% depending on agency and document complexity. A commonly cited summary puts it at 1–4% of fields even for trained staff, with error rates spiking to 18–40% of fields when workloads are high or processes are complex.
Now attach a price. Gartner estimates poor data quality costs organisations an average of $12.9–15 million a year, and at the micro level studies suggest each individual data error costs between $50 and $150 to resolve once you add investigation, correction, revised invoices, updated reports and follow-up communication. Invoice processing is the best-documented case: IOFM reports that 3.6% of all invoices contain at least one data entry error requiring correction, and the average manual processing cost per invoice runs to $15.97.
Apply that to a modest UK services firm processing 300 supplier invoices and 400 customer records a month. At a conservative 2% error rate, that is fourteen wrong fields a month, roughly 170 a year, at £40–£120 each once you count the chase, the correction and the apology. Somewhere between £7,000 and £20,000 a year, none of it a line item anywhere, all of it absorbed as “just how it is.”
There is a well-worn data quality rule of thumb behind why this compounds: a pound to prevent an error at entry, ten to correct it in the system, a hundred once it has reached a customer, an auditor or HMRC. A misspelt company name is nothing. The same misspelling in a VAT return is something else.
A 2% error rate sounds like a rounding error until you notice it is compounding daily across a system that HMRC now wants a summary of every three months.
The Human in the Loop: Why Full Automation Keeps Falling Just Short
Given all of this, the obvious question is why any of it still requires a person. The tooling is remarkable. Bank feeds categorise. OCR reads invoices. Large language models parse unstructured documents that would have defeated a rules engine five years ago.
The honest answer, backed by the 2026 evidence, is that the tooling is excellent at the middle of the distribution and unreliable at the edges — and it is the edges that cost money.
Start with the accuracy threshold that actually matters. Character-level accuracy is a misleading metric, because a system can score 99% character accuracy and still extract an invoice total incorrectly, and for financial fields and identity documents the 2026 benchmark is 99.9% field-level accuracy — the threshold required for straight-through processing, where documents move through a workflow with no human review at all. Below that, you are not automating; you are generating a review queue.
How close does the technology get? It depends entirely on the input. Tesseract, the most widely deployed open-source engine, hits 98–99% on clean 300+ DPI printed documents but is poor at handwriting, scene text and complex layouts. Current benchmarks put character error rates below 1% for clean printed text but at 3–5% for handwriting recognition, with resolution below 300 DPI causing a measurable drop and some studies putting degradation at 20% or more for poor scans.
There is also a quietly damning finding about how these systems are evaluated. On OCR Arena, a community platform where users vote in blind head-to-head comparisons, frontier vision-language models win the matchups while models that top automated benchmarks rank near the bottom — the likely explanation being that automated benchmarks over-weight specific document types and formatting, while humans care about readability across the full range of messy real-world documents. The guidance that follows is worth reading twice: do not rely on published benchmark numbers alone, test on your actual document types, and keep a fallback path, because no OCR model handles every scan, language, layout and handwritten note reliably.
The academic work points the same way. A 2026 paper describing a multi-agent document processing pipeline reported results on a production dataset of 955 real-world documents, achieving a 97.0% full-pipeline automation rate, with the full configuration including human-in-the-loop supervision reaching 98.5% document-level accuracy and a modelled FTE reduction of roughly 70% at a volume of 100,000 invoices a year. Note what that architecture assumes rather than what it eliminates. Human supervision is not the legacy component being phased out. It is a named part of the configuration that produces the best result.
Benchmark work in regulated sectors reaches the same conclusion from the other direction. An analysis of document AI on real pharmaceutical content found that case studies showing 70–90% accuracy on heterogeneous tasks demonstrate these systems can reliably handle the majority of typical content while human experts remain in the loop for exceptions, and warns that relying solely on traditional OCR or generic benchmarks risks error. Even the newest commercial releases are built around this assumption rather than against it — Mistral’s OCR 4, launched in June 2026, returns bounding boxes, typed-block classification and inline confidence scores that drive source-grounded citations, redactions and human-in-the-loop verification.
There is a regulatory layer too. The Data (Use and Access) Act has reshaped the UK’s automated decision-making rules, creating a more permissive framework for decisions based solely on automated processing while adding new safeguards alongside it, with the ICO’s AI guidance applying accuracy, lawfulness and data minimisation across the AI lifecycle. Commentary on the accuracy principle warns the obligation will only intensify as automated systems amplify the downstream consequences of inaccurate input data.
That last phrase deserves emphasis. Automation does not neutralise a bad input. It distributes it faster.
Which brings us to what a trained human actually contributes. Not typing speed — software wins that outright. What a person contributes is the small set of judgements that sit outside the training distribution:
- Recognising that the invoice marked £1,240.00 is from a supplier who has never invoiced above £200, and querying it before it is paid.
- Knowing that “J. Hartley Ltd”, “Hartley (J) Limited” and “Hartleys” are one customer, because they spoke to the office manager last month.
- Spotting that a bank feed has silently duplicated a fortnight because a connection dropped and reconnected.
- Noticing that a form submission carries a Manchester postcode with a London address line, and picking up the phone rather than saving both.
- Understanding that a supplier statement does not reconcile because a credit note was issued in a prior quarter, which is exactly the sort of context no model has.
None of these is glamorous. All of them are the difference between a clean quarterly submission and a correction.
The 2026 automation stack is superb at the 97% and dangerous at the 3%. The whole argument for a trained assistant is that they own the 3% — and know which 3% it is.
What the Work Actually Looks Like, Week by Week
A well-run administrative VA engagement is not a queue of ad hoc requests. It is a rhythm. The ones that fail are almost always where the client treats the assistant as an inbox rather than an operator with a standing remit.
Daily. Inbox triage against agreed rules. Bank feed categorisation, with anything ambiguous flagged rather than guessed. New enquiry records created under a consistent naming convention. Calendar management and confirmations. Receipt and invoice capture from wherever they have landed, including the WhatsApp thread.
Weekly. Supplier invoice entry and matching against purchase orders. Duplicate detection and merging in the CRM. Chasing documents clients promised and did not send. Updating the pipeline so Monday’s numbers are the actual numbers. A short written note on anything that looked wrong.
Monthly. Bank and supplier statement reconciliation. Expense reports. Data hygiene passes — dead records archived, formatting normalised, missing fields chased. Preparing the pack the accountant actually wants rather than the shoebox they usually get.
Quarterly. MTD preparation from records maintained continuously rather than assembled in a panic. Compliance calendar checks — confirmation statement dates, verification deadlines, insurance renewals. A data quality review: what broke, why, and which intake process needs changing so it stops breaking.
The strategic value sits in that last item. A good administrative assistant does not just process errors. After three months they can tell you that 40% of your bad records come from one web form with no validation, and that fixing the form is worth more than any amount of downstream correction.
That is not data entry. That is process ownership — and it is invisible unless someone is watching the same data every day.
The South African Advantage
Every offshore support conversation eventually reduces to the same three variables: when they work, how they communicate, and what they cost. South Africa is unusual in that it does not force a trade-off between them.
The Timezone Is the Whole Argument
South Africa runs on GMT+2 — just two hours ahead of the UK, giving a full six to eight hour overlap every working day, which supports real-time collaboration on Teams, Slack and Zoom without overnight gaps. Cape Town does not observe daylight saving, so the gap narrows to one hour during British Summer Time and widens to two in winter, and it never inverts.
For administrative work this matters more than for almost any other outsourced function, because admin is a clarification-heavy discipline. “Is this the same client?” “Capex or opex?” “Chase this or leave it?” Each question is thirty seconds of your time and unblocks an hour of theirs. In a same-day timezone the answer arrives and work continues. Across a seven or eight hour gap — the Philippines sits at GMT+8 — every clarification costs a day of elapsed time, and a week’s work becomes a fortnight’s.
VAConnect frames it plainly on the UK page: South Africa and the UK share more than a history, they share a working day — meaning a VA starts when you do and delivers as though they are in the office next door. A UK marketing manager quoted in a Clutch review made the comparison directly, saying their South African assistant “might as well be in the next office” after eighteen months of scheduling gymnastics with providers seven to eight hours away.
English That Reads British
Data entry is not language-neutral work. Somebody typing your customer records is also writing your chase emails, your supplier queries and your holding replies. Every one of those is a small trust signal.
South Africa ranks 13th globally for English proficiency and first in Africa on the EF English Proficiency Index 2025, which makes it a structurally stronger language match for UK buyers than most Asian alternatives. More to the point, South African business English sits naturally close to British usage — the spellings, the date order, the register of understatement that reads as professional in Britain and evasive almost nowhere else. VAConnect states that all its VAs have native-level English fluency and that for UK client-facing roles it specifically matches candidates with British English proficiency and an understanding of UK business culture and communication norms.
The practical consequence is that nobody has to proofread. When your assistant writes to a supplier, it sounds like your business wrote to a supplier.
A Sector With Genuine Depth, Not a Freelancer Pool
The most underrated part of the South African case is that it is not a cottage industry but a national sector with institutional weight behind it.
BPESA reported roughly 150,000 workers employed in South Africa’s global business services sector as of 2024, up from 65,000 in 2019 — a 131% increase in five years — with 20,518 net new jobs created in full-year 2024, around 90% of them going to young workers. Revenue tracked the same curve: from USD 1.04 billion in 2019 to an estimated USD 2.91 billion in 2024, a 180% increase in five years. And the direction of demand is not ambiguous — UK businesses account for 48% of new GBS employment, and US buyers have ranked South Africa as their leading offshore destination. The sector is now recognised as a preferred offshore delivery location for markets including the UK and US, with the Western Cape alone posting 13,056 jobs in 2025.
What this means for a UK business hiring one person is that the person you hire has been trained in an ecosystem that already knows what UK compliance work looks like. They are not learning that a confirmation statement is not a bank statement on your time.
Cost, Without the Quality Trade-Off
The arithmetic changed in Britain on 6 April 2025 and has not changed back. Employer National Insurance rose from 13.8% to 15%, the secondary threshold fell from £9,100 to £5,000, and while the Employment Allowance more than doubled to £10,500 with its £100,000 eligibility cap removed, the net effect was to raise the baseline cost of every hire. Both figures were maintained for 2026/27. As one analysis put it, the fixed additional cost per employee from the threshold reduction makes multiple part-time roles comparatively more expensive than fewer full-time ones.
Set that against UK admin pay. The median UK administrative salary in 2026 sits at £27,500, with London highest at £31,400. Add employer NI, pension, equipment, holiday and sick cover, and a £27,500 administrator lands somewhere north of £33,000 fully loaded — for a role that, in most small firms, is genuinely needed for perhaps three days a week.
South African delivery is costed differently. Industry analysis puts South African BPO at 55–65% cost savings versus UK, US and Australian in-house hiring. VAConnect’s own UK positioning makes the comparison directly, noting that a full-time PA in London runs £35,000–£50,000 plus NI, pension and office space, while South Africa offers the same calibre of professional on the same working hours at a different cost base.
The distinction that matters is that this is currency arbitrage, not a quality discount. You are not buying a less capable person. You are buying a comparably capable person whose cost of living is denominated in rand.
Managed, Not Matched: Why the Model Matters More for Admin Than for Anything Else
Here is where the marketplace approach quietly fails, and it fails worst on exactly this category of work.
Data entry and admin are continuity businesses. Their entire value comes from the same person seeing the same data repeatedly until they know it better than you do. The moment that person changes, you are not just re-hiring — you are re-teaching every convention, every exception, every supplier quirk, every reason that one client’s invoices are always coded differently. On a freelance platform, that reset is a normal part of the year.
VAConnect’s argument is that continuity has to be engineered rather than hoped for. The company handles recruitment, training, performance reviews and backup cover, so the client gets the output without the overhead of managing another hire, and reports 98% client retention. That sits on top of four proprietary platforms: a dedicated South African VA talent portal that sources and pre-screens candidates through skills testing, background checks and cultural fit assessment before anyone reaches a shortlist; VAVarsity, the training platform every VA is upskilled through before touching client systems; and the VAPIness and Atomic Energy programmes covering happiness, accountability and performance. The company is explicit that 98% retention does not happen by accident — it is engineered.
The failure case is covered too: if a placement is not performing to the agreed standard, VAConnect rematches at no additional cost and manages the full transition. On speed, most matches are filled within two to three weeks.
The outcome data on admin specifically is the part worth dwelling on. Jonathan Perry, a partner at Perry & Associates, in a verified Clutch review after a 20-month placement, reported that “our VAConnect VA handles 60% of what used to take an entire admin team” — with the firm’s admin headcount going from three to one.
One trained, permanent, timezone-aligned assistant absorbing 60% of what previously took three people. That is not a productivity gain. That is a different operating model.
The First Ninety Days, Honestly
It is worth being straight about the ramp, because unrealistic expectations are the main reason good placements get judged badly.
Week one. Access, tools and observation. The assistant is watching how your data actually flows rather than how the process document says it flows. Expect meaningful output on well-defined tasks — inbox triage, receipt capture, straightforward record creation — but not independent judgement yet.
Weeks two to four. The conventions get written down. This is the phase where a good assistant builds the standard operating procedure that did not previously exist, because the rules were living in your head. Expect the first “why do you do it this way?” questions, and expect a couple of them to be uncomfortable.
Weeks four to eight. Full ramp on the recurring rhythm. Daily and weekly tasks run without prompting, and the error queue shrinks because things are caught at intake rather than at reconciliation.
Weeks eight to twelve. The shift from processing to ownership — where the assistant starts telling you what is broken upstream. Acting on that advice is where the compounding return lives.
One caveat worth stating plainly: an assistant fixes capacity fast and data quality slowly. Clean records take a quarter to establish and a year to fully pay off, because the value shows up in decisions you make with numbers you can trust. Anyone promising a transformed database in three weeks is selling something.
The Gap Has Become Genuinely Uncomfortable
Step back and look at the two businesses side by side.
Business A: the owner does the admin. Eleven hours a week, at whatever an owner’s hour is worth. Records assembled quarterly under deadline pressure. A 2–4% error rate nobody measures. Compliance dates tracked in memory and a calendar reminder. MTD submissions built from a reconstruction rather than a record. Every clarification routed through the busiest person in the company.
Business B: a trained, timezone-aligned assistant owns the work. Records maintained continuously, so the quarterly submission is a review rather than a rescue. Errors caught at intake, where they cost a pound rather than a hundred. A compliance calendar owned by someone whose job it is. The owner’s eleven hours redirected to the 3.6 days a month that were meant to go to sales.
The two businesses look identical from the outside. One of them has a founder who can take a Friday afternoon.
What is striking is not that the second model works. It is how quickly the gap has opened, and how much of it is now structural rather than effort-based. The compliance load went up in 2026. The cost of a UK hire went up in 2025 and stayed up. The tools got dramatically better at the middle of the distribution and no better at the edges — which means the residual human work is now concentrated, high-stakes, and precisely what you do not want your most expensive person doing at 21:40 on a Tuesday.
The businesses that worked this out early are not working harder. They stopped asking their founder to be a data entry clerk.
The Comparison
| DIY / Owner-Handled | Generic Freelancer or AI Tool | VAConnect Managed Admin VA | |
|---|---|---|---|
| Weekly hours recovered | None — it is your 11 hours | Some, offset by management time | 25–35 hours of dedicated capacity |
| Timezone overlap with UK | N/A | Often 7–8 hrs offset (GMT+8) | GMT+2 — 6–8 hrs of daily overlap |
| Continuity of knowledge | Total, but locked in your head | Resets with every contractor change | Same VA long-term; 98% retention |
| Error catching at intake | Whenever you get to it | Tool flags, nobody investigates | Trained human owns the exception queue |
| Handles the 3% edge cases | Yes, badly, at 21:40 | No — this is where tools fail | Yes, by design |
| British English communication | Yes | Variable; often needs proofing | Native-level, matched for UK register |
| MTD quarterly preparation | Panic reconstruction | Not in scope for most freelancers | Continuous records, review not rescue |
| Compliance calendar ownership | Memory and hope | Unowned | Owned and tracked |
| UK GDPR accuracy duty | Your exposure entirely | Unclear accountability | Managed under structured data handling |
| Cover for illness or leave | Work stops | Work stops | Backup cover managed by agency |
| Training before touching systems | N/A | None | VAVarsity, before system access |
| Vetting depth | N/A | Profile and reviews | Skills testing, background and fit checks |
| If it is not working | It is still your problem | Re-post, re-hire, re-train | Free rematch, transition managed |
| Fully loaded cost | Your hourly rate × 11 × 52 | Low rate + 10–15 hrs/wk managing | ~55–65% below UK in-house equivalent |
| Time to productive output | Immediate, at your expense | 2–8 weeks, high variance | Week one output, full ramp 2–4 weeks |
| Process improvement over time | Rare — no capacity to review | Rare — no incentive | Quarterly data quality review |
Three days until the first MTD deadline. If the records that submission depends on are currently spread across a bank feed, a shoebox and a WhatsApp thread, the problem is not the software. It is that nobody owns the inputs.
That is a solvable staffing problem, and it takes about two to three weeks to solve.
Book a 30-minute discovery call with VAConnect →
Sources
- American Express SME Business Barometer 2026 (survey of 1,000 UK micro, small and medium business owners), reported July 2026
- Fyxer, Admin Burden Index 2026 (survey of 5,000 UK and US office workers, November–December 2025)
- Superscript research with 500 UK SME owners, via Small Business Charter
- HMRC / GOV.UK, Making Tax Digital for Income Tax guidance and rollout thresholds; STEP and ByteStart coverage of the April 2026 phase
- Companies House identity verification rollout under the Economic Crime and Corporate Transparency Act 2023; ACSP and deadline analysis, 2026
- Information Commissioner’s Office, A Guide to the Data Protection Principles and Principle (d): Accuracy; DUAA complaints regime guidance, June 2026
- Lido, Data Entry Error Rates (review of accuracy research across four decades); DigiParser manual data entry error rate report
- InfoSeeMedia, The Hidden Operational Risks of Manual Data Entry and Document Processing, June 2026 (Gartner data quality cost figures)
- Parsli, 67 Data Entry Statistics & Automation Trends (2026) — IOFM invoice error and processing cost benchmarks
- LlamaIndex, OCR Accuracy Explained (2026 field-level accuracy benchmarks and straight-through processing thresholds)
- The Definitive Guide to OCR in 2026: From Pipelines to VLMs (OCR Arena blind evaluation findings)
- MADP: A Multi-Agent Pipeline for Sustainable Document Processing with Human-in-the-Loop, arXiv 2026
- IntuitionLabs, Pharma Document AI & OCR Accuracy: A Benchmark Analysis
- Mistral AI, OCR 4 release notes, June 2026
- BPESA GBS sector job creation and revenue data, 2024–2026; Afrishore BPO South Africa BPO statistics; ITWeb coverage of BPESA sector reporting
- EF English Proficiency Index 2025
- Pin-Point UK Administrative Salary Guide 2026; HMRC / Xero / PayFit employer National Insurance guidance 2025–26 and 2026–27
- vaconnect.co.uk and vaconnect.co.za — UK service pages, managed model, VAVarsity, VAPIness, Atomic Energy, retention and replacement policy
- Verified Clutch reviews — Jonathan Perry, Partner, Perry & Associates; UK marketing manager review cited on vaconnect.co.uk