Bandwidth Planning for Growing UK Businesses: A Practical Guide to Sizing Your Internet Connection in 2026

Almost every UK business sizes its internet connection exactly once. Someone signs a lease, an installer quotes what is available at the postcode, a number is chosen that sounds generous, and that number is still on the invoice four years later. Business bandwidth planning is the discipline of revisiting that decision deliberately, on a schedule, against what the organisation actually does today rather than what it did on move-in day. It is unglamorous work, and it is one of the few infrastructure decisions where both overspending and underspending carry a visible cost.

This guide covers how bandwidth demand actually changes as a business grows; why upload capacity now behaves differently from download capacity in a cloud-first estate; what each common application really consumes; how symmetric and asymmetric connections differ in ways that matter beyond the headline number; what UK circuits cost in 2026 across FTTP, SoGEA and dedicated Ethernet; and a repeatable framework for right-sizing connectivity as headcount changes. The aim is to give you a method you can run yourself each year, not a single recommendation that expires the moment you hire five people.

What bandwidth planning actually means in 2026

Bandwidth planning is the process of matching the capacity, symmetry, resilience and contractual guarantees of an internet circuit to the measured and forecast demand of the people and systems behind it. The word most often skipped in that definition is measured. The overwhelming majority of connection-sizing conversations in UK small and mid-sized businesses begin with a headcount and a guess, when the organisation is already sitting on months of usable utilisation data from its firewall or router.

Four attributes describe any business circuit, and they are frequently conflated into the single number on the marketing page. The first is downstream capacity, the rate at which data arrives. The second is upstream capacity, the rate at which data leaves, which on most UK broadband products is a fraction of the downstream figure. The third is contention, the degree to which that capacity is shared with other subscribers on the same infrastructure, and whether any portion is contractually guaranteed. The fourth is the service wrapper: the fix-time commitment, the availability target, whether the address is covered by a formal SLA with service credits, and how quickly a fault reaches an engineer rather than a queue.

Two circuits advertising “1 Gbps” can differ on every one of those attributes. A gigabit FTTP broadband service typically delivers up to 1000 Mbps down and 115 Mbps up, on a shared bearer, with best-efforts repair and no guaranteed throughput. A gigabit dedicated Ethernet circuit delivers 1000 Mbps in both directions, uncontended, with a fix-time commitment measured in hours and financial credits if it is missed. The first might cost £60 a month and the second £600. Neither is the correct answer in the abstract; the correct answer depends on what breaks when the line degrades, and who is waiting.

The reason this matters more in 2026 than it did five years ago is that the traffic profile of a typical UK office has inverted. In a server-in-the-cupboard estate, the internet carried email, web browsing and a handful of remote sessions, and almost all of it flowed inbound. In a cloud-first estate with Microsoft 365, hosted telephony, cloud backup, SaaS line-of-business applications and daily video calls, a substantial and growing share of traffic flows outbound. The connection that was correctly sized for the first pattern is frequently mis-sized for the second, and the failure shows up as choppy calls and stalling file uploads rather than as slow web pages.

The UK infrastructure picture underneath all of this has changed faster than most procurement cycles. Full-fibre coverage has expanded to the large majority of UK premises through a combination of Openreach build and alternative network operators, which means many business addresses that were limited to a copper VDSL service three years ago now have two or three genuinely competitive options. At the same time the retirement of the legacy PSTN has pushed voice, alarm lines, lifts and card terminals onto data circuits, and SoGEA has replaced the old broadband-plus-line-rental arrangement with a single data-only order. The practical consequence is that a contract signed before roughly 2023 is likely to be both more expensive and less capable than what is available at the same postcode today, and no provider has a commercial incentive to tell you that unprompted.

It is also worth being clear that bandwidth demand does not grow linearly with headcount. It grows in steps, and the steps are triggered by decisions rather than by hiring. Moving a file server to SharePoint, adopting cloud desktops, switching to a hosted phone system, installing IP CCTV, or simply enabling a default-on sync client across the estate each shift the baseline in a single week. A business can add fifteen people with almost no measurable change and then double its upstream peak the month it migrates a shared drive. This is why sizing against headcount alone fails: the headcount is the visible variable, and it is rarely the one doing the work.

Where the bandwidth actually goes - per-application demand

Sizing arguments tend to stall because nobody in the room can say what an application consumes. The figures below are steady-state per-user or per-stream demands for the applications most commonly found in a UK office, expressed as a share of a 100 Mbps reference so they can be compared at a glance. They describe sustained load during active use, not the burst a single file transfer can produce when nothing else is competing.

Two things stand out immediately. The first is that voice, which causes more complaints than any other application when a connection is stressed, consumes almost nothing. A G.711 call with RTP, UDP and IP overhead runs at roughly 87–100 Kbps in each direction; thirty simultaneous calls need around 3 Mbps. Voice is not a capacity problem, it is a jitter and prioritisation problem, and no amount of additional bandwidth will fix it if the circuit has no quality-of-service policy and the backup job is allowed to saturate the uplink at 10:30 on a Tuesday.

The second is the asymmetry of the demand itself. Cloud backup, file sync, video calls and offsite CCTV are all substantially or entirely upstream, and together they account for the largest numbers in the table. Web browsing and SaaS, the traffic most people picture when they imagine “internet usage”, sits near the bottom. This is precisely the inversion that catches businesses out: they buy more download capacity because the download number is the one on the tariff, while the constraint sits on the upload side where the number is fifteen times smaller.

A third point is easy to miss because it concerns behaviour rather than volume. The workloads in the table fall into two categories that stress a circuit very differently. Voice, video and interactive applications are latency-sensitive and rate-constant: they need a modest, steady allocation and they degrade immediately if it is not available. Backup, sync and update traffic is elastic and greedy: given a free link it will consume everything available, because that is what the transport protocols are designed to do. Put the two on the same circuit without a prioritisation policy and the elastic traffic will reliably crowd out the sensitive traffic, not occasionally but every single time the two coincide. This is the mechanism behind the classic complaint that calls are fine first thing and unusable by eleven.

It is also worth being precise about what “per user” means, because this is where estimates go wrong in both directions. A 60-person office does not have 60 people on video calls at once. Depending on the sector, concurrent video participation typically peaks somewhere between 15% and 40% of headcount, with professional services, agencies and anything sales-led clustering at the top of that range and operational, field-based or shift-based businesses at the bottom. Multiplying headcount by a per-user figure without applying a concurrency factor produces a number two to six times larger than reality, which is the single most common route to an oversized and overpriced circuit.

Bandwidth planning by the numbers - the UK 2026 picture

The figures below describe the position we most often encounter when auditing connectivity for UK organisations in the 20–250 employee band. They are useful mainly as a sanity check: if your own numbers differ sharply in either direction, that difference is the thing worth investigating.

The first and last of those numbers interact in a way that causes real operational pain. A connection reviewed only when it fails, or only when the office moves, is being reviewed under time pressure - and dedicated circuits are the one piece of IT infrastructure that cannot be expedited by paying more. Openreach and alternative network operators work to published lead times, and where a new duct, a road crossing or a wayleave agreement with a landlord is involved, three months is an ordinary outcome rather than a worst case. The organisations that avoid this are not the ones with bigger budgets; they are the ones that started the conversation a quarter earlier. We covered the adjacent version of this problem, where the circuit is the critical path on a move date, in our guide to the hidden costs of an office move and IT relocation budget.

The 8:1 ratio deserves comment too, because it is not a law of physics. It is a product of how GPON fibre-to-the-premises is engineered and sold in the UK. The underlying technology is capable of far more symmetric behaviour than the retail tariffs expose; the split exists because consumer demand historically ran one way and because symmetric capacity is the product differentiator that justifies dedicated-circuit pricing. Knowing that the ratio is a commercial decision rather than a technical ceiling is useful when you are evaluating whether to pay for symmetry or engineer around the lack of it.

Scoring your current connection - where most UK businesses actually sit

Before choosing a target, it is worth scoring the position you are in. The assessment below covers the four attributes that determine whether a circuit is adequate, and it is deliberately structured so that a non-technical finance or operations lead can complete it with help from whoever holds the router password. Score each row honestly; the value is in finding the one or two that are genuinely red rather than in producing a flattering total.

The pattern across these three cards is consistent and worth naming. Capacity problems are visible and get attention. Quality problems are invisible in a speed test and get blamed on the application vendor. Contract problems are invisible until the day the line fails, at which point the difference between a five-hour fix commitment and a best-efforts broadband fault becomes the only thing anyone cares about. A business that scores well on raw capacity and poorly on the other two cards has bought the wrong thing, and buying more of it will not help.

The monitoring requirement in the second card is not optional if you want any of this to be repeatable. Jitter, packet loss and latency have to be sampled continuously to be meaningful, because the complaint that reaches the service desk is always retrospective: the call was bad twenty minutes ago. Our guide to proactive network monitoring for UK businesses covers how to establish that baseline and how to keep the alerting readable after the first month.

Symmetric versus asymmetric - the decision that actually matters

Most bandwidth conversations are framed as a question of how much. The more consequential question is usually what shape. An asymmetric connection allocates most of its capacity downstream; a symmetric connection provides equal capacity in both directions. For a cloud-first business the second is often worth more than a headline-number upgrade to the first, and it is almost always the more expensive option, so the trade-off deserves to be made explicitly rather than by default.

The honest position is that a large number of UK businesses do not need a dedicated circuit and are sold one anyway, while a smaller number need one badly and have been told their broadband is fine. The distinguishing test is not headcount. It is whether the business has a sustained upstream requirement that cannot be moved outside working hours. A 90-person accountancy practice whose backup runs at 01:00 and whose client calls are mostly audio can run comfortably on gigabit FTTP. A 30-person video production company pushing rushes to cloud storage during the working day cannot, regardless of how the download figure looks.

There is a middle path that is frequently overlooked. Several UK providers now sell symmetric products over FTTP infrastructure - often marketed as symmetric business fibre or a “soft” dedicated service - at 300/300 or 500/500 Mbps for somewhere between £100 and £200 a month. These sit between the two columns above: they deliver the symmetry without the uncontended guarantee or the five-hour fix time. For a business whose problem is genuinely upload capacity rather than availability risk, this is often the correct and considerably cheaper answer, and it is worth asking for explicitly because it rarely appears on the first quote.

Two ancillary details frequently decide the question in practice, and neither appears in the capacity comparison. The first is addressing: dedicated circuits come with a routed block of static IP addresses as standard, while an increasing number of broadband products place customers behind carrier-grade NAT unless a static address is specifically ordered. If you host anything on-premises, terminate a site-to-site VPN, or maintain IP allow-lists with a supplier or a client, that detail matters more than the throughput figure. The second is the demarcation point for fault diagnosis. On a dedicated circuit the provider monitors the service end to end and will usually detect a fault before you report it; on broadband the first diagnostic step is frequently a request that you reboot the router and test again tomorrow.

The other structural option is to stop treating the circuit as a single purchase. Two diverse asymmetric circuits from different carriers, bonded or managed by an SD-WAN appliance, can deliver more aggregate capacity and better real-world availability than one mid-sized leased line at a similar total cost. What it will not deliver is a single contractual fix-time commitment covering the service as a whole, which is precisely the thing a regulated business or one with contractual uptime obligations to its own clients may need. The choice between those two shapes of resilience is a business decision, not a technical one.

The bandwidth review timeline - what a proper process looks like

A connectivity review that produces a good decision takes about twelve weeks of elapsed time and perhaps three days of actual work. Most of the elapsed time is waiting: for utilisation data to accumulate, for surveys to come back, for a provider to confirm whether excess construction charges apply at your postcode. The sequence below is the one we run, and it is deliberately front-loaded with measurement so that the commercial conversation happens after the requirement is known rather than before.

The step organisations most often want to skip is weeks 2 to 5, and it is the one that carries the most value. Four weeks of waiting feels like inertia when a director has already decided the internet is slow. But a decision made on a week of data, taken during a quiet period or an unusually busy one, is a decision made on noise. Circuits are bought on three- and five-year terms; spending a month to get the requirement right is a reasonable ratio.

It is also worth setting expectations about week 7 explicitly, because the result is frequently anticlimactic. A meaningful share of “we need more bandwidth” conversations resolve into a single laptop running an unthrottled cloud sync of a 400 GB archive folder, a security camera uploading at full bitrate to an offsite recorder, or a backup job whose window has quietly crept into the working day as the dataset grew. Those are configuration fixes costing nothing. Finding them before signing a five-year contract is the return on the whole process.

What UK business connectivity costs in 2026

The table below gives typical UK monthly pricing on a 36-month term, excluding VAT and excluding any excess construction charges. Prices vary materially by postcode, by how much competing infrastructure has been built in the area, and by whether the building already has fibre terminated in the comms room. Treat these as the shape of the market rather than as a quote.

The fourth row contains the most useful commercial idea in the table and the one most often missed. A dedicated circuit is delivered on a bearer, which is the physical capacity of the fibre presented to your building, and sold at a committed rate, which is the portion of that bearer you are contracted and billed for. Ordering 100 Mbps on a 1 Gbps bearer costs a little more upfront than 100 Mbps on a 100 Mbps bearer, but it means any future upgrade to 200, 500 or 1000 Mbps is a billing change completed in days rather than a new installation taking three months. For any business with a credible growth plan, paying for the larger bearer is the cheapest insurance in connectivity.

Excess construction charges are the genuine wildcard and the reason indicative quotes should never be trusted for budgeting. If the building has no existing fibre and the nearest chamber is across a road, the civils cost is passed through. We have seen ECC quotes come back at zero, at £2,800, and at £27,000 for three buildings on the same industrial estate. The only way to know is to ask for a formal survey, which is free and takes two to four weeks, and to do it early enough that a large number does not derail a project timeline. This is the same category of budget risk we set out in the IT roadmap and technology strategy guide: the cost that is unknowable until surveyed belongs in the plan as a range, not as an omission.

Contract term is the other lever, and it cuts both ways. Moving from a 12-month to a 60-month term typically reduces the monthly rental by 20–35% and frequently waives the installation charge, which on a dedicated circuit can be worth several thousand pounds. It also commits the business to a capacity decision for five years, over which time both the requirement and the market price will move. A reasonable compromise for most organisations is a 36-month term on the larger bearer: long enough to secure meaningful discount and absorb the install cost, short enough that the decision is revisited before the technology landscape shifts again. Where a lease break or a likely office move falls inside the term, check the provider’s site-relocation terms before signing, because the cost of moving a circuit mid-contract varies enormously between providers and is rarely volunteered.

One further contractual point. Price per megabit falls steeply with capacity, which creates a genuine temptation to over-buy because the bigger number looks like better value. It is better value per unit and worse value in absolute terms if the capacity is never used. The discipline is to size against the 24-month forecast from the timeline above, buy the larger bearer, and take the committed rate you actually need - which preserves the upgrade option without paying for it until the growth arrives.

The upload problem, quantified

If there is a single number that captures why bandwidth planning has changed, it is the share of a modern office’s peak traffic that flows outbound. The figure below is what we typically measure at the busiest fifteen minutes of the working day in a cloud-first UK organisation with Microsoft 365, hosted telephony and cloud backup in place.

Forty-three per cent of peak traffic heading outbound is not remarkable in itself. It becomes remarkable when you set it against a connection that provides one-eighth as much upstream capacity as downstream. On a 1000/115 Mbps FTTP service, a peak that is 43% upstream means the upload path reaches saturation while the download path is still comfortably below a quarter of its capacity. Every diagnostic tool the business reaches for - the speed test, the router dashboard, the provider’s status page - reports a healthy connection, because all of them lead with the download figure.

This is the mechanism behind the most common complaint pattern in UK offices: video calls that degrade in the late morning, file uploads that stall, and a Teams session that drops to audio-only for one person while everyone else is fine. None of those symptoms points at the circuit in an obvious way. All of them are consistent with an upstream path that is briefly and repeatedly full. The diagnostic that settles it takes two minutes: look at the upstream utilisation graph for the period in which the complaint occurred, and see whether it is flat-topped. A flat top is a saturated link, and no application-level troubleshooting will resolve it.

Sizing benchmarks - a per-user framework you can apply

With the measurement work done, the arithmetic is straightforward. The framework below expresses the sustained per-user demand each workload places on a connection during core hours, as a proportion of a 10 Mbps-per-user reference. Multiply the relevant rows by your active headcount, apply the concurrency factors discussed earlier, and add the site-level workloads that do not scale with headcount at all.

Worked through, the method looks like this for a 70-person professional services firm. Sixty standard knowledge workers at 1.8 Mbps and ten video-led roles at 4 Mbps gives 148 Mbps of nominal demand. Applying a 35% concurrency factor - because not everyone is transacting simultaneously - brings that to roughly 52 Mbps. Add 5 Mbps of daytime backup, 3.5 Mbps of amortised update traffic and a 10 Mbps guest allowance, for about 70 Mbps at peak. Then add 40% headroom for growth and burst behaviour, giving a target near 100 Mbps sustained in each direction.

The instructive part is what that number means against real products. Downstream, 100 Mbps is comfortably inside a £50-a-month FTTP service. Upstream, it exceeds the 115 Mbps ceiling of that same service once burst behaviour is included, and sits precisely at the entry point for a 100 Mbps dedicated circuit at around £285 a month. The firm’s decision is therefore not “how fast” but whether the upstream requirement is real and sustained, or whether it can be flattened by moving backup to 01:00 and throttling sync clients. That is a £2,800-a-year question, and it is answerable with data.

The same arithmetic run for a 25-person operational business produces a very different answer and is worth stating for contrast. Twenty operational users at 0.7 Mbps and five knowledge workers at 1.8 Mbps gives 23 Mbps nominal; at 40% concurrency that is roughly 9 Mbps, plus a small guest allowance and an overnight backup that never touches core hours. The requirement lands near 20 Mbps with headroom included, which any current FTTP or even SoGEA service covers several times over. For this business the connectivity conversation is not about capacity at all - it is entirely about resilience, because the constraint that would actually hurt is the circuit being down for two days rather than being slow for two hours. Recognising which of those two problems you have is the whole substance of the exercise, and the two lead to opposite purchasing decisions at similar price points.

Two notes on applying the concurrency factor, because it is the input people get wrong. Concurrency is not an average across the day; it is the ratio at the busiest fifteen minutes, which in most UK offices falls between 10:00 and 11:30 or immediately after lunch. And concurrency for video specifically should be measured, not assumed, because it has moved sharply in both directions since 2020 depending on whether an organisation settled into hybrid working or returned largely to the office.

A readiness benchmark for connectivity planning

Pulling the three assessment cards together into a single figure gives a rough indication of how well-founded an organisation’s connectivity position is. The gauge reflects what we typically find on a first review of a UK business in the 20–250 employee band that has not formally reassessed its circuit since installation.

A score in the high forties has a recognisable shape. Raw capacity usually scores acceptably, because bandwidth has become cheap enough that even an unconsidered purchase tends to land somewhere workable downstream. Prioritisation scores poorly, because QoS requires someone to have configured it deliberately and nobody owns that task by default. Resilience scores worst of all, not because businesses reject the idea of a secondary circuit but because failover that has never been tested is functionally the same as no failover, and testing it requires deliberately breaking a working connection during business hours - a task that is always scheduled for next quarter.

The practical reading is that the cheapest improvements available to most organisations are not capacity upgrades. Configuring quality-of-service to protect voice and video, moving bulk transfers out of core hours, and testing the failover path are three pieces of work that together cost a day or two of engineering time and move the score further than a doubling of the line rate would.

Common bandwidth planning mistakes

The failures below account for the large majority of connectivity problems we are asked to investigate. None involves exotic technology, and each is straightforward to avoid once it has been named.

• Sizing on headcount alone. Multiplying staff by a per-user figure with no concurrency factor and no application mix produces a requirement two to six times too large, and the resulting circuit is sold on a five-year term.
• Reading averages instead of percentiles. A monthly average of 15% utilisation is entirely compatible with a link that saturates for forty minutes every afternoon. The 95th percentile and the sustained-peak figure are the numbers that describe user experience.
• Ignoring the upload ceiling. The most common real constraint in a cloud-first estate, and the one that is invisible in every speed test and provider status page because those lead with the download figure.
• Buying capacity to fix a prioritisation problem. Choppy calls on an unsaturated circuit are a QoS, jitter or routing issue. Doubling the line rate changes nothing and costs a great deal.
• Letting backup windows drift into working hours. A backup job sized correctly at 200 GB three years ago may now be running until 10:30. Nobody is alerted to this, because the job still completes successfully.
• Ordering the committed rate on a matching bearer. Saving a small monthly amount on the bearer converts every future upgrade from a two-day billing change into a three-month installation.
• Treating the indicative quote as the budget. Excess construction charges are unknowable without a formal survey and have a genuine range from zero to five figures. Budget a range and survey early.
• Never testing failover. A secondary circuit that has not been failed over in the last six months should be assumed not to work. Most do not, usually because of a static route, a firewall rule or a DNS dependency nobody revisited.

A real-world example - when more bandwidth was the wrong answer

A Leeds-based insurance brokerage with 84 staff across two floors had spent eighteen months escalating the same complaint: video calls degraded mid-morning, and the shared document library felt slow. The incumbent provider had already upgraded them once, from a 330/50 Mbps FTTP service to a 1000/115 Mbps package, at which point the download figure trebled and the complaints continued unchanged. The next quote on the table was a 500 Mbps dedicated Ethernet circuit at just over £500 a month on a five-year term, with a fifteen-week delivery window.

The review began with four weeks of interface polling, which had not previously been collected at all. The downstream 95th percentile came in at 214 Mbps, comfortably inside a quarter of the available capacity, which explained why every speed test the brokerage ran looked excellent. The upstream picture was the opposite: a flat-topped graph pinned at 113–115 Mbps between roughly 09:40 and 11:20 every weekday, and again briefly after 14:00. The link was not slow. It was full, in one direction, for about two hours a day.

Flow analysis attributed the peak to three sources. A cloud backup job, sized in 2022 for 900 GB, was now protecting 4.1 TB and had drifted so that its incremental pass ran until 11:00. Thirty-one endpoints were running an unthrottled document sync against a library that had grown past 600 GB after a departmental migration. And a twelve-camera CCTV system installed the previous year was uploading continuously to an offsite recorder at full bitrate, consuming around 34 Mbps upstream around the clock, which nobody in IT had been told about because the installation had been procured by facilities.

We were three signatures away from committing to a five-year leased line. What actually fixed it was moving the backup window, putting a cap on the sync client and turning the camera bitrate down to something sensible for a car park. The call quality problem disappeared inside a fortnight and we are still on the same broadband service.

The remediation cost nothing beyond about a day of engineering time. Backup was rescheduled to 00:30 with a bandwidth cap applied during business hours; the sync client was throttled to 20% of the uplink between 08:00 and 18:00 via policy; the CCTV bitrate was reduced and the retention shifted to local recording with overnight offsite replication. Upstream 95th percentile fell from 115 Mbps to 61 Mbps. The brokerage kept the 1000/115 service and diarised an annual review.

The point of the story is not that dedicated circuits are unnecessary. It is that the brokerage had no way of knowing which of the two answers was correct, because it had never measured anything. Had the upstream peak turned out to be 110 Mbps of genuine, irreducible business traffic, the leased line would have been the right purchase and the £500 a month would have been well spent. The measurement did not tell them what to buy; it told them what they were actually short of, which is the prerequisite for the buying decision rather than a substitute for it. It is worth adding that where a guaranteed fix time genuinely is required, that requirement should be established against the business’s own obligations - the same logic we set out in our guide to IT support SLA response times.

The bandwidth planning checklist - twelve points to work through

Run this against your own estate. Each item is answerable in an afternoon, and the ones you cannot answer are the findings.

1. Confirm what you are actually paying for. Find the current contract and record the committed rate in each direction, the bearer size if it is a dedicated circuit, the fix-time commitment, the end date and the notice period. A surprising number of businesses cannot produce this document.
2. Verify that utilisation is being recorded. Check that WAN interface counters are polled at five-minute granularity in both directions and retained for at least twelve months. If they are not, fix this first; everything else depends on it.
3. Pull the 95th percentile for both directions. Take the last full month. Note the figure and, separately, the duration and timing of any sustained peak above 80% of the line rate.
4. Check whether the upstream graph is flat-topped. A flat top during working hours is a saturated uplink and the single most likely explanation for degraded call quality.
5. Attribute the peak to applications. Use flow data or firewall logs to produce a one-page breakdown: backup, sync, video, voice, updates, CCTV, everything else.
6. Audit the backup window. Confirm when the job starts, when it finishes, how large the protected dataset now is, and whether a bandwidth cap applies during business hours.
7. Find the traffic nobody told IT about. Offsite CCTV, digital signage, building management systems, EV charge points and guest networks all consume capacity and are frequently procured outside IT.
8. Confirm QoS is configured and effective. Voice and interactive video should be prioritised above bulk transfer. Verify by observing call quality metrics during a deliberate large upload, not by reading the configuration.
9. Forecast 24 months of headcount and systems change. Include planned migrations, any move toward cloud desktops, and any new upload-heavy workload. Size against this figure, not today’s.
10. Price the upgrade path before you need it. Ask what a move from your current committed rate to the next two tiers would cost and how long it would take. If the answer involves a new installation, consider the bearer question now.
11. Test the failover path end to end. Physically disconnect the primary circuit during a planned window and confirm voice, email, line-of-business applications and remote access all recover. Record how long it took.
12. Diarise the next review and the contract notice date separately. Twelve months for the review, and a reminder at least ninety days before notice is due, so renewal is a decision rather than an automatic rollover.

At a glance - bandwidth planning summary

How Cloudswitched approaches connectivity sizing

Cloudswitched works with UK organisations to baseline existing circuits, attribute peak demand to real applications, and price the available options across Openreach and alternative network operators at a given postcode. That includes the unglamorous parts: confirming what the current contract actually commits to, running formal excess construction charge checks early enough that the number is not a surprise, and testing failover paths that have been assumed to work. The output is a documented requirement and a costed set of options, so the decision sits with the business rather than with whoever quoted first.

Frequently Asked Questions

How much internet bandwidth does a business actually need per employee?

There is no single figure, which is why per-employee rules of thumb tend to mislead. A realistic starting point for a standard knowledge worker on Microsoft 365 with occasional video is around 1.5–2 Mbps sustained; a video-led sales or client-services role runs nearer 4 Mbps; a cloud desktop user sits around 3 Mbps; and design or media production can exceed 8 Mbps. Those figures must then be reduced by a concurrency factor, because not everyone transacts simultaneously - in most UK offices the peak concurrency falls between 15% and 40% of headcount. Finally, add the site-level workloads that do not scale with staff numbers at all, such as cloud backup, offsite CCTV and endpoint update traffic. The result is usually far lower than a naive headcount multiplication suggests.

What is the difference between symmetric and asymmetric bandwidth?

Asymmetric connections provide substantially more download capacity than upload capacity; symmetric connections provide the same in both directions. Most UK business broadband is asymmetric, typically at a ratio near 8:1 - a 1000 Mbps download service commonly offers 115 Mbps upload. Symmetric services are usually delivered as dedicated Ethernet, or occasionally as a contended symmetric fibre product at a lower price point. The distinction matters because a cloud-first estate sends a large and growing share of its traffic outbound: video calls, cloud backup, file synchronisation, hosted voice and offsite CCTV are all upload-weighted. If those workloads cannot be moved outside working hours, the upload figure becomes the real constraint regardless of how large the download number is.

How do I know if my current internet connection is too slow?

Speed tests are the wrong instrument, because they measure an instantaneous burst on an otherwise idle link and lead with the download result. Instead, look at utilisation graphs from your firewall or router across a full month, sampled at five-minute intervals, in both directions. Two things indicate a genuine capacity problem: a 95th percentile above roughly 60–70% of the line rate, and any period where the graph is flat-topped during working hours. A flat top means the link is saturated for that period. If neither appears but users still report problems, the issue is more likely to be prioritisation, jitter, routing or an application-side fault than raw capacity.

Should a growing business buy more bandwidth than it currently needs?

Buy the headroom, but buy it in the right form. Sizing to roughly 30–40% above your 24-month forecast peak is sensible, because contract terms typically outlive forecasts and because burst behaviour is not captured in percentile figures. Where a dedicated circuit is involved, the more valuable form of headroom is the bearer rather than the committed rate: ordering 100 Mbps on a 1 Gbps bearer costs modestly more and converts any future upgrade into a billing change completed within days, instead of a fresh installation taking three months. Paying for committed capacity you will not use for two years, by contrast, is simply an overpayment.

What does a business leased line cost in the UK in 2026?

On a 36-month term and excluding VAT, a 100 Mbps dedicated Ethernet circuit on a 1 Gbps bearer typically runs £250–£350 per month, and a full 1000/1000 Mbps service usually falls between £500 and £900. Price per megabit falls steeply as capacity rises, which is why the larger tiers look like better value per unit. The material variable is not the monthly rental but excess construction charges: where a building lacks existing fibre and civils work is required, the one-off cost can range from nothing to five figures. That figure is only knowable through a formal survey, which is free and takes two to four weeks, so request one early rather than budgeting from an indicative quote.

How long does it take to install a new business internet connection?

It depends entirely on the product. SoGEA and FTTP broadband services are typically delivered within 10–25 working days where the infrastructure is already present. Dedicated Ethernet circuits are a different proposition: 45–90 working days is normal, and where a road crossing, new duct or landlord wayleave is involved the upper end is realistic rather than pessimistic. These lead times cannot be shortened by paying more, which makes connectivity the most common critical path in an office move. Starting the conversation a full quarter ahead of the date you need service is the practical minimum, and earlier is better where the address has no existing fibre.

Does VoIP need a lot of bandwidth?

No - this is one of the most persistent misconceptions in connectivity planning. A G.711 call consumes roughly 87–100 Kbps in each direction once RTP, UDP and IP overhead is counted, so thirty concurrent calls need around 3 Mbps. Even a hundred simultaneous calls fits inside 10 Mbps. Voice quality problems are almost never capacity problems; they are problems of jitter, packet loss, and the absence of a quality-of-service policy that protects real-time traffic from bulk transfers. If calls degrade when a backup job or large upload is running, adding bandwidth will postpone the symptom rather than resolve it. Configure prioritisation first.

What is a 95th percentile and why use it instead of an average?

The 95th percentile is the level below which utilisation sits for 95% of measured intervals, discarding the top 5% as brief bursts. It is the standard measure in connectivity because it describes the capacity a link genuinely needs to sustain, without being distorted either by momentary spikes or by the long quiet periods that drag an average down. A circuit averaging 15% utilisation across a month can still be completely saturated for forty minutes every afternoon, and it is those forty minutes that generate complaints. The percentile figure catches that; the average conceals it entirely.

Can two broadband circuits replace a leased line?

Sometimes, and for a comparable total cost. Two diverse circuits from different carriers, managed by an SD-WAN appliance or bonded, can deliver more aggregate capacity and better practical availability than a single mid-sized dedicated line, because a fault on one carrier does not take the site offline. What that arrangement cannot provide is a single contractual fix-time commitment with service credits covering the service as a whole. For a business with regulatory obligations, contractual uptime commitments to its own clients, or life-safety systems now running over IP, that guarantee may be the actual requirement. The choice is a business judgement about risk rather than a technical comparison.

How does cloud backup affect bandwidth requirements?

Substantially, and in a way that degrades quietly over time. A backup job sized correctly against a 900 GB dataset three years ago may now be protecting several terabytes, with the incremental pass running well into the working day. Nothing alerts you to this, because the job still reports success. During its window a cloud backup can consume 30–100 Mbps of upstream capacity at a site level. Two controls resolve most of the impact: schedule the job to start early enough that it completes before staff arrive, and apply a bandwidth cap during core hours so that any overrun cannot saturate the uplink. Both are configuration changes rather than purchases.

What bandwidth do Microsoft Teams and Zoom video calls use?

A one-to-one HD call typically runs at 1.2–1.5 Mbps in each direction, while group calls with multiple video streams commonly reach 2.5–4 Mbps per participant, and a meeting-room system rendering several 1080p streams can exceed that. The figures are symmetric in nature: each participant both sends and receives. This is why video is the workload that most often exposes an asymmetric circuit’s upload ceiling. Both platforms degrade gracefully by reducing resolution or dropping to audio, which is useful for continuity but means a saturated uplink presents as inconsistent quality for individual users rather than as an obvious outage.

How often should bandwidth be reviewed once it has been sized correctly?

Annually as a standing item, and additionally whenever one of three triggers occurs: headcount changes by more than about 20%, a significant system moves to cloud hosting, or a new upload-heavy workload such as offsite CCTV or cloud desktops is introduced. Diarise the contract notice date separately from the review, at least ninety days before notice is due, so that renewal becomes an active decision rather than an automatic rollover onto a tariff set years earlier. The review itself is not onerous once monitoring is in place: pulling the percentile figures and comparing them against the forecast takes a couple of hours.

Related reading

These guides cover the adjacent decisions that most often come up alongside a connectivity review.

• Network monitoring for UK businesses: a proactive guide
• The hidden costs of an office move: IT relocation budget
• IT support SLA response times: a UK business guide
• Building an IT roadmap and technology strategy
• Cyber Essentials Plus: what the technical audit covers