Latency

The invisible tax on digital experiences — and why milliseconds translate directly into millions.

Amazon calculates that every 100 milliseconds of latency costs it 1% in sales — roughly $1.6 billion annually at its current revenue scale. Google found that an additional 500 milliseconds of delay in search results reduced traffic by 20%. Latency — the time it takes for data to travel from one point to another and back — is not merely a technical metric; it is a business-critical variable that affects revenue, customer satisfaction, and competitive positioning. In an era where AI-powered applications must respond in real time and global users expect instant gratification, latency management has become a board-level concern.

What It Is

Latency is the time delay between a request being sent and a response being received. In computing and networking, it is typically measured in milliseconds (ms), though for high-frequency trading and certain scientific applications, it is measured in microseconds (μs) — one-thousandth of a millisecond. Think of latency as the "travel time" for data. Just as a letter mailed from London to Singapore takes days to arrive, a data packet travelling from a server in London to a user in Singapore takes time — typically 150-250 milliseconds for a round trip, depending on the route and network conditions.

Latency is influenced by several factors: the physical distance between sender and receiver (data cannot travel faster than the speed of light in fibre optic cables, which imposes a theoretical minimum of about 100ms for a London-to-Singapore round trip), the number of intermediate network hops (each router or switch adds processing time), server processing time, and the efficiency of the application itself. In modern web applications, a typical user request might traverse 10-20 network hops between their device and the origin server, with each hop adding 1-5 milliseconds of delay. The total end-to-end latency for a simple web page load averages 3-8 seconds globally, though users perceive anything beyond 2 seconds as slow.

Distinguishing latency from bandwidth is crucial. Bandwidth is the capacity of the pipe — how much data can flow simultaneously. Latency is how long it takes for the first drop to arrive. A motorway with ten lanes (high bandwidth) does not help if the journey is 500 miles (high latency). Many organisations invest in bandwidth when their real problem is latency, particularly for interactive applications where small amounts of data must be exchanged rapidly.

Why It Matters for Business

The financial impact of latency is well-documented across industries. In e-commerce, a 1-second delay in page load time reduces conversions by 7%, according to research by Akamai. For an online retailer generating $100 million in annual revenue, that translates to $7 million in lost sales. In financial services, latency is measured in microseconds and carries even higher stakes. Trading firms invest hundreds of millions of dollars in low-latency infrastructure — including laying dedicated fibre optic cables and building data centres physically adjacent to stock exchanges — because a 10-microsecond advantage in trade execution can be worth over $100 million annually in arbitrage opportunities.

In the AI era, latency takes on new dimensions. Large language model inference typically requires 500-3,000 milliseconds to generate a response, and users expect conversational interactions to feel natural — which means sub-2-second response times. Organisations deploying AI-powered customer service, real-time translation, or interactive assistants must carefully optimise their inference infrastructure for latency, often placing models at edge locations close to users rather than in centralised data centres. Autonomous vehicles require inference latencies under 100 milliseconds to make safe driving decisions at motorway speeds, where a car travels nearly 3 metres in the time it takes a model to process a single frame.

Latency also affects employee productivity. Research by Sandisk found that slow enterprise applications cost the average knowledge worker 6 hours per week in waiting time. For a company with 10,000 employees, that represents 3 million hours of lost productivity annually — equivalent to over $150 million in labour costs at average enterprise salary levels. Internal systems with high latency — CRM platforms, ERP systems, internal tools — silently drain organisational efficiency.

Enterprise Use Cases

Cloudflare operates one of the world's largest edge networks specifically to combat latency, with data centres in over 330 cities across 120 countries. By caching and serving content from locations physically close to users, Cloudflare reduces average round-trip latency from 100-200ms to 10-30ms for cached content. Companies using Cloudflare's enterprise services report 30-50% improvements in page load times. In gaming, Riot Games (maker of League of Legends) built a custom network backbone to reduce latency for its 180 million monthly players. In competitive online gaming, even 20 milliseconds of additional latency can disadvantage players, so Riot invested in dedicated server infrastructure in over 30 regions globally, reducing average player latency from 80ms to under 35ms.

Netflix operates a global content delivery network called Open Connect, with servers deployed inside internet service provider facilities worldwide. By pre-positioning popular content close to viewers, Netflix achieves an average start-up latency of under 2 seconds — compared to the 5-10 seconds typical of traditional streaming architectures. This investment in latency reduction is credited with reducing subscriber churn by an estimated 5-10%. In financial services, Citadel Securities processes over 25% of all US equities trading volume with average order-to-execution latency under 10 microseconds, enabled by custom hardware and co-located servers in the same buildings as exchange matching engines.

Vendor Landscape

The latency optimisation market spans several categories. Content Delivery Networks (CDNs) — Cloudflare, Akamai, Fastly, and AWS CloudFront — reduce latency by caching content at edge locations. Cloudflare leads with the broadest global footprint; Fastly differentiates with sub-millisecond cache purging for dynamic content. Enterprise pricing typically ranges from $5,000-50,000 per month depending on traffic volume. Edge computing platforms — AWS Wavelength, Azure Edge Zones, Google Distributed Cloud — allow organisations to run compute workloads at the network edge, reducing latency for applications that cannot rely on cached content alone. Network monitoring and optimisation vendors include ThousandEyes (acquired by Cisco for $1 billion), Datadog Network Monitoring, and Catchpoint, which provide visibility into latency across complex application architectures.

Strategic Considerations

Establish latency budgets for every customer-facing application — explicit targets for maximum acceptable latency at each tier of the experience. Measure real user latency, not synthetic benchmarks; the gap between lab conditions and real-world performance is often 3-5x due to network variability, device differences, and geographic distance. Prioritise latency optimisation for revenue-critical paths: checkout flows, search results, and customer-facing AI interactions deliver the highest ROI for latency improvements. For global organisations, invest in edge infrastructure (CDN, edge compute) early — the performance difference between serving users from a single region versus a global edge network is dramatic. Consider the latency implications of technology choices: adopting a microservices architecture can improve development agility but adds network latency between services. Mitigate this through service mesh technology, strategic service colocation, and careful API design.