Andrew Crossley

How can AI reduce operating costs?

SHORT ANSWER

AI can reduce operating cost when it shortens repeated, reasonably consistent tasks such as support triage, data extraction, research, first-draft responses and QA. Do not start with a generic savings percentage. Measure the current handling time, error or correction rate and cost of the target task, pilot the AI workflow with human review, then compare the full post-AI cost including model usage, tools and review time.

Answered by Andrew Crossley, AI Product Consultant & Product Partner · Updated 2026-08-22

Why it matters

Cost pressure is a common reason companies explore AI, but generic company-wide saving percentages are rarely useful without a measured task baseline.

Task-level savings are easier to defend because the company can show the before state, the new workflow, the correction rate and the net cost rather than extrapolating from a demo.

How it works in practice

  1. 1

    Inventory the repeated tasks

    Use real work logs or workflow data. Look for high-frequency tasks with clear inputs, a reviewable output and enough consistency to create a useful baseline.

  2. 2

    Measure the before state

    Record handling time, volume, error or rework rate and labour cost for the specific task before automating it. Without that baseline there is no honest saving to report.

  3. 3

    Pilot one workflow with review

    Keep a human in the loop while measuring how often the AI output is accepted, corrected or rejected. The correction work is part of the cost, not an exception to the calculation.

  4. 4

    Redeploy the reclaimed capacity explicitly

    Name what the time is now spent on: more customer conversations, faster case resolution, avoided backlog or delayed hiring. Unassigned time saving can disappear into the working week.

  5. 5

    Report net cost and outcome

    Model spend, tool subscriptions, integration, monitoring and review time all count. Compare the total with the baseline and include any change in quality or customer outcome.

Common mistakes

  • Announcing a saving before measuring the baseline.
  • Automating exception-heavy work first simply because the demo looks impressive.
  • Removing human review before the error and correction pattern is understood.
  • Counting time saved without deciding what that capacity will now produce or avoid.

FROM EXPERIENCE

Contact-centre lessons applied to AI

Contact-centre work taught me that cost per contact is shaped by why customers make contact, not only by how quickly an agent handles the conversation. Sometimes the best saving is removing the reason for the contact rather than automating the response.

AI is useful here as both measurement and execution: classify the contact reasons, fix avoidable product friction, then automate the repeated residual work where the baseline shows a real opportunity.

Frequently asked

What is a realistic AI cost saving?

There is no credible universal percentage. Measure it for the target workflow: baseline handling cost minus model, tooling, integration and review cost, adjusted for any change in quality or output volume.

Does AI cost reduction always mean redundancies?

No. In a growing small company the value may be avoiding or delaying the next hire, reducing backlog, increasing capacity or moving existing people onto higher-value work.

How long until an AI automation pays back?

Calculate payback from the specific implementation cost and measured monthly net saving. A high-volume simple workflow can repay quickly; a low-volume or exception-heavy workflow may never justify the build.

IN SHORT

  • Start with repeated tasks that have a measurable before state and reviewable output.
  • Use net task cost and quality rather than a generic savings percentage.
  • Fix avoidable root causes first, automate the residual work second, and redeploy reclaimed capacity explicitly.

Work with Andrew

A short call to work out whether the problem is strategy, scope or speed.

Work with Andrew

THE FRAMEWORK

The Crossley Method: idea to first revenue in seven stages

See the full method
  1. STAGE 1DiscoverWeek 1
  2. STAGE 2ValidateWeek 2
  3. STAGE 3PrototypeWeek 3
  4. STAGE 4Build MVPWeeks 3-4
  5. STAGE 5LaunchWeek 5
  6. STAGE 6First RevenueWeek 6
  7. STAGE 7ScaleOngoing

MORE ANSWERS

AI and product development

Will AI replace product managers?

No, but it is removing a large part of what product managers used to spend their week doing. AI already writes specs, summarises research, drafts tickets, analyses feedback and builds prototypes. What it cannot do is decide what a company should refuse to build, hold accountability for a metric, or persuade a room. Roles weighted towards documentation are shrinking; roles weighted towards judgement are becoming more valuable.

How do startups use AI?

Startups use AI in three places: inside the product as a feature that completes a user task, inside operations to remove repeated manual work, and inside go-to-market for research, content and support triage. For a small company, internal operational work is often the easiest first use to measure because the baseline task already exists and the team can compare time, quality and cost before and after.

Which AI tools should founders use?

Founders need four categories, not forty tools: a frontier chat assistant for thinking and drafting, an AI-assisted build tool for prototypes and MVPs, a research and synthesis tool for customer and market work, and automation for repeated operational tasks. Pick one per category, use it deeply, and add another tool only when a specific recurring task justifies the subscription and context-switch cost.

How much does it cost to build an AI application in the UK?

For a UK founder, a focused commercial AI MVP with one core journey typically sits around £12,000–£25,000 in the pricing model used on this site. A prototype or narrow validation build can be roughly £3,000–£8,000. Multiple user roles, RAG or agentic workflows, several integrations, sensitive-data requirements or bespoke design can push the build above £25,000. Model and infrastructure running costs should be measured separately per successful customer task.