The four things to measure in an AI ROI case
A credible ROI case for AI in a public sector service area covers four dimensions. Presenting all four is more persuasive to a finance director or chief executive than a single headline figure, because it shows the full picture of what changed and why it matters.
Dimension 1 - Officer time saved
The question it answers: How much officer time did this free up, and what is that worth?
How to calculate it: Time saved per case or transaction (in minutes or hours) × number of cases or transactions per month × officer hourly cost (salary + on-costs ÷ working hours). Express as hours per year and as annual salary equivalent. Cross-check against the deployment cost. The payback period is typically expressed in months.
What Arto measures: Run volume and outcome distribution tracked from the first workflow execution. Per-run time saving benchmarks from the POC Flow Specifications (see Section 3) can be applied to your council's actual transaction volume to calculate annual time saved.
Dimension 2 - Compliance risk avoided
The question it answers: What legal, financial or reputational exposure has been reduced?
How to calculate it: Identify the specific compliance risks the workflow addresses. For planning: statutory clock failures and appeal costs. For SEND: Ombudsman upheld complaints and tribunal applications (average costs documented in LGO annual reports). For enforcement: Ombudsman compensation awards (£500–£2,000 per upheld case) and enforcement agent fees on vulnerable cases (£310 minimum per case). For revenues: arrears accumulation from delayed billing. Express as risk reduction, not as guaranteed savings.
What Arto measures: Assurance case records and audit trail demonstrate governance on every run.
Dimension 3 - Demand reduced
The question it answers: Have we reduced inbound demand on a service, freeing capacity or avoiding staffing cost?
How to calculate it: Baseline demand measurement (calls, cases, transactions) before deployment. Same measurement after a defined period. Express as percentage reduction and as absolute volume reduction. Apply the cost-per-contact or cost-per-case metric from the council's service costing model to convert volume reduction to financial value.
What Arto measures: Monitoring dashboard tracks run volume and outcome distribution. Contact centre demand reduction is measured against the organisation's own baseline. The service lead inputs their baseline volume and the dashboard shows reduction against it.
Dimension 4 - Capacity freed for higher-value work
The question it answers: What can officers now do that they could not do before, and what is that worth?
How to calculate it: This dimension is harder to quantify but often the most persuasive for senior leaders. Identify what officers were previously doing that is now automated (validation letter drafting, manual chasing, calculation tasks) and what they are now doing instead (complex casework, resident contact, professional judgement work). Express qualitatively in the leadership presentation, with supporting evidence from officer feedback.
What Arto measures: Run History tab shows what each workflow execution processed. Over time, the Monitoring dashboard shows the distribution of case outcomes, showing the ratio of automated resolutions to human-review cases and how officer capacity is being reallocated.
Estimated time savings from Arto workflow proof of concept specifications
Service area | Workflow | Estimated time saving | Scale at a typical council | Annual impact estimate |
Planning | Pre-submission application validation | 30–45 minutes per invalid application | ~30% of applications nationally are invalid at first submission. Council receiving 1,500 applications per year: ~450 invalid applications. | ~225–338 officer hours per year on validation letters alone. At median planning officer salary, 200+ officer days. |
Planning | Consultation analysis | 3–5 hours per consultation summary | High-volume contested applications generate 100+ consultation responses. Councils processing 200+ committee applications per year. | Significant reduction in officer hours on committee report drafting. Estimate using your committee report volume × hours per report. |
Children's Services (SEND) | EHC plan annual review | 4–6 hours per week per SEND coordinator in chasing and scheduling | Each coordinator manages 150–300 plans. Statutory compliance improvement from ~36% to near-100% on-time reviews. | Full-time equivalent hours recovered per coordinator. Compliance risk: Ombudsman investigations of late EHC reviews. |
Children's Services (MASH) | Referral triage | 20–40 minutes per referral for structured triage analysis | Typical MASH team receives 50–150 referrals per week. | Structured triage analysis available for every referral. Consistency improvement addresses Ofsted quality concern. |
Revenues and Benefits | Change of circumstances | 45–90 minutes per change reduced to under 5 minutes for standard automated cases | Typical billing authority processes 400–600 changes per week. 70–80% of standard changes fully automated. | At 500 changes per week and 70% automation rate: 350 changes per week automated. At 45 minutes each saved: ~260 officer hours per week. |
Revenues and Benefits | Enforcement vulnerability screening | Reduction in Ombudsman complaints, compensation payments and enforcement agent fees on vulnerable cases | LGO upholds thousands of enforcement complaints annually. Compensation: £500–£2,000 per upheld case. Enforcement agent fees: £310 minimum per case. | Financial risk mitigation depends on enforcement queue volume and proportion of vulnerable households. Primary ROI case is risk avoided, not time saved. |
Contact Centre | Demand reduction | 28% reduction in contact centre demand | Redcar and Cleveland Council deployment. Live result, not a POC estimate. | Live result - Redcar and Cleveland Council. Within three months of deployment. |
What the Monitoring dashboard records automatically
Arto Supported Flows generate measurement data from the first workflow execution. The Monitoring dashboard surfaces this data in real time. A service lead can produce a current-state picture of what the workflow is doing at any point from the day it goes live, without manual reporting or data extraction.
These measurements create the evidence base for an ROI presentation. Run volume multiplied by estimated time saving per case produces a projected hours-saved figure. Outcome distribution shows the proportion of cases handled automatically versus those requiring officer involvement. The governance score demonstrates that efficiency has not come at the cost of compliance.
-
Run volume:
total workflow executions by service area and by time period. The volume trend chart shows the last 7 days by service area and by outcome.
-
Outcome distribution:
how many workflow executions were resolved automatically, referred to officer review (HITL queue), or escalated. This is the metric that shows what proportion of cases the AI is handling and where officer time is being applied.
-
Governance score:
derived from Assurance Designer completion across active workflows. Provides the evidence that governance standards are being maintained alongside operational performance.
-
Active alerts:
workflow executions that have generated a governance issue requiring attention. Ensures that operational performance data is read alongside governance health.
-
Run log:
a timestamped record of recent individual workflow executions. The starting point for case-level audit and for reviewing the detail behind the volume figures.
Structuring an AI ROI presentation for senior leadership
Senior leaders in UK local government are not persuaded by technology claims. They are persuaded by evidence that a problem has been solved, that the legal position is clear, and that another council has done it successfully. An effective AI ROI presentation includes five elements.
1. The problem in your service area - before AI
A specific, quantified description of the operational problem the workflow addresses. Not 'we are under pressure' but: 'Our planning team writes 450 validation letters per year at an estimated 35 minutes each, representing 263 officer hours on pure administration. This is equivalent to one full planning officer month per year on form-checking.'
Why it lands well: Quantified problem statements are taken seriously because they show the service lead understands the issue precisely. A vague problem framing signals the proposal has not been properly assessed.
2. The workflow and what it changes
A plain-English description of what the AI workflow does and what changes as a result. 'Arto's planning validation workflow reads every incoming application and checks it against the council's Local Validation Checklist automatically. Invalid applications are identified within seconds and a validation letter drafted without officer involvement.'
Why it lands well: Senior leaders and finance directors do not need to understand how AI works. They need to understand what specifically changes in the process: what officers stop doing, what they start doing, and what residents experience differently.
3. The governance position
Confirmation that the AI deployment is legal, governed and approved. 'The DPO has signed off the deployment. Data is processed on UK-hosted infrastructure. Every workflow execution produces an immutable governance record and a named officer reviews and approves all decisions affecting residents.'
Why it lands well: For public sector senior leadership, governance is the threshold question. A proposal that does not address governance clearly will be blocked. Addressing it proactively signals that the deployment is properly managed and removes the primary reason for refusal.
4. The projected return - with your council's numbers
Your council's specific numbers applied to the POC time-saving benchmarks. 'We process 480 changes of circumstances per week. The POC specification estimates 70–80% automation. At 70%, that is 336 changes per week processed automatically at an estimated saving of 45 minutes each, approximately 252 officer hours per week. Applied to our Grade 3 revenues officer salary, the annual time saving is equivalent to [X] FTE.'
Why it lands well: Organisation-specific projections built from your own data are more persuasive than generic benchmarks. The process of building them also demonstrates that the service lead has done their due diligence. The finance director can challenge the assumptions and the service lead can explain each one.
5. The live result from another council
Redcar and Cleveland Council reduced contact centre demand by 28% within three months of deployment using Arto's contact centre workflow.
Why it lands well: Named, attributed, specific results from comparable organisations are the most persuasive evidence available to a public sector senior leadership audience. 'Another council did this and it worked' addresses both the risk concern and the feasibility question simultaneously.
The live result: Redcar and Cleveland Council
28% reduction in contact centre demand
Redcar and Cleveland Council deployed Arto's contact centre workflow and reduced inbound contact centre demand by 28% within the first three months. This is a live deployment result, not a projected saving or a proof of concept estimate. It is the result a named council achieved from a specific AI workflow in a specific service context.
The full detail of the deployment is in the contact centre use case and the Redcar case study: which workflows were configured, what the governance looked like, and what changed operationally.
Where to go from here
Explore the workflow library
Pre-built workflows for planning, children's services, revenues and benefits, contact centre, housing and adult social care, with estimated time savings for each.
Workflow libraryTalk to us about your service area
We can help you build an ROI projection for your council's specific transaction volumes and staff costs.
Get startedRead the contact centre case study
The full Redcar and Cleveland deployment: what the workflow did, what the governance looked like, and what the 28% figure represents.
Case study