Case Study · AI Automation Program
AI Automation ROI Tracking Platform Case Study
Bringing automation ROI into one shared view
How Decision Foundry and a business process services provider built a shared platform for tracking what automation is delivering and deciding where to automate next.
The result: one shared view of the AI automation program across every customer, project and workstream, with realized and projected savings kept separate and every new opportunity scored on five weighted factors.
The customer
The client is a business process services provider delivering operations work for clients across insurance, finance and accounting, mortgage, healthcare, data management, software and voice.
Decision Foundry works with the client as an automation partner. We design and build automations that the client's teams use to deliver services to their own customers, covering work such as policy servicing, loan document handling, eligibility checks and accounts payable.
Over time the automation program grew to span several customers and verticals, each with its own workflows, owners and stage of delivery. That growth is what led to this project.
At a glance
- Client
- A business process services provider
- Industry
- Business process services (BPO)
- Verticals served
- Insurance, F&A, Mortgage, Healthcare, Data Management, Software, Voice
- Solution
- A shared platform for automation ROI tracking and opportunity prioritisation
- Built with
- Next.js, Supabase (PostgreSQL, Auth), AWS Amplify, Google and Microsoft sign-in
The problem
As the program expanded, information about it lived in several places: delivery trackers, spreadsheets, and regular conversations between the DF delivery team and client business owners. Each source served its purpose well, but bringing them together to answer portfolio-level questions took time, and the picture changed as projects moved forward.
Client and DF leadership wanted a shared, dependable answer to two questions:
- What is automation delivering today? Hours and cost saved by automations that are live, kept separate from what is expected from work still in progress.
- Where should we automate next? A consistent way to look across the client's process inventory and decide which opportunities to take forward.
The aim was not another reporting layer. It was one place that both teams trusted and could use in the same conversation.
Challenges faced
Separating delivered value from expected value.
Automations can show measurable time savings while still in testing, but those savings should only count as realised once the automation is in production. Combining the two would overstate results.
Working with incomplete inputs.
Hourly rates are not always available for every process. The platform needed to show a blank rather than an estimate, so that every cost figure on screen comes from actual inputs.
Using a shared vocabulary.
The client and DF use some terms differently. For example, the client refers to user acceptance testing as “Calibration”. The platform uses the client's terms, with clarifications where helpful, so both teams read the same screen the same way.
Giving the right access to the right people.
Users from two organisations needed different levels of access: administrators, delivery owners who update their own work, and viewers. These permissions had to be enforced in the database, not just hidden in the interface.
Prioritising a large process inventory.
The client's process inventory covers revenue, headcount and vendor costs across many customers. Comparing opportunities fairly needed a consistent and explainable scoring method.
Keeping it focused.
Client leadership asked for a simple first version without notifications or extra features. New capabilities were added only in response to their feedback.
The solution: one platform for the AI automation program
Together with client leadership, we built a web platform with two connected parts:
Automation ROI tracking:
every automation workstream, organised by customer, project and workflow, with its delivery stage, owners, and before-and-after effort. The platform calculates hours saved, annual saving and automation efficiency, and labels each figure as Realized (live) or Projected (in progress).
Automation opportunity platform:
The client's process inventory, imported from Excel, with monthly revenue, headcount and vendor cost for each process. Each opportunity is scored on five weighted factors and placed in a High, Medium or Low band, with a short plain-language explanation of the score.
The platform was built in short iterations. Client stakeholders reviewed working versions and their feedback shaped what came next: the customer-level view, how stages are shown, and the opportunity scoring model.
Inside the platform: tracking AI automation ROI
1. Customer Portfolio
The starting point for leadership. Portfolio totals for annual savings, hours saved and average efficiency are shown at the top, each split into live and pipeline values. Below them, short insight cards point to the customer with the highest efficiency, the largest contributor to savings, the most activity in build, and any customer that needs attention. Customers are grouped into those already realising ROI and those still building, and each card shows its own savings and delivery stages.
2. Automation ROI Tracker
Every automation workstream in one table, organised into three bands: business context (vertical, owner, customer, project), delivery progress (workflow documentation, discovery, feasibility, customer approval, build, testing, deployment) and business impact (hours saved, annual saving, efficiency). Live workstreams are shown in full colour and in-progress ones are muted, so delivered and expected value are easy to tell apart. Where an hourly rate has not been provided, the saving shows a dash instead of an estimate.
3. Project Detail
A single project (here, a policy servicing engagement) with its automations, owners, stages and client discussion status. Summary figures count only deployed automations. Each row shows whether its hours and savings are Realized or Projected, so that expected results are never presented as delivered ones.
4. Automation Opportunity Inventory
The client's process inventory, imported directly from the Excel template the operations teams already maintain. Each process shows its customer, service, delivery model (in-house or vendor), headcount, monthly invoice value, vendor cost and margin, along with month-by-month revenue for the year. It can be sorted by revenue, headcount, cost or margin to see where the largest opportunities are.
5. Opportunity Scoring
Each process is rated on five weighted factors: automation value potential (30%), feasibility (25%), operational improvement potential (20%), opportunity clarity (15%), and risk and compliance impact (10%). Each rating has a plain-language description, and the combined score places the process in a High, Medium or Low band, along with a short summary explaining why. This gives the client and DF a consistent, transparent basis for deciding what to automate next.
How it was built
Secure sign-in with existing Google and Microsoft accounts for both DF and client users.
Role-based access (Admin, Owner, Viewer), enforced by database-level security policies.
A single source of truth: data is entered and maintained by the teams themselves, with Excel import for the process inventory. No external system integrations are required.
A transparent calculation engine: hours saved = hours before − hours after; annual saving = hours saved × hourly rate × 12; efficiency = hours saved ÷ hours before. Every formula is visible and covered by automated tests.
Hosted on AWS Amplify with a Supabase PostgreSQL database.
The outcome for the AI automation program
The platform gave the client and DF a common reference point for automation: one place to see what is live, what is in progress, what it is delivering, and which opportunities are worth taking forward. Conversations about the program could start from the same numbers, with realised and projected value clearly separated.
Get In Touch