Digital transformation ROI: a measurement framework for leaders

In short: Digital transformation ROI (return on investment) compares the net benefit of a digitisation, automation or AI project with its total cost over a fixed period, usually one to three years. To make the number credible, measure a baseline before the project starts, count hidden costs such as training, data work and staff time, and only put a money value on benefits you can show are actually reused.

What is digital transformation ROI?

It is the net benefit of a technology project expressed as a percentage of what the project cost. The formula fits on one line:

ROI = (Total benefits − Total costs) / Total costs × 100%

The arithmetic is easy. The inputs are not. Costs tend to be understated because leadership looks only at the software invoice. Benefits are spread across teams and come in different shapes: a few hours saved per day, fewer wrong deliveries, a report that arrives two days earlier. Turning those into one figure requires assumptions, and vague assumptions are exactly what a finance team will strike out.

This gap between "useful" and "profitable" is not limited to small firms. McKinsey's The state of AI in 2026 survey found that 80% of respondents said AI had improved their individual productivity, while only about 37% reported any contribution from AI to their organisation's EBIT, roughly unchanged from the previous year. Earlier, Gartner (2024) predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing unclear business value alongside poor data quality, weak risk controls and rising costs.

The four steps below work for management software, process automation and AI projects alike.

Step 1: Measure the baseline before you start

The baseline is how the process performs before the new tool arrives: how long it takes, how many people it uses, how often it goes wrong. Without it, nobody can answer "how much better is it?" six months later, because everyone remembers the old way differently.

For each process in scope, the process owner records three to five measures over two to four normal working weeks:

• Time to handle one unit of work (an order, an invoice, a candidate profile), timed or taken from software logs rather than estimated.
• Monthly hours spent by each role involved.
• Error rate, plus the average cost of fixing one error.
• Volume: orders, documents or quotes per month.
• A customer-facing measure where relevant, such as response time or complaints.

Avoid peak season and holidays. Whoever measures before should measure after, the same way. Choosing which processes to measure belongs to the assessment phase described in where to start digital transformation.

Step 2: List all six cost categories

Most project budgets cover the first three categories below and forget the last three, which sit in your own payroll.

Cost category	What it includes	Where it gets missed
Licences and subscriptions	SaaS fees per user, usage-based AI model fees	Fees rise as users or AI usage grow
Implementation and integration	Configuration, API connections to accounting, inventory and sales systems	Extra work when the real process differs from the brief
Training	Sessions, materials, staff time spent learning	Hours of the people being trained
Data	Cleaning, standardising product and customer codes, migration	Time of the people who know the data
Change management	Manager time, coaching, running old and new processes in parallel	Productivity often dips before it rises
Run cost	Maintenance, support, updates, a system owner, AI monitoring	Recurs every year, not just in year one

Convert internal time to money using a fully loaded hourly cost (salary, insurance and allowances divided by working hours). Then split costs into one-off and recurring columns, because the payback calculation needs that split. Change management deserves its own plan; see change management in digital transformation.

Step 3: Put a money value on benefits

Benefits usually fall into four types, each with its own conversion rule. The general rule: only count a benefit if you can say where it goes.

Benefit type	How to value it	Condition
Time saved	Hours saved × hourly cost	Freed hours are reassigned to valuable work, or avoid a new hire
Fewer errors	Errors avoided × average cost per error	A baseline error count exists
More revenue	Extra revenue × gross margin	The uplift can be separated from seasonality, pricing and promotions
Lower risk	Probability of incident × estimated loss	Assumptions stated; usually shown separately, not added to headline ROI

Time saved is the easiest benefit to inflate. Two hours less data entry per day only becomes money when the manager fills those hours with other work, such as calling back past customers, or when volume grows without hiring. Write down where the freed time will go in the project proposal itself. Revenue, likewise, counts at gross margin: VND 100 million of extra sales at a 20% margin adds VND 20 million to the benefit side.

Step 4: Calculate ROI and payback

With the inputs ready, two figures answer most board questions:

• ROI = (Total benefits − Total costs) / Total costs × 100%, over a fixed period such as three years.
• Payback period = One-off costs / (Monthly benefit − Monthly recurring cost), plus the implementation months with no benefit yet.

A hypothetical example

This example is invented to show the method; it is not data from a real company. A distributor with about 50 staff has three people typing orders received by chat and email into its sales system, about four hours each per day. AI reads the messages and drafts the orders; staff only check and approve. Assume a fully loaded cost of VND 60,000 per hour.

Item	Assumption	Amount
Baseline	3 people × 4 h × 22 days = 264 h/month
After rollout	3 people × 1.5 h × 22 days = 99 h/month
Time benefit	165 h × VND 60,000	VND 9.9 million/month
Error benefit	Fewer wrong deliveries and returns	VND 2 million/month
One-off cost	Implementation 80m, data 15m, training 10m, change management 10m	VND 115 million
Recurring cost	Subscription and AI fees 3m, run cost 1m	VND 4 million/month

With two months of implementation before benefits begin:

• Year one: benefits of 10 × 11.9 = VND 119 million against costs of VND 163 million. Year-one ROI is about −27%.
• Over three years: benefits of about VND 404.6 million against costs of VND 259 million, an ROI of about 56%.
• Each live month nets VND 7.9 million, so the project pays back around month 18.

A negative first year is normal for projects with a large one-off cost. Tell the board in advance so nobody calls it a failure at month six.

Now assume the manager only manages to reuse half of the freed hours. Monthly benefit drops to about VND 6.95 million, the net gain to VND 2.95 million, and payback stretches to around month 44, almost four years. One assumption moved payback from about a year and a half to nearly four years. Every proposal should therefore show an expected case and a cautious case, and approvers should question the assumption with the biggest effect rather than the headline ROI.

Leading and lagging indicators

ROI is a lagging indicator: you only see it after months. Leading indicators move early and tell you whether the project is on track.

Type	What it shows	Example KPIs	Review
Leading	Are people really using it?	Weekly active users, share of orders going through the new flow	Weekly
Leading	Output quality	Share of AI drafts approved without edits, share handed to a person	Weekly
Lagging	Operating cost	Cost per order, hours per month	Monthly
Lagging	Customer quality	Wrong deliveries, complaints, response time	Monthly
Lagging	Financial	Revenue per employee, margin, ROI, payback	Quarterly

If weekly usage is low after a month, the benefits in the spreadsheet will never arrive. Catch it early and fix the process or the training. Building a dashboard for these KPIs is covered in the data-driven business.

Common measurement mistakes

1. No baseline. The most expensive mistake, because it cannot be fixed afterwards.
2. Treating hours saved as money saved without saying where the hours go.
3. Ignoring internal costs such as data work, training and parallel running.
4. Looking only at year one, when subscriptions run for years.
5. Crediting the project with all growth instead of comparing with the same period last year or a group that has not adopted the change.
6. No owner for the numbers. Each KPI needs a named person, usually the head of the team that runs the process.

When ROI should not be the main test

Three kinds of spending need other criteria alongside ROI. Legal obligations, such as e-invoicing or personal data protection under Vietnam's Personal Data Protection Law (No. 91/2025/QH15), are about compliance at the lowest cost. Data foundations, such as one clean product list or one customer database, rarely pay back on their own but make every later AI and reporting project cheaper; judge them as part of the whole digital transformation roadmap. Small four-to-six-week experiments should be judged on what you learned and whether to continue, not on a three-year ROI model.

On the other hand, a process that takes a few hours a month and rarely goes wrong is not the first place to spend money, even if the ROI comes out positive. Business process automation explains which processes to automate first.

FAQ

What is a good digital transformation ROI? There is no universal threshold. Compare ROI and payback with your cost of borrowing and with other investments on the table, and agree the acceptable threshold before you calculate.

How soon can we measure the ROI of an AI project? Leading indicators such as usage and draft approval rates are visible after two to four weeks. ROI needs several months of stable operation, so set the first ROI review at around six months.

Do small businesses need an ROI model? Yes, in a compact form: one page, one process, a few measures. The smaller the business, the more each spend needs a clear reason.

Should soft benefits like staff satisfaction go into ROI? Record them, but keep them separate from the headline figure. Mixing in benefits that are hard to verify undermines the whole model with your finance team.

A vendor promises a very high ROI. Should we trust it? Ask for the inputs: where the baseline comes from, what hourly cost was used, and whether internal costs are included. An ROI figure without its assumptions is not ready for a decision.

---

HAT Solutions helps businesses set the baseline, build a cost and benefit model with two scenarios before choosing tools, and track KPIs after rollout. Send us one process you are thinking of automating and talk to us about how to measure it. Related reading: AI process automation vs RPA and AI for business.