How to measure the ROI of a domain-specific document AI honestly: the no-bullshit guide

In short.
Three hundred, five hundred, a thousand per cent: AI return-on-investment promises are often storytelling. An honest measurement takes three steps: establish the real cost of the current situation, measure the gain observed after deployment, and only then calculate the ratio.

Three hundred, five hundred, a thousand per cent: most AI ROI promises are storytelling. An honest measurement takes three steps, in that order. Figures that look more like sales arguments than rigorous measurements.

Optivalue.ai reports an average ROI of 428% in year 1. We are going to explain exactly how this figure is calculated, what assumptions it rests on, and how you can (with your own data) build your own calculation before committing to anything.

Because an ROI you cannot reproduce is not an ROI. It is a promise.

Step 1: identify the real cost of your current situation

Before measuring what AI saves you, you need to measure what your organisation is losing today. This is the step most business cases skip, and it is precisely where the calculations become dishonest.

The cost of expert time

Start by identifying the profiles who handle questionnaires, due diligence, audits or tenders in your organisation. For each one, estimate:

• The number of questionnaires handled per month
• The average time per questionnaire (from receipt to submission)
• The fully loaded daily cost of the profile concerned

Monthly processing cost formula:

Monthly cost = Σ (no. of questionnaires × average time in days × fully loaded daily cost)


A concrete example, senior CISO:

• 8 questionnaires per month
• 2 days on average per questionnaire
• Fully loaded daily cost: €700
• Monthly cost: 8 × 2 × 700 = €11,200
• Annual cost: €134,400 for this profile alone

If your organisation has 3 profiles exposed to this kind of workload (CISO, head of legal, compliance manager), you probably have €300,000 to €500,000 in annual cost from handling questionnaires alone, before factoring in the costs of errors and missed opportunities.

The cost of errors and expired certifications

Questionnaires handled manually contain errors. Certifications cited after their expiry date, inconsistencies between the legal and technical sections, outdated data that was never updated. These errors have a cost:

• Lost tenders on non-compliance grounds (estimate your average deal size × award rate)
• Correction time after submission, when the error is flagged by the customer
• Reputational risk during an audit or due diligence

This cost is hard to quantify precisely, but it is real and systematically ignored in conventional business cases.

The opportunity cost

Every day an expert spends compiling answers is a day they do not spend on their strategic projects. For a CISO: the NIS2 roll-out, the IAM overhaul, the business continuity plan. For a head of legal: securing key contracts, regulatory watch, internal training.

Opportunity cost formula:

Opportunity cost = no. of days tied up × value created per day on strategic projects


This value is subjective, but the exercise of estimating it forces executives to answer an uncomfortable question: at what price do you accept that your best people do administrative work?

Step 2: measure the real gain after deployment

Once the current situation has been costed, you can build the post-deployment scenario. A word of caution: gains must be measured, not projected.

The gain in processing time

This is the most immediate and most measurable gain. Optivalue.ai customers observe on average an 85 to 92% reduction in processing time per questionnaire.

Time gain formula:

Monthly time gain = Current monthly cost × reduction rate


Example, same CISO:

• Current monthly cost: €11,200
• Observed reduction rate: 90%
• Monthly gain: €10,080
• Annual gain: €120,960

But be careful: this gain is only a real gain if the time freed up is reallocated to value-adding activities. If your CISO uses that time to handle more questionnaires at the same unit cost, you have not improved your ROI: you have increased your processing capacity.

Both outcomes are valid. They are not measured in the same way.

The gain in quality and reliability

Every answer generated by Optivalue.ai is sourced: document, page, date. The expired-certification error becomes structurally impossible: the system automatically flags any reference whose validity date has passed.

Quality is hard to put a direct monetary value on. Here is how to approach it:

• Current error rate on questionnaires (estimated or measured on a sample)
• Average cost of an error (correction time, commercial risk, customer impact)
• Expected reduction after deployment

Quality gain formula:

Quality gain = no. of questionnaires × current error rate × average cost of an error × reduction rate


The gain in coverage

Before Optivalue.ai, how many questionnaires did your team leave unanswered for lack of time? How many tenders were not submitted? How many suppliers were not assessed?

This coverage gain can be considerable, particularly for Procurement and Sales teams:

Example, bid management team:

• 4 tenders turned down this year for lack of capacity
• Average value of a tender: €800,000
• Usual success rate: 35%
• Estimated lost revenue: 4 × 800,000 × 35% = €1,120,000

Step 3: calculate the ROI

You now have the two variables you need.

ROI formula:

ROI (%) = [(Total annual gains − Annual cost of the solution) ÷ Annual cost of the solution] × 100


Worked example:

• Annual time gain: €120,960
• Estimated quality gain: €15,000
• Coverage gain (partial, conservative): €50,000
• Total gains: €185,960
• Annual Optivalue.ai cost: €35,000 (annual subscription)
• ROI = [(185,960 − 35,000) ÷ 35,000] × 100 = 431%

That is how the 428% figure is built. Not a promise extrapolated from an ideal case. A reproducible calculation, on conservative assumptions, with variables you can adjust to your context.

The practical recommendation: start with a quantified pilot

Before deploying at scale, choose a representative use case and measure it over 30 days.

1. Select 5 typical questionnaires from your organisation
2. Measure the current processing time on these 5 questionnaires
3. Process them with Optivalue.ai and measure again
4. Calculate your own reduction rate: it will differ from ours, and that is normal
5. Extrapolate over 12 months with your real volumes

Optivalue.ai is operational in under 7 minutes. The first questionnaire is processed the same day. You will have your pilot data within a week: not in six months.

Ready to build your own calculation? Our teams will help you set up your business case with your real data. Request a personalised demonstration →