Short answer: training ROI is (net benefit ÷ total cost) × 100 – but the number is only as defensible as its two inputs. The Phillips methodology adds the discipline: count all the costs (participant wages during training are usually the biggest, most-forgotten line), monetize only benefits you can trace to a metric, and then discount for attribution – the honest admission that training wasn’t the only thing driving the improvement. Done that way, a “70% ROI” survives a CFO’s questions. Done sloppily, a “300% ROI” gets your next training budget cut, because nobody believes it.
The five levels, in one table
Donald Kirkpatrick’s four evaluation levels have been the L&D standard since 1959; Jack Phillips added the fifth:
| Level | Question | Evidence |
|---|---|---|
| 1. Reaction | Did they like it? | Post-course survey |
| 2. Learning | Did they learn it? | Test / demonstration |
| 3. Behavior | Are they doing it at work? | Observation, 30-90 days later |
| 4. Results | Did a business metric move? | Error rate, output, retention, sales |
| 5. ROI (Phillips) | Was the metric movement worth the cost? | Monetized Level 4 vs full cost |
The practical insight for a small business: most training evaluation stops at Level 1, and happy-sheet scores have close to zero correlation with business impact. You don’t need a measurement department to reach Level 5 – you need one metric chosen before the training, and the arithmetic below.
Step 1 – Count the full cost
Direct costs are the easy half. The complete inventory:
| Cost | Example (12-person customer-support training) |
|---|---|
| Trainer / course fees | ₹1,20,000 |
| Materials, platform, certification fees | ₹25,000 |
| Venue / travel (if any) | ₹15,000 |
| Participant wages during training – 12 people × 24 hours × ₹450/hr loaded | ₹1,29,600 (round ₹1,30,000) |
| Coordination and admin time | ₹20,000 |
| Total cost | ₹3,10,000 |
Note the shape: wages-during-training are ~42% of the total, and they never appear on an invoice. Leaving them out understates cost by nearly half and produces the inflated ROI numbers that destroy credibility. (The Training Budget Calculator itemises these lines; the Learning Hours Calculator checks whether the hours are even available.)
Step 2 – Monetize the benefit (conservatively)
Pick the Level 4 metric before training, measure it 90 days after. Standard monetizations:
- Productivity: (output gain %) × affected salary base
- Error/rework reduction: errors avoided × average cost per error
- Retention: reduced departures × cost per departure
- Sales: margin (not revenue) on attributable incremental sales
For our support team, 90 days later: average handle time down 12%, escalations down 20%, and one agent who had signalled leaving stayed. Raw monetization:
| Benefit | Computation | Annualised |
|---|---|---|
| Productivity gain | 12% × (12 agents × ₹4.5L salary base) | ₹6,48,000 |
| Escalation reduction | 300 fewer escalations × ₹400 each | ₹1,20,000 |
| One retention save | ~60% of ₹5.5L salary | ₹3,30,000 |
| Raw total | ₹10,98,000 |
Step 3 – The attribution discount (the step everyone skips)
Did training cause all of that? Almost certainly not – a new macro tool shipped the same quarter, and the retained agent also got a raise. Phillips’ answer where controlled studies aren’t feasible (they never are at small-business scale) is participant estimation with a confidence adjustment: ask participants and managers what share of the improvement came from the training, then multiply by their confidence in that estimate.
Suppose managers attribute 60% of the gains to training, with 80% confidence:
Adjusted benefit = 10,98,000 × 0.60 × 0.80 ≈ ₹5,27,000
Yes, it’s an estimate stacked on an estimate – but it’s a documented, conservative one, which beats both the naive 254% ROI (raw ÷ cost) and the vague “training is valuable” argument. Write the attribution assumptions down; they’re the first thing a skeptical CFO asks about, and having them ready is what makes the number credible.
Step 4 – Compute and interpret
ROI = (5,27,000 – 3,10,000) ÷ 3,10,000 × 100 ≈ 70%
BCR = 5,27,000 ÷ 3,10,000 ≈ 1.70 (₹1.70 back per ₹1 spent)
Read against the interpretation bands (same ones the Training ROI Calculator uses): negative = reassess design or assumptions · 0-50% = common for early-stage and soft-skills programs, track longer · 50-100% = solid, benefit meaningfully outweighs cost · 100%+ = strong outcome. 70% with conservative attribution is a program worth repeating – and worth presenting as “₹1.70 per rupee, after discounting for what training didn’t cause,” which is a sentence stakeholders trust.
Prefer BCR for non-financial audiences: “₹1.70 back per ₹1” lands faster than “70% ROI,” and it can’t be confused with the raw-vs-net ROI ambiguity.
When NOT to run the ROI math
Phillips himself is explicit that Level 5 isn’t for everything. Skip the ROI calculation for:
- Compliance training – the benefit is avoided legal risk; the decision is made by law, not ROI.
- Sub-₹50,000 programs – measurement effort exceeds the stakes; Level 3 (are they doing it?) is enough.
- Long-horizon leadership development – benefits mature over years; a 90-day ROI reads falsely negative. Use Level 3 behavior evidence and revisit at 12-18 months.
Reserve full Level-5 treatment for the few programs that are expensive, repeatable, or contested – that’s where a defensible number changes a decision.
Closing the loop with the skill gap
The cleanest ROI cases start from a skill gap analysis: the gap defines the metric (SQL 2→4), the training budget defines the cost, the 90-day re-rating defines Level 2/3 evidence, and the metric movement defines Level 4. When the pipeline runs in that order, the ROI number is a by-product of decisions you were making anyway – not a retrofitted justification.
FAQ
What’s a good training ROI benchmark? There’s no audited industry number – published claims vary wildly because attribution practices vary wildly. Use the bands above, and compare your own programs against each other rather than against vendor marketing.
Over what period do I count benefits? Match the skill’s decay: technical skills 12-24 months, compliance one year (it resets), leadership 2-3 years. Be conservative – a benefit window longer than the average tenure of the trained group is fiction.
Can I count improved morale? As narrative, yes; as rupees, only through a measurable proxy (retention, absenteeism). Unmonetized soft benefits belong in the writeup as upside, not in the ROI number.
One employee’s certification – same method? Yes, at Quick-Mode scale: certification cost + their wages for study hours vs the measurable change (billable rate, error rate, retention risk). The calculator handles participants = 1.
Calculators referenced: Training ROI · Training Budget · Learning Hours · Skill Gap Estimator. Benefit monetization is an estimate, not an audit – document your attribution assumptions and keep them conservative.
