How to Do a Training Needs Analysis and Build an Annual Training Plan

What is a training needs analysis, and why does it matter now?

A training needs analysis (TNA) is a structured way to find the gaps between what people can do and what the organisation needs them to do, and to decide which of those gaps training should close. It is the step that turns a training budget from a list of events into a set of decisions you can explain.

The pressure on skills is well documented. The World Economic Forum's Future of Jobs Report 2025 says workers can expect about 39% of their existing skill sets to be transformed or become outdated between 2025 and 2030, and estimates that 59 in every 100 workers will need training by 2030. In LinkedIn's Workplace Learning Report 2025, 49% of L&D and HR professionals agreed their executives are concerned that employees lack the skills to execute business strategy.

Concern does not automatically produce a good plan. A plan built from requests tells you what people would like to attend. A plan built from analysis tells you what people must be able to do next year for the business to hit its goals, and the cheapest reliable way to get there.

What are the steps in a training needs analysis?

A practical TNA runs in nine steps. Most organisations can complete the first eight in about eight working weeks before the budget deadline, then use the ninth all year.

What are the three levels of training needs analysis?

In 1961 William McGehee and Paul Thayer described needs analysis at three levels, and it is still the clearest way to organise the work. Organisation analysis asks where the business needs capability, based on goals, risks and change. Role or task analysis asks what each role must do well, and to what standard. Individual analysis asks who already can and who has a gap.

Skipping a level causes predictable problems. Without the organisation level you train people well in things the business does not need. Without the role level your individual ratings have no standard to compare against. Without the individual level you send everyone on everything.

For the role and individual levels, keep the skills framework short: five to eight priority skills per role group, each with four behavioural levels such as Aware, Working, Skilled and Expert. Borrow wording from public frameworks such as SFIA for digital roles, the EU's ESCO classification, the US O*NET database or Singapore's Skills Frameworks, then adapt it to your own work.

How do you know if it is really a training problem?

Training fixes one kind of problem: people not knowing how. When the cause is unclear expectations, missing feedback, poor tools, a process that makes the wrong way easier, or incentives that reward the wrong behaviour, a course costs money and changes nothing.

Performance analysis in the tradition of Robert Mager and Peter Pipe, and Thomas Gilbert's work on the environment, give you a simple test. Describe the gap in observable terms and price it. Then check the environment and the person. The single most useful question is whether people could do the task now if they really had to. If the answer is yes, the problem is rarely skill.

• Do people know clearly what good looks like, and get timely feedback?
• Do they have the tools, data, systems and time to do it?
• Does the process make the right way the easy way?
• Is doing it right noticed, and doing it wrong not rewarded?
• Have they done it correctly before, and do they do it often enough to stay in practice?

Which data and surveys should you use?

Start with what the organisation already knows: operational KPIs, quality and error data, customer complaints, safety incidents, audit findings, exit interviews, engagement scores by team, performance review themes and assessment results. Existing data is cheaper and usually more trustworthy than new opinions, and it often shows the cause as well as the size of a gap.

Then fill the gaps with a short survey, a skills audit and focus groups. Keep the employee survey under ten minutes and rate only the priority skills. Ask about obstacles and time as well as skills: in the CIPD's Learning at Work survey 2023, only 39% of L&D professionals believed employees are given time away from their day-to-day role for learning. Avoid questions about learning styles, which are not supported by evidence; ask about devices, languages, shifts and what has helped before.

Use self ratings plus manager ratings, and hold a calibration conversation wherever they differ by two levels or more. Report results only for groups of five or more, and say in writing that ratings are used to plan support, not for pay or performance decisions.

How do you prioritise training needs with the business?

Score every need on the same criteria, with the business owners in the room. A practical set is business impact, number of people affected, urgency, risk or compliance, and strength of evidence, each from 1 to 5, combined with weights you agree in advance. Put mandatory training in first, send needs that are not about skill to their owners, and draw a line.

Ask owners to score independently before the meeting and discuss only where scores differ by two points or more. Needs below the line are parked with a reason and a review date, not lost. When a leader pushes a favourite programme, score it on the same criteria in front of everyone and show where it lands. If you want to test your own judgement in this area, AssessAll's Training Needs Analysis Work Sample for Learning and Development Teams and Human Resource Business Partners is designed for exactly this kind of decision.

How do you choose solutions and cost the annual plan?

Design from what people must do on the job, not from a favourite format. Rare or simple tasks need a job aid in the flow of work; skills that need judgement under pressure need practice with feedback spread over weeks; equipment skills need hands-on practice. Research on retrieval practice and spaced practice supports building recall and spacing into every programme. The 70-20-10 idea is a useful reminder that most development happens through experience and other people, and Jobulary's explainer, Learn by doing: the 70-20-10 model explained, shows how to use it without treating the numbers as a formula.

Then decide whether to build, buy, borrow or blend, comparing total cost over the programme's life and fit to your context. Cost every programme fully: external fees, internal trainer days at a rate agreed with finance, content, venue and travel, and learner time if your organisation counts it. ATD's 2025 State of the Industry report found an average of 13.7 formal learning hours per employee in 2024, down from 17.4 in 2023. Use benchmarks like this for context, not as a target, and hold back a contingency for needs that appear during the year.

How do you get the plan approved?

Share findings before you share the plan, so nobody meets a finding for the first time in the approval meeting. Then write the plan as a decision document: priorities linked to goals, the programme portfolio, mandatory training, a calendar built around the business, the budget, how you will measure, governance and the decisions requested. Add a one-page summary that a leader can read in two minutes.

Walk the finance director and two key leaders through the plan before the meeting. Know in advance what you would defer if asked to cut, and cut whole programmes from the bottom of the ranked list rather than trimming everything. The kit's message library includes the exact words for responding to a cut request.

How do you measure the plan and re-plan during the year?

Agree what you will measure before each programme starts, in proportion to the decision it informs. Kirkpatrick's four levels give a shared language; Jack Phillips's ROI Methodology adds return on investment for large programmes; Robert Brinkerhoff's Success Case Method explains what made the difference; and Will Thalheimer's LTEM sets a higher bar than attendance and opinions.

Track delivery, completion and spend monthly, results quarterly, and hold a quarterly review with the business to start, stop, defer or fund new needs from contingency. Rules also change during the year. In the EU, for example, the AI Act's AI literacy provision was amended by Regulation (EU) 2026/1744, in force from 27 July 2026, so that providers and deployers must take measures to support a sufficient level of AI literacy among staff operating AI systems. Check local rules on mandatory training, levies and data for every country you plan for.