"We Need a Course on X": The Most Expensive Sentence in L&D
On Monday morning an email lands from the VP of Sales. Representatives aren't logging their customer calls, forecasts are unreliable, and the message ends with a sentence every learning team has read a hundred times: "We need a course on the CRM."
There are two ways to respond. The first is to say yes. You schedule a kickoff, collect the vendor's slides, and six weeks later 600 salespeople click through a polished module and pass a quiz. Completion hits 94%. Three months on, call logging has barely moved.
The second is to treat the sentence as what it really is: a proposed solution disguised as a problem. Someone has diagnosed the cause as "people don't know how" and prescribed training before anyone checked. Sometimes that diagnosis is right. Often it isn't, and being wrong costs far more than it appears.
This matters more now than it did a few years ago. AI tools can draft an outline, a script and a quiz in an afternoon, which makes saying yes cheaper than it has ever been. Cheap yeses are dangerous. If the diagnosis is wrong, you simply deliver the wrong answer faster and at greater scale. When production is nearly free, the scarce professional skill is the thoughtful "not yet".
Why the sentence is so expensive
Start with the visible costs. Suppose the course takes six weeks to design and two hours of every learner's time. For 600 reps, that is 1,200 hours of selling time redirected to a screen, before you count the build effort, the platform, the reminder emails and the managers chasing completions. These are illustrative numbers, but most organisations are surprised when they run the arithmetic.
The hidden costs are worse:
The real problem survives. If reps skip logging because the screen takes nine clicks and their bonus depends only on calls made, no training touches either cause. Yet everyone now believes the issue has been handled.
Credibility erodes. When the course fails, the story that forms is "training doesn't work", not "we trained for the wrong problem". The next legitimate request meets more scepticism and less budget.
Attention is spent. Learners forced through content they didn't need learn to treat L&D messages as noise.
Success is mismeasured. A 94% completion rate looks like victory, so nobody investigates why behavior hasn't changed.
None of this means stakeholders are being foolish. Training is the lever they know L&D controls, so when something goes wrong they reach for it, the way anyone reaches for the tool they have been handed. Our job is not to scold them. It is to widen the toolbox.
Turn the solution back into a problem
The first move is to translate the request into a performance gap. A usable gap statement has four parts: who should be doing what, what they do instead, how often, and what it costs.
"We need a course on the CRM" becomes something like this: "Sales reps should log every customer call within a day. Today about half of calls go unlogged, mostly for mid-sized accounts. Forecast accuracy suffers, and managers spend Fridays reconstructing pipelines by phone." (This is a composite example, not a real client.)
Notice what changed. The request now has a measurable behavior, a baseline and a business consequence. It no longer contains a solution, so the full range of options reopens. And you can eventually answer the only question that matters: did the number move?
Next comes the single most useful question in performance consulting, a classic from the field's early practitioners: could these people do it today if their job depended on it? That one question splits the problem in two.
Only one of the six is a problem training can solve. Here is what the others look like on the ground:
Unclear expectations. A manager says "keep your notes tidy", but nobody has defined what good looks like. The fix is to write the standard, with examples.
Missing information or feedback. A team lead never learns whether their handover notes were useful, and nobody tells a rep their data was wrong. The fix is timely feedback that closes the loop.
Misaligned incentives. The bonus rewards call volume, so two minutes spent logging costs money. People do what they are rewarded for, whatever the policy says. The fix is to change what is measured and recognised.
Capacity or fit. Impossible workloads, or people placed in roles that don't suit them. The fix lies in staffing, selection and workload.
Skill gap. People genuinely haven't learned the task, or it is so rare that they forget it between uses. This is where training or a job aid belongs.
In practice, real cases mix several causes. A team might have an unclear standard, a clumsy tool and a skills gap all at once. The point of the scan is to see the whole picture before you commit your budget to one slice of it.
A five-day investigation
You don't need a three-month study. A time-boxed week usually reveals enough to make a sound recommendation.
Day one: pin down the gap and the data. Gather whatever already exists: error rates, audit results, support tickets, complaint themes, system logs. Confirm the baseline, and agree with the requester which number should move.
Day two: compare strong and typical performers. Find two or three people who do the task well and two or three who struggle. What do the strong ones do differently? Often it is something neither group would mention in a survey, such as a workaround, a habit or a relationship.
Days three and four: go to the front line. Shadow people at work. Interview them with "tell me about the last time you did this" instead of "why don't you do this?", because the first gets stories and the second gets defensiveness. Ask what gets in the way, what shortcuts exist and why. From these conversations, build a failure-mode list: the specific mistakes, what triggers each one and what it costs.
Day five: sort and recommend. Place your findings against the six causes, judge roughly how much each contributes, and shape options.
Where AI helps, and where it can't
AI is a useful assistant here. It can transcribe interviews, cluster open comments into themes, draft interview guides and play devil's advocate. A prompt like "list ten alternative causes for this problem that training would not fix" is a good way to widen your thinking.
What it cannot do is see your workplace. An AI-generated learner persona is a plausible fiction, an average of everything written about people like your learners. It is not a description of the people on your floor. Use AI to prepare for the interviews and to organise what you hear, but never to replace them. The conclusion about root cause should be yours, tested against what you actually saw.
How to say "not yet"
The investigation is half the job. The other half is a conversation with a stakeholder who asked for a course and is about to hear something else. Three moves help.
First, validate the problem before you challenge the solution. Something like: "You're right that unlogged calls are hurting forecast accuracy. I want to make sure whatever we build actually fixes it." You are now on their side of the table.
Second, ask for a bounded investigation rather than refusing. "Give me a week to talk to a dozen reps and look at the data. If it's a skills issue, we'll build exactly the course you're asking for, and it will be better aimed." Most leaders will trade a week for a better outcome.
Third, return with options, not a verdict. Present a short menu: a process fix, an incentive change, a job aid, a short practice experience, or a combination, each with estimated effort and likely effect on the metric. Let the stakeholder choose with their eyes open.
If they still insist on the course, build the smallest honest version. Write down the assumption ("we are assuming reps lack the skill"), agree a metric and a review date, and measure. If the number doesn't move, you have created the evidence for the conversation that should have happened first, without winning a fight.
When the answer really is yes
Sometimes it is a skill gap, and then you should build with confidence. Training is the right tool when three conditions hold: people genuinely can't do the task today, the stakes justify the investment, and the behavior is learnable.
Even then, the best answer is rarely a content dump. Design practice from the failure modes you found: realistic decisions, visible consequences and useful feedback. For tasks people perform rarely, a job aid such as a checklist at the point of need often beats a course, because nobody should have to remember something they use twice a year.
The question behind the question
The sentence "we need a course on X" will never stop arriving, and it shouldn't. It is a sign that people trust L&D enough to ask for help. But the designers who become indispensable will be those who can hear it and calmly ask: what are we actually trying to change, and how will we know?
Answer that honestly, and sometimes the best thing you deliver is not a course at all. It is a fixed process, a better incentive, a clearer standard or a feedback loop that finally works. That is the work that moves the numbers, and it is the work that no tool can do for you.
Keep the conversation going; join this community👇
Comments