How to get a new sales rep to their first real call in 14 days, and what they have to pass first

Most onboarding plans are built in thirty-day blocks, and the blocks describe what the company will show the rep rather than what the rep will be able to do. Week one is access and product. Week two is shadowing. Week three, more product. By the second month the plan has stopped describing anything and turned into a quota.

The trouble with that shape is that it is untestable. A new hire can sit through every block and the plan will never once have asked them to say the company's value out loud. Nobody finds out whether they can until a real buyer is on the line. There is no more expensive place to find out.

The schedule below is the part you can copy. The gate at the end of it is the part that has to be yours, and it is the whole argument of this article.

The Bridge Group's 2026 study of 158 B2B companies puts average ramp to full productivity at 6.2 months, the highest in the study's history, and attributes the rise to deal complexity and larger buying committees. The authors note their sample skews toward companies with engaged revenue leadership. Whatever onboarding is currently doing, it is getting slower.

Fourteen days gets a rep to the first call and nowhere near quota

Fourteen days buys one thing: a rep who can hold a whole conversation with a buyer and reach a commitment, tested before a real buyer is exposed to them. Quota attainment, pipeline and product depth take months. Saying which of the two you mean belongs in the first paragraph of the plan.

Keeping those apart matters more than it sounds. A leader who hears "14-day onboarding" and plans capacity around it will be out by nearly two quarters. A leader who hears "14 days to the first supervised call, six months to full productivity" can plan both, and can tell the new hire the truth on day one.

We do not yet have our own ramp figure at a sample size worth printing. When we do, it will appear here with that sample size attached.

Roleplay beats the product deck and loses to a live call

Roleplay replaces the product deck. It does not replace listening to live calls, and the evidence on that is unusually blunt. Sitzmann's meta-analysis of simulation training, covering 65 samples and 6,476 participants, found simulation beat passive instruction by d = 0.38 and beat no instruction at all by d = 0.61. Against a comparison group that was also actively engaged, simulation came out worse, d = −0.19, with the confidence interval excluding zero.

Shadowing a live call is active. So the popular framing, roleplay in place of shadowing, is the one version of this the research declines to support. Cut the slides, keep the listening, and the claim survives contact with anyone who checks.

Two design rules fall out of the same analysis, and both are cheap to follow:

Of those two rules the paper's own between-group statistic is strongest for keeping simulation as a supplement, and weakest for uncapping attempts, where the cell holds only ten samples. Both are worth following. Only one of them is worth arguing about.

There is a second reason roleplay is rare, and it is not scepticism. In the sixth edition of Salesforce's State of Sales, fielded across 27 countries with 5,500 sales professionals in March and April 2024, roleplay sessions came last of eight enablement practices at 26% adoption, behind strategy reviews at 53% and performance reviews at 51%. Salesforce's own explanation is that personalised enablement of this kind is data-intensive and time-intensive, so teams may find it harder to run consistently or scale as headcount grows. That is a cost problem, and it is the exact cost a simulator removes. We can stand up one buyer profile from calls your team has already lost and let a rep run it.

The paper also states two limits on itself, and they change what you can do with it. Published studies in the set averaged d = 0.52 while unpublished ones averaged −0.10, which is publication bias in plain sight. And only 5% of the samples were employees; 77% were undergraduates, average age 23, with the difference across populations reported as not significant. The operational reading is narrow. Use these numbers to decide what to cut from a week, and never to promise a board a number.

The first week should contain almost nothing the rep only watches

Cut anything the rep consumes passively and nobody checks. Usually that is the product presentation, the recorded webinar library, and the ride-along with no task attached. What goes in their place is smaller and duller: three things the rep has to produce out loud, repeatedly, with somebody looking at the output.

Pull five objections from calls your team lost last quarter, in the buyer's own words, and stop at five. A new hire who can handle the five that actually happen is more useful on day fifteen than one who has read a thirty-item catalogue.

Every day of the plan needs an output and a person who looks at it

A fourteen-day ramp contains five stages, and each one ends in something the rep produced out loud with a named person looking at it. Two days on the product in their own words, three on stage drills with the prompt visible, four with the prompt stripped back, three on whole calls, two on the gate.

Stripping the prompts back is the part people skip. In our own trainer the scaffolding comes off in four steps, full text, then anchors, then goal only, then nothing, and a stage stays open until the rep has cleared it at the hidden level. Practice with the answer on screen tests whether the rep can read. The gate is for something else.

A gate with no written second-failure path is a ceremony

Five things are worth checking before a new rep talks to a buyer: how they open, how deep their questions go, how they handle the objections that actually occur, whether they close on a commitment, and whether they talk too long. Write the pass condition for each one before the rep starts, in terms somebody could disagree with, and write what happens when they fail it twice. Writing the failure column against your own five checks is something we can do with you.

Naming a certification step is the easy half. The failure path is the half that decides whether the gate holds, because that is the afternoon when somebody has to either send a new hire back to drills or wave them through.

Two of those five are worth setting as hard gates, meaning a miss fails the whole run whatever else went well. We default to closing on a commitment and to anything a regulator would want a human to have signed. Everything else can be amber. If every line is a hard gate, nobody passes and the bar quietly gets lowered in private. Setting it lower honestly costs less.

A founder running an immigration services company described their version of this to us on a call. A new hire gets two weeks of material, then sits a roleplay assessment, and access to leads is granted only after both. Fail the assessment twice and the hire ends. Whatever you make of the last part, the structure is this article's thesis told by somebody who built it with no software at all.

Whoever scores the gate should not be what the rep drilled against

The test that decides whether a new rep may talk to a buyer should be scored by a human, or at least by a different instrument. When the same thing both trains and measures, the measurement comes out flattering. We know this from our own build. Twice.

Until August this year, the instructions telling the model who to play were assembled in the browser and sent up with each turn. Any trainee who opened developer tools could read exactly who they were selling to: what moved the buyer, what shut him down, what counted as a win. The comment we left in the code when we fixed it is the shortest statement of the problem I can offer. The simulator stops training after that, because the student knows the answers to the exam they are sitting.

Moving the assembly to the server was only half the fix. The server now discards any instruction the browser sends, because the client could otherwise push its own version and swap the role underneath. And a missing scenario now fails loudly, because a model with no instruction plays anyone, and the roleplay quietly turns into a chat.

The second one is smaller and worse. The language of the session arrived from the client and went into the evaluator's prompt, so a trainee could put an instruction in that field and influence their own score. The part that took us a second pass was where to check it: the validation has to happen before the row is written, not before the model is called, or the raw string sits in the column and resurfaces the next time the attempt is re-scored.

Both bugs are the same bug wearing different clothes, and nothing about it is specific to our stack. Any tool that assembles the buyer's instructions in the browser has the first one. Any tool that pipes a user-supplied field into the evaluator's prompt has the second. The fix is a procedure: the final gate runs against something the rep has not practised against, and a person signs it off.

We built the first version of the roleplay trainer this article describes for ourselves. We were hiring a sales team, nobody had the time to teach them one at a time, and the alternative was letting people learn on live buyers.

What to do next

On what day did your last three hires first speak to a real buyer, and what did anyone check before that happened? If the second half has no answer, the gate is the thing to build first, ahead of any content.

Then write five rows. The check, the pass condition, how it is checked, what happens on a second failure. Five rows on one page is a complete onboarding gate. The fourteen-day schedule is the easy part and the table above will do. The five rows encode what your buyers actually do, so they have to be yours. After day fifteen this stops being onboarding and becomes the weekly coaching loop, which runs on the rep's own recordings.

The gate is the part worth stealing from this article. If you would rather see one working before you write your own, we will build a buyer profile out of calls your team has already had and run a rep through it while you watch. Forty-five minutes, nothing to prepare.