Lessons from a dozen AI workshops: what non-technical teams actually retain

Run enough AI workshops for non-technical teams and a pattern separates itself: the room learns two very different kinds of things, and only one kind survives the week. One kind is judgment — a working sense of when to reach for AI, how to ask for what you want in your own words, and how to catch it when the answer is wrong. The other kind is recipe — a memorized prompt string, a sequence of menu clicks, a template someone copied off the projector. Judgment carries to Monday. Recipes are mostly gone by Friday. The forgetting curve is not gentle: people lose roughly half of newly learned material within an hour and up to 70% within a day when nothing reinforces it. A brokerage that measures a workshop’s return in prompt templates has bought the part that decays fastest. Across CRE teams — brokers, property managers, acquisitions analysts — the retained lessons are consistent enough to name, and so are the things that vanish.

Two pointers before the lessons. The full arc of getting a lean team fluent — the ninety-day sequence, not the single afternoon — sits in our CRE AI training playbook, and the case for why small shops can out-operate larger competitors is in the small CRE firm manifesto. This piece is narrower: drawing on the pattern that emerges across a dozen AI workshops, it reports what non-technical CRE staff carried out the door, and what they left on the floor.

The One Distinction That Predicts Retention

If you remember one thing, remember the split between judgment and recipe. It predicts retention better than the trainer, the platform, or the length of the session.

Judgment is transferable. A property manager who understands that AI is good at rewriting and summarizing but bad at inventing facts applies that instinct to a tenant email, a lease clause, and a vendor complaint she has never seen. She is not recalling a specific answer; she is recalling a rule about where the tool helps and where it lies. That rule survives because it is short, it is hers, and she uses it every day.

Recipe is brittle. A twelve-line prompt template written by the trainer works in the room and fails the following week, because the person never internalized why it was structured that way. When the situation shifts — a different property type, a different tone, a different ask — they cannot adapt it, so they abandon it. What looked like the most valuable takeaway is the first thing to go.

This is why the NAR 2025 Technology Survey shows 68% of Realtors using AI while only 17% report a significant positive impact. The gap is not access. Most training loads people up with recipes and sends them home, and recipes decay. The workshops that move the impact number teach judgment and treat prompts as disposable examples, not as the product.

What Actually Survives the Week

The retained lessons are remarkably consistent across rooms, and share a trait: each is a small mental model the person can restate in their own words and use without looking anything up.

• “AI drafts, I decide.” The stickiest idea. People keep the frame that the tool produces a first pass and they own the final version. It survives because it maps onto how brokers and managers already work — reviewing junior work — so it needs no new habit, just a new junior.
• “Ask like you’d ask a sharp new hire.” Once someone stops trying to write “prompts” and starts describing the task the way they would to a capable assistant — context, an example, the format they want back — output quality jumps and the skill sticks, because there is no syntax to forget.
• “Never trust a number or a name it gives you.” The verification instinct. People retain that AI will state a cap rate, a square footage, or a case citation with total confidence and be wrong. Burned once in the room by a confident fabrication, they keep the reflex to check anything factual.
• “Give it the source, don’t make it remember.” Teams retain that pasting the actual lease, rent roll, or email thread produces grounded output, while asking from memory produces plausible fiction. This survives because the failure mode is so vivid when shown live.
• “Start it, then talk to it.” Treating the tool as a conversation — pushing back, saying “shorter” or “more formal” — rather than expecting a perfect one-shot answer. People keep this because it feels like delegation, which is familiar.

None of these are tool-specific. They work in ChatGPT, Claude, Gemini, or Microsoft Copilot without modification, which is why they endure even when the interface changes.

What Reliably Evaporates

The failure catalog is just as consistent — name it plainly so you do not pay for it.

• Exact prompt strings. Anything longer than a sentence that had to be memorized is gone within days. If the workshop’s value lived in a handout of prompts, the value left with the handout.
• Tool-specific button paths. “Click the paperclip, then the model selector, then attach” — interface mechanics evaporate the moment the interface updates, which for these products is often.
• Features tied to a login the person doesn’t own. If the demo used a paid seat, team workspace, or connected data source the attendee cannot access on Monday, the capability was never transferred. People retain what they can do, not what they watched.
• Anything the person never touched. Watching the trainer retains almost nothing. The skills that survived were the ones each attendee performed with their own hands on their own real work. Passive demos are the most expensive form of forgetting.
• Enthusiasm without a first task. The room ends energized, and the energy dissipates by Wednesday if no one leaves with one specific job to do with the tool that week. Motivation is not retention.

The through-line: the more a lesson depended on memorization, a specific screen, or someone else’s hands, the faster it died. It is also why “just watch some tutorials” fails — passive video is the purest form of the demo-without-doing problem. The broader reasons these programs miss are worth reading before you book: see why most AI training programs fail at real estate firms.

Why Role Changes What Sticks

Retention is not uniform across a CRE firm, because the tasks are not. What a team keeps depends on what they do all day, and a workshop that ignores role teaches half the room things they will never use.

• Leasing and sales brokers retain AI for the writing they already resent: first-draft LOIs, listing descriptions, comps write-ups, follow-up emails. The judgment that sticks is “draft fast, then make it sound like me.” They drop data-pipeline and spreadsheet framing — not their job, not retained.
• Property managers keep AI for volume triage: summarizing long tenant threads, drafting maintenance and notice communications, pulling key terms out of a lease so they do not re-read forty pages. What sticks is “summarize and extract from the document in front of me.”
• Acquisitions and investment staff retain AI for reading fast: digesting an offering memorandum, sanity-checking assumptions, restructuring notes into a memo. The lesson that sticks is “read faster, but verify every number” — and they keep the verification reflex hardest, because their numbers carry the most risk.

The case against generic training is this: a mixed room retains only the lessons that touch each person’s actual work, so a session pitched at “AI for real estate” in the abstract retains poorly for everyone. Domain-specific practice on the team’s own documents is what converts a demonstration into a habit — the mechanics are in the anatomy of an effective CRE workshop.

The Cognitive Science, Briefly

The field observations line up with what learning research has said for a century, which means the fixes are known, not improvised.

The forgetting curve, first measured by Hermann Ebbinghaus, describes how newly learned information decays rapidly without reinforcement: a steep drop in the first day, then a longer tail. A single workshop sits at the top of that curve and falls. Nothing about AI changes it — the content is new, the memory science is old.

Two mechanisms flatten it, both well supported in the literature. Retrieval practice — being made to recall and produce, rather than watch — builds durable memory far better than passive review; that is why the hands-on lessons survived and the demos did not. Spaced repetition — revisiting the skill at intervals over the following weeks — beats a single massed session even at equal total time. A ninety-minute workshop plus two short check-ins weeks apart retains far better than one two-hour block.

The translation is blunt: retention is a design choice, not a talent. A workshop that makes people do the work and schedules light reinforcement will stick; one that shows slides and hands out prompts will not. Sequencing that reinforcement into a rollout is the subject of our adoption framework for small CRE firms, which treats practice on real work as its own stage.

What to Demand So a Workshop Retains

Turn the lessons into buying criteria. Market rates for CRE team workshops run roughly two to fifteen thousand dollars depending on size and depth; six things separate a session that changes behavior from a pleasant afternoon that does not.

1. Hands on keyboards, on your own work. Every attendee spends most of the session doing their own real tasks, not watching. If the agenda is mostly presentation, retention will be mostly zero.
2. Judgment over recipes. “When to use it, how to ask, how to verify” is the durable curriculum; a binder of prompts is the perishable one.
3. Role-relevant tasks. Brokers practice LOIs and comps; managers practice tenant comms and lease extraction; acquisitions practice memo drafting and verification.
4. Tools they can use Monday. The session runs on accounts and access your team already has, so nothing demonstrated is stranded behind a login they lack.
5. A verification habit, taught explicitly. Given how confidently these tools fabricate, the workshop must burn in the check-everything-factual reflex — most of all for confidential deal data.
6. Reinforcement built in. One session is a spike, not a change. Ask what happens in the two to four weeks after; even light follow-ups convert it into a habit.

Demand these six and you get durable behavior change. Book a one-off show and you get the 68/17 gap. The difference is not the trainer’s talent; it is whether the design respects how memory works.

Where to Start

Which lessons stick at your firm depends on your team’s actual workflows — and mapping those is a short, concrete exercise. A free AI-readiness assessment is a working session that identifies which roles have the most to gain, which tasks a trained team would genuinely change, what confidential-data rules you need in place first, and whether you need a full hands-on workshop, a lighter touch, or just clearer rules for the tools people already use. Book a free AI-readiness assessment and you will leave knowing what durable training would look like at your firm — and what to refuse to pay for.

Frequently Asked Questions

What do non-technical teams actually retain from an AI workshop?

They retain judgment, not recipes. The durable takeaways are small mental models — “AI drafts, I decide,” “ask like you’d brief a sharp new hire,” “never trust a number it gives you without checking,” and “give it the source document rather than making it remember.” Each survives because it is short, restatable in the person’s own words, and used daily. Memorized prompt strings and tool-specific clicks fade within days.

Why does so little of a typical AI training stick?

Because most training teaches the perishable part. A polished prompt template feels valuable, but it decays fastest — the person never internalized why it was built that way and cannot adapt it when the situation shifts. Sessions built on presentation and handouts sit at the top of the forgetting curve; sessions built on hands-on practice retain, because the skill was retrieved under the learner’s own effort.

How long before workshop skills fade without reinforcement?

Fast. Research on the forgetting curve shows people lose roughly half of newly learned material within an hour and up to about 70% within a day when nothing reinforces it, with the majority gone within a month. That decay is why a single workshop produces a spike rather than a change. Spaced follow-ups over the following weeks flatten the curve and convert the spike into a habit.

How many people from a small firm should attend?

Train the people who do the repetitive writing and reading — usually most of a 4–20 person office. Retention depends on each person practicing their own real work, so passive observers gain little. If budget forces a subset, pick one person per role who will use it daily and can model it for peers, then plan a second round.

How do we measure whether a workshop worked?

Measure behavior on real work two to four weeks out, not attendance. Useful signals: how many people used the tool on a real task last week, whether a recurring workflow like LOIs or tenant summaries now runs faster, and whether people can describe when they reach for AI and how they verify it. Login counts and satisfaction scores are vanity metrics that do not predict retention.

Does role-specific training retain better than generic training?

Yes, noticeably. Retention tracks whether the practiced task matches the person’s actual job. Brokers keep AI for LOIs, listing copy, and comps; managers for tenant-thread summaries and lease extraction; acquisitions staff for reading offering memoranda and verifying numbers. A generic “AI for real estate” session retains poorly because half the room never practices what they will use.

What does an effective AI workshop cost?

Market rates for CRE team workshops run roughly two to fifteen thousand dollars, depending on team size, length, and how customized the exercises are to your workflows. The cheaper end tends to be generic and presentation-heavy — the format that retains poorly. You are not paying for slides; you are paying for hands-on practice on your own work plus reinforcement afterward. A session with no follow-up is overpriced at any number.

What should happen in the weeks after the workshop?

Light, spaced reinforcement. Revisiting a skill at intervals beats a single long session even at equal total time. Practically: one or two short follow-ups in the two to four weeks after, each tied to a real task, plus one named workflow every attendee is expected to run with AI that first week. That is the difference between a spike of enthusiasm and a durable change.