Behavior-Over-Time Graphs: The One-Minute Sketch That Reveals a System's Hidden Dynamics

A behavior-over-time graph (BOTG) is a rough, hand-drawn line chart of how one key variable has moved across a meaningful stretch of history. Sketching it before drawing a causal loop diagram matters: exponential growth implies a reinforcing loop, leveling off implies a balancing loop, and oscillation implies a balancing loop wrestling with a delay.

Sketch the shape of your key metric's trajectory before naming a single loop. Exponential growth points to a reinforcing loop, a flattening curve points to a balancing loop, and oscillation points to a balancing loop fighting a delay.

What a Behavior-Over-Time Graph Actually Is

A BOTG is a diagnostic sketch, not a dashboard chart: one variable, plotted against time, drawn quickly enough that you capture pattern instead of precision. The y-axis rarely needs exact units. What matters is whether the line rises, falls, bends, or repeats — because that shape is the first clue to the system's underlying loop structure.

The term comes out of System Dynamics, the field Jay Forrester founded at MIT in the late 1950s to model industrial and organizational behavior mathematically. Donella Meadows later popularized a related idea in Thinking in Systems: A Primer: the reference mode — the historical (and expected) pattern of behavior that any credible model of a system has to reproduce. John Sterman's Business Dynamics, the standard MIT Sloan textbook on the field, treats sketching the reference mode as the very first deliverable of a systems engagement, before a single stock or flow gets defined.

Reference Mode vs. Point Metrics

Most PM dashboards show point metrics: this week's activation rate, this month's churn. A BOTG asks a different question — not "what is the number," but "what is the number doing." That distinction is the foundation of systems thinking generally; if you haven't yet, our complete guide to systems thinking for product managers covers the broader vocabulary of stocks, flows, and loops that BOTGs feed into.

Why a Sketch Beats a Precise Chart

It's tempting to open your analytics tool and export a clean line chart instead of drawing one by hand. Resist that instinct for the first pass. A hand-drawn BOTG forces you to commit to a shape from memory — exponential, S-curve, oscillating — before the tool's smoothing, log scales, or truncated axes bias your read. Precision tools are excellent for confirming a hypothesis; they're poor for forming one, because they answer the question you already asked instead of surfacing the pattern you hadn't noticed. Sketch first, verify with real data second.

The Canonical Curve Shapes and the Loop Structures Behind Them

Six curve shapes recur constantly in product metrics, and each is a fingerprint for a specific loop structure. Exponential growth means a reinforcing loop dominates; a curve flattening toward a ceiling means a balancing loop dominates; oscillation means a balancing loop is fighting a delay.

The first three rows describe growth that hasn't met resistance yet, or growth that has met its match cleanly. The last three describe systems where a delay causes the correction to overshoot the target — the more subtle, and more common, pattern in mature products. For a deeper walkthrough of why reinforcing loops compound and balancing loops correct, see our piece on reinforcing versus balancing loops in growth systems.

Distinguishing a Ceiling From a Collapse

A goal-seeking curve and an overshoot-and-collapse curve often look identical for their first two-thirds — both rise, then bend. The difference shows up at the top: a goal-seeking curve bends before it hits the limit, because the balancing loop has time to act. An overshoot-and-collapse curve keeps rising past the limit and only bends once the constraint bites hard, usually because of a delay in noticing it. If you can't yet tell which one you're in, wait for one more data point before committing to a diagnosis.

Delays are the hidden variable in four of the six shapes above. A balancing loop with a short delay corrects smoothly; the same loop with a long delay overshoots, corrects, and overshoots again — visible as oscillation or overshoot-and-collapse. We go deep on this mechanism, especially as it plays out in retention curves, in how delays in feedback loops shape retention and churn.

How to Sketch a BOTG in Under a Minute

Sketching a BOTG takes four steps: pick one variable, choose a timeframe long enough to show a full cycle, draw the line from memory or a quick data pull, and add a second line for what you expected or wanted to happen. The comparison between the two lines is where the insight lives, not the values on either axis.

1. Pick the one variable that matters. Weekly active users, support ticket backlog, NPS, whatever the conversation is actually about — not everything you could measure.
2. Choose a timeframe with at least one full cycle. A week of data can't show a quarterly pattern; go back far enough to see the shape repeat or complete.
3. Draw the actual line freehand. From memory, a spreadsheet glance, or a quick query — precision isn't the point, direction and inflection are.
4. Add a dotted reference line. What you expected to happen if nothing changed, and optionally a third line for what you want to happen.
5. Layer in one or two related variables. The interaction between two lines — say, signups and support load — is usually more diagnostic than either line alone.

Where the Historical Line Comes From

If you don't remember the shape, thirty seconds in your analytics tool usually gets you close enough — a single line chart of the past two to four quarters, default axis, no filters. For qualitative variables, a scroll through old retro notes, support tickets, or Slack threads works just as well; you're reconstructing a shape, not auditing a precise number.

The Three-Line Technique

Donella Meadows taught a version of the BOTG with three lines on one sketch: what actually happened, what you'd expect if current trends simply continued, and what you actually want to happen. The gap between line one and line two tells you a loop has already bent the trend. The gap between line one and line three tells you how much work is left. Most PM conversations jump straight to line three — the target — without ever sketching line two, which is exactly how a healthy S-curve gets mistaken for a failing one.

Multiple Variables on One Chart

A single-variable BOTG tells you the shape of one story. A two- or three-variable BOTG tells you which story is driving which. If weekly signups climb exponentially while onboarding completion flattens, you're looking at a reinforcing loop (referrals) running into a balancing loop (support or onboarding capacity) — and the gap between the two lines is your leverage-point candidate. Our guide to finding leverage points in a product system picks up exactly at this juncture.

Turning a Shape Into a Dynamic Hypothesis

A curve shape becomes useful the moment you translate it into a plain-English hypothesis about loops and delays — what Sterman calls the dynamic hypothesis. Name the reinforcing loop that could produce the rise, the balancing loop that could produce the flattening, and the delay that could produce the oscillation, then test the story against what you already know.

Consider a B2B SaaS product's weekly active users (WAU). For the first two quarters after launch, WAU grows exponentially — each new user refers a few more, so the natural read is a reinforcing loop: referrals feed awareness, awareness feeds signups, signups feed more referrals.

By month seven, the curve bends into an S-shape. Growth hasn't reversed, but it has slowed sharply, which suggests a balancing loop has entered the picture — most likely a finite addressable market, or an onboarding team that can only process so many new accounts per week without quality dropping. The shape alone tells you a constraint exists; it doesn't yet tell you which one.

Six months later, WAU starts to wobble — up after each feature release, down a few weeks later, repeating. That's the signature of a balancing loop working against a delay: a release drives usage up, but the lag between "feature ships" and "support team fully ramps to handle it" means service quality trails the growth spike, and users churn a few weeks after the fact. The gap between cause and correction is what turns a smooth correction into an oscillation.

If the variable you're tracking is qualitative — sentiment, frustration, delight — a BOTG still works; you just plot felt intensity instead of a count. That's effectively the same primitive behind an emotion curve in a customer journey map: time on the x-axis, experience on the y-axis, and the shape revealing exactly where the friction sits.

None of this requires a data science team. It requires drawing three shapes on a whiteboard and asking, out loud, what kind of loop could have produced each one — the same discipline whether the variable is WAU, churn, or deployment frequency.

Common Mistakes When Reading a BOTG

Most misreads of a BOTG come from zooming in too close, trusting a single data point, or ignoring how long effects take to show up. Each mistake below produces a confidently wrong loop diagram — the fix is almost always to widen the window, add a second variable, or ask how much delay separates cause from effect.

That third mistake — ignoring the delay — is worth dwelling on. In a well-known systems-thinking inventory, MIT researchers Linda Booth Sweeney and John Sterman found that a majority of well-educated participants, including graduate students, misjudged the basic stock-and-flow dynamics of something as simple as a bathtub filling and draining. If trained analysts get delay-driven dynamics wrong on a bathtub, it's safe to assume a team will get them wrong on a product metric too — which is exactly why the BOTG's second line, the reference or expected behavior, matters so much.

Smoothing Without Lying to Yourself

A little smoothing helps a BOTG; too much hides the very oscillation you're trying to diagnose. A reasonable rule: smooth over a window shorter than the suspected cycle length, never longer. If you suspect a monthly oscillation, a 7-day rolling average is fine; a 90-day rolling average will flatten it into what looks like steady, boring growth — and you'll draw the wrong loop entirely.

When the Shape Says "Wrong Job," Not "Wrong Loop"

Not every flat or declining curve is a loop problem. Sometimes a BOTG plateaus because the product serves a shrinking population well, or serves the wrong need for the population it has. Before committing to a loop-based fix, sanity-check the assumption against the underlying need — our Jobs to Be Done complete guide walks through how to separate a structural problem from a positioning one.

From Sketch to Model: Encoding the Dynamic Hypothesis

A behavior-over-time sketch is only useful once it becomes a testable hypothesis you can share, critique, and refine with a team — a napkin drawing doesn't survive a Slack thread. The natural next step is turning each named loop into an explicit causal loop diagram, with polarity and delay marked, so a teammate can challenge the structure rather than the vibe.

This is exactly where a behavior-over-time sketch earns its keep: it is the dynamic hypothesis, in its roughest form, before you've committed to any loop. In Prodinja's Systems Engineering editor, you encode the loops you hypothesized from the shape — naming reinforcing and balancing relationships, marking delays — and the editor's causal-loop detection surfaces the structure back to you, so you can check the diagram against the sketch that started it, rather than the other way around.

Key Takeaways

• A behavior-over-time graph is a fast, rough sketch of one variable's trajectory — the diagnostic step before any causal loop diagram.
• Exponential growth implies a reinforcing loop dominates; a curve that flattens toward a ceiling implies a balancing loop dominates.
• Oscillation and overshoot-and-collapse both signal a balancing loop fighting a meaningful delay between action and effect.
• Plot two or three related variables together, not one — the interaction between lines is usually more diagnostic than any single line.
• Widen your timeframe before trusting a shape; a single data point or a short window can't reveal a pattern.
• A BOTG becomes valuable the moment it turns into a shareable, testable dynamic hypothesis about loops and delays.