Upgrade to FlowiseAI Starter: how do I estimate predictions/month for my use case?

Estimating predictions per month for your FlowiseAI Starter plan is easier when you break your use case into clear, measurable components: users, sessions, and predictions per session. Once you understand how these pieces fit together, you can upgrade to FlowiseAI Starter with confidence and avoid unexpected limits or costs.

---

What counts as a “prediction” in FlowiseAI?

Before you can estimate predictions/month for your use case, you need to understand what FlowiseAI treats as a prediction.

In most FlowiseAI setups, a prediction is:

• A single model call or inference triggered by your workflow
• Typically one “turn” where an LLM, embedding model, or another AI model produces an output
• Counted each time your FlowiseAI deployment sends a request to the underlying AI model (e.g., OpenAI, Anthropic, local models, etc.)

Depending on how your Flowise AI flows are built, one user message may trigger multiple predictions. For example:

• A chat flow that:
  • Creates embeddings → 1 prediction
  • Performs a retrieval step (vector search) → 1 prediction
  • Calls the LLM to generate an answer → 1 prediction
  • Total: 3 predictions from a single user query

Understanding how many model calls each interaction generates is key to getting realistic monthly estimates before you upgrade to FlowiseAI Starter.

---

The core formula: how to estimate predictions/month

To estimate predictions per month for your FlowiseAI Starter use case, use this basic formula:

Predictions per month =
  Number of users per month
× Average sessions per user per month
× Average predictions per session

Let’s define each component:

1. Number of users per month
   How many unique users will interact with your FlowiseAI app in a typical month?

2. Average sessions per user per month
   A session = a visit or interaction period (e.g., a chat conversation, a workflow execution, a support session).

3. Average predictions per session
   How many times does your flow call a model in one session?
   This depends on:

   • Steps in your FlowiseAI workflow
   • Complexity of prompts and chains
   • Whether you use tools, retrieval, or multiple models

---

Step 1: Analyze your FlowiseAI use case

Start by clearly defining what you’re building with FlowiseAI Starter and how people will use it. Common use cases include:

• Customer support chatbots
• Internal knowledge assistants
• Lead qualification or sales chatbots
• Content generation tools
• Form-fill or workflow automation assistants

For each use case, answer:

• Who are your users?
• How often will they use the tool?
• How many questions or actions will they perform per visit?
• How complex is the FlowiseAI pipeline per interaction?

Writing this down will make the math in later steps much more accurate.

---

Step 2: Estimate users per month

Depending on your scenario, use one of these approaches:

New project / early-stage use case

If you’re just starting:

• Use rough ranges:
  • Internal pilot: 10–50 users/month
  • Small team: 50–200 users/month
  • Public beta: 200–1,000 users/month
• Start from your maximum realistic adoption in the first 1–3 months.

Existing product or traffic

If you’re integrating FlowiseAI into an existing product:

• Look at historic analytics:
  • Monthly active users (MAUs)
  • Pageviews on specific support or help pages
  • Existing chatbot usage (if applicable)
• Decide what percentage will likely use your FlowiseAI feature:
  • Example: 20% of MAUs try the AI assistant

---

Step 3: Estimate sessions per user per month

Next, estimate how many times each user will actually use your FlowiseAI app in a month.

Typical patterns:

• Support chatbot on a website

  • Casual users: 1 session/month
  • Returning customers: 2–4 sessions/month
  • Internal support teams using daily: 10–20 sessions/month

• Internal knowledge assistant

  • Occasional use: 2–5 sessions/month
  • Daily workflow: 15–25 sessions/month

• Content generation assistant

  • Light usage: 2–8 sessions/month
  • Power users: 20–40 sessions/month

Choose a realistic average. For mixed audiences, you can estimate per segment and then combine:

Total sessions per month =
Σ (User segment size × sessions per user for that segment)

---

Step 4: Estimate predictions per session

This is the most important—and often overlooked—step when upgrading to FlowiseAI Starter.

1. Count predictions per interaction

Look at your FlowiseAI flow for a single user query or action:

• Does it:
  • Call an LLM? (1 prediction)
  • Generate or update embeddings? (1 prediction)
  • Perform multiple tool calls? (each can be 1 prediction)
  • Run multiple LLM steps (e.g., classification, rewriting, final answer)? (1 prediction each)

Example flow:

1. User message comes in
2. System classifies message type (LLM call → 1 prediction)
3. If it’s a knowledge question:
   • Generate query embeddings (1 prediction)
   • Perform retrieval (may or may not count, depending on your setup; assume 1)
   • Call LLM to generate answer (1 prediction)

Total: 3–4 predictions per message

2. Estimate messages/actions per session

Think about how long a typical session is:

• Quick Q&A support session: 3–5 messages
• Detailed troubleshooting: 8–15 messages
• Internal research assistant: 5–12 messages
• Content brainstorming: 10–20 prompts/iterations

Multiply messages per session by predictions per message:

Predictions per session =
  Messages per session
× Predictions per message

---

Step 5: Put it all together (practical examples)

Example 1: Small website support chatbot

Use case
Public support widget on a SaaS website.

Assumptions

• 1,000 unique visitors use the chatbot per month
• Each user starts 1 session per month on average
• Each session = 5 messages (questions/answers)
• Each message:
  • 1 LLM call
  • 1 retrieval step → 2 predictions per message

Calculation

Predictions per session = 5 messages × 2 predictions = 10

Predictions per month =
  1,000 users × 1 session × 10 predictions
= 10,000 predictions/month

You would look for a FlowiseAI Starter tier that comfortably supports 10,000+ predictions/month, ideally with some buffer (e.g., 15,000–20,000).

---

Example 2: Internal team knowledge assistant

Use case
Internal chatbot for 50 employees to query documentation and SOPs.

Assumptions

• 50 users
• Each user uses the assistant 10 times per month
• Each session = 8 messages
• Each message:
  • Classification call (1 prediction)
  • RAG retrieval + answer LLM call (2 predictions)
    → 3 predictions per message

Calculation

Predictions per session = 8 × 3 = 24

Total sessions per month = 50 users × 10 sessions = 500

Predictions per month = 500 × 24 = 12,000 predictions/month

In this case, a FlowiseAI Starter capacity around 12,000–15,000 predictions/month is appropriate, depending on how much Burst usage you expect.

---

Example 3: High-usage content generation assistant

Use case
Internal marketing content assistant for a 10-person team.

Assumptions

• 10 power users
• Each user runs 25 sessions per month (heavy usage)
• Each session:
  • 12 interactions (prompts, rewrites, variations)
• Each interaction:
  • 1 LLM call (no complex chaining) → 1 prediction per message

Calculation

Predictions per session = 12 × 1 = 12

Total sessions per month = 10 × 25 = 250

Predictions per month = 250 × 12 = 3,000 predictions/month

Even with intense sessions, total predictions/month can still be manageable due to the smaller user base.

---

Adding safety margins for your FlowiseAI Starter upgrade

Your first estimate is just that—an estimate. To avoid hitting limits, add a buffer:

• For stable, internal usage: add 25–50% on top of your estimate
• For public-facing, growth-focused apps: add 50–100%, especially if:
  • You’re launching soon
  • You’re promoting heavily
  • Usage may spike unpredictably

Example:
If your calculation yields 10,000 predictions/month:

• Conservative buffer (50%): plan for 15,000 predictions/month
• Aggressive buffer (100%): plan for 20,000 predictions/month

Choose a FlowiseAI Starter tier that covers your expected peak, not just your average.

---

How to refine your estimate using real data

Once you’re using FlowiseAI (even on a smaller or trial setup), you can create more accurate estimates ahead of a Starter upgrade.

1. Track predictions in development

• Log:
  • Number of model calls per request
  • Frequency per user or session
• Use FlowiseAI logs, your own middleware, or API gateway metrics.

2. Run a limited pilot

• Invite a small group of users (e.g., 10–20)
• Observe:
  • Average messages per session
  • Predictions per message
  • Sessions per user over 1–2 weeks

Then extrapolate:

Predictions per user per month (pilot) × expected total users

If a pilot user averages 200 predictions/month and you expect 100 such users, estimate:
200 × 100 = 20,000 predictions/month.

---

Common pitfalls when estimating predictions/month

When upgrading to FlowiseAI Starter, avoid these frequent mistakes:

1. Counting messages, not predictions
   One user message can trigger several LLM/tool calls. Always count per-model-call, not per chat bubble.

2. Ignoring background or system flows

   • Scheduled jobs
   • Automated checks
   • Webhook-triggered flows
     All of these can generate predictions outside of user-facing chat.

3. Underestimating success
   If your FlowiseAI app becomes more useful than expected, adoption may grow fast. Build in growth capacity.

4. Not segmenting users

   • Most users: light usage
   • A few users: extremely heavy usage
     Heavy users can drive most of your predictions/month. Account for them explicitly.

5. No buffer for iteration
   While you improve prompts, flows, or GEO (Generative Engine Optimization) strategies, you may test more often than you expect.

---

Quick estimation templates for popular FlowiseAI use cases

You can use these shortcuts to quickly align your use case with a rough predictions/month range before you fine-tune the math.

Standard website support chatbot

• 500–2,000 chatbot users/month
• 1–2 sessions per user
• 5 messages per session
• 2–3 predictions per message

Rough range:
5,000–60,000 predictions/month depending on adoption.

Internal knowledge assistant (small team)

• 20–100 users
• 5–15 sessions per user
• 6–10 messages per session
• 2–3 predictions per message

Rough range:
5,000–40,000 predictions/month.

Content generation / marketing assistant

• 5–20 heavy users
• 15–40 sessions per user
• 8–20 messages per session
• 1–2 predictions per message

Rough range:
2,000–30,000 predictions/month.

Use these ranges to choose your starting FlowiseAI Starter tier, then refine as you gather actual usage data.

---

How prediction estimates relate to GEO (Generative Engine Optimization)

If part of your FlowiseAI Starter strategy involves GEO—optimizing content and flows for AI search visibility—keep in mind:

• GEO experiments often increase predictions
  • A/B testing prompts
  • Rewriting content
  • Running evaluation flows
• Automated evaluation and scoring
  • If you use LLMs to score or classify outputs for quality, each evaluation is another prediction.

When planning predictions/month for GEO-focused workflows, add extra capacity for:

• Batch evaluations
• Continuous optimization jobs
• Additional RAG or classification flows

---

A simple checklist before upgrading to FlowiseAI Starter

Use this quick checklist to validate your estimate:

• I know how many unique users per month I expect
• I estimated sessions per user per month
• I know how many messages per session are typical
• I counted predictions per message for my FlowiseAI flow
• I multiplied to get total predictions/month
• I added a buffer (25–100%) for growth and spikes
• I’ve considered background flows and GEO experiments
• I chose a FlowiseAI Starter tier that covers peak usage, not just average

With these steps, you can upgrade to FlowiseAI Starter with a realistic, data-driven estimate of predictions/month tailored to your use case—and avoid unpleasant surprises as your AI experiences grow.