Start with the decision the model has to support. An attribution model is a set of credit rules, and different rule sets are correct for different questions. Pick the model that answers your actual budgeting question, then confirm you have enough conversion volume to run it without fooling yourself.
Step 1: Name the Decision
Three decisions drive most attribution work, and each points to a different model.
- Where to spend next quarter's demand creation budget. First-touch credit is the closest fit, because it isolates which channels open new accounts. - Which conversion assets to invest in. Last-touch or time-decay credit is the closest fit, because it isolates what moves a buyer who is already engaged. - How to split a shared marketing number across the full journey. Position-based or algorithmic credit is the fit, because both distribute credit across the record instead of concentrating it.
If you cannot name the decision, no model will help. You will produce a report that gets debated and never used.
Step 2: Match the Model to Your Data Volume
Treat the bands below as a working guide, not a benchmark. The condition that actually decides the choice is whether you have enough conversions to estimate weights without fitting noise.
| Deals closed per year | Workable model | Why |
|---|---|---|
| Under 50 | Single-touch plus self-reported source | Too few conversions to fit weights |
| 50 to 300 | Position-based or time-decay | Rules are transparent and defensible |
| 300 or more | Algorithmic with 12+ months of history | Enough events to estimate weights |
Step 3: Test the Model Against Closed Revenue
A model earns its place by predicting something. Hold out a quarter, run the model on the prior period, and check whether the channels it favored produced opportunities that converted at a higher win rate and closed at their forecast value. A model that ranks channels but shows no relationship to closed revenue is a reporting exercise.
Step 4: Freeze the Definitions
Write down the attribution window, the definition of a qualifying touch, and the rule for matching contacts to accounts. Publish those three rules with the report. Most attribution arguments are definition arguments in disguise, and definitions that shift by quarter destroy forecast accuracy in any model that consumes the output. Treat attribution rules the way you treat close-date rules and stage-exit criteria, as governed fields with owners. Sales forecasting best practices cover the same discipline on the pipeline side.
Frequently Asked Questions
What is the best attribution model for B2B SaaS?
For most B2B SaaS teams, position-based attribution is the practical default. It credits the touch that created the account and the touch that created the opportunity, which maps to how demand creation and demand capture budgets are actually split.
How much data do you need for algorithmic attribution?
Algorithmic models fit weights from historical conversions, so they need enough closed-won and closed-lost opportunities to separate signal from noise. Teams closing a few dozen deals a year will overfit and should stay with rule-based models.
Should the attribution model be the same for every channel?
Yes. Applying different credit rules per channel makes the totals impossible to reconcile against bookings. Use one model for the shared report and run separate incrementality tests when a specific channel is in question.
How often should you change attribution models?
Change the model when the go-to-market motion changes, not when a quarter looks bad. Switching models mid-year breaks every trend line you have and gives each team a version of history that supports its own budget.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like how do you choose an attribution model? into prescriptive action for your team.
Schedule a Demo