Software to Predict Which Donors Will Lapse: What Actually Works in 2026
Donor lapse prediction software watches giving and engagement signals to flag supporters likely to stop giving before they do. The real options are Dataro, Virtuous Insights, Blackbaud's analytics tools, CRM-native scores like Bloomerang's, and Gratefully, which flags at-risk donors with cited evidence from your own records. None of the vendors publish accuracy figures, so ask hard questions before you buy.
- Why lapse prediction matters right now
- What lapse prediction software actually does
- The signals every model watches
- The tools compared
- Scores vs evidence: two ways to predict a lapse
- What vendors will not tell you
- Seven questions to ask before you buy
- What to do when a donor gets flagged
- When you do not need prediction software
- Frequently asked questions
Why lapse prediction matters right now.
The sector is keeping fewer of the donors it wins. The Fundraising Effectiveness Project's Q4 2025 report puts overall donor retention at 43.3%, barely up from 43.1%, while donor counts fell an estimated 3.6% in 2025, the fifth straight year of decline. That is a full-year figure. The year-to-date figure through Q1 2026 is 18.0%, measured under a methodology FEP revised in the same report. Full-year and year-to-date numbers are not comparable, so check which basis any retention benchmark is on before you hold your own against it. Giving dollars grew 5.0% across 2025, and that growth came overwhelmingly from major and supersize gifts. Through the first quarter of 2026 it broadened, with donor counts and dollars both growing across Small, Midsize, Major and Supersize bands, and only Micro donors still declining.
The catch is that retention moved the other way. It fell for Small, Midsize, Major and Supersize in the same quarter, and Micro was the only band where it improved. More donors came in, and the ones already there held on slightly less well. That is the pattern lapse prediction is supposed to catch, and a revenue total is designed to hide.
The sharpest finding is about second gifts. In the Q1 2026 data, donors who gave once the previous year were retained at 7.4%. Donors who gave twice were retained at 19.4%. The largest single jump anywhere in the dataset sits between the first gift and the second, which is exactly where most files leak. Blackbaud's own benchmarking puts median new-donor retention around 24%. Three out of four first-time donors never give again, and most organizations find out only after the year-end report.
That is the case for prediction: a lapse is rarely sudden. Donors drift, and the drift leaves tracks in data you already hold. Software that reads those tracks buys you the one thing retention work needs most, which is time to act while the donor still feels connected.
What lapse prediction software actually does.
Donor lapse prediction software analyzes each donor's history and behavior, compares it against patterns that preceded past lapses, and surfaces the supporters most likely to stop giving so your team can intervene early. Depending on the tool, the output is a probability score, a ranked list, or a flagged donor with the evidence behind the flag.
Three things it is not:
- It is not a CRM. Every serious option sits on top of the donor database you already run and reads from it.
- It is not wealth screening. Tools like DonorSearch Ai model giving likelihood and response likelihood for prospects; that is acquisition intelligence, not lapse detection. The two pair well but answer different questions.
- It is not a guarantee. A prediction is a probability, and every model produces false alarms and misses. The value is in shifting your team's attention earlier, not in being right every time.
The signals every model watches.
Across every vendor we reviewed, the same core signals come up. If you want to sanity-check any tool's output, or build a manual early-warning report while you evaluate, this is the list.
| Signal | What it looks like in the data | Why it predicts lapse |
|---|---|---|
| Recency | Months since the last gift, against that donor's own rhythm | The single strongest drift indicator |
| Frequency deviation | A December-every-year donor missing December | Deviation from personal baseline beats any global rule |
| Giving trend | This year's total vs last year's at the same point | Downgrades usually precede departures |
| Gift downgrades | Smaller amounts at the same frequency | Commitment shrinking before it stops |
| Engagement decline | Email opens, event attendance, portal or website visits falling off | Attention fades before money does |
| Tenure and lifecycle | First-year donors vs 8-year veterans | New donors lapse at far higher rates and need different treatment |
| Gift count | How many separate gifts the donor has ever made | The strongest single predictor in the sector data, and the one already sitting in your CRM |
| Acknowledgment latency | Days between gift and thank-you | Delayed acknowledgment significantly raises lapse risk |
| Relationship gaps | Promised follow-ups that never happened, unanswered questions in notes | Invisible to score-only models, visible in your records |
Gift count deserves a note, because it outperforms most of what vendors sell. In the Q1 2026 figures, donors who had given once were retained at 7.4%, twice at 19.4%, three to six times at 44.6%, and seven or more times at 88.1%. Sorted by gift size instead, the same donors run only from 10.0% to 32.9%.
You do not need software to produce that split. Any CRM can count gifts per donor. If you are evaluating a prediction tool, run this cut on your own file first, because it sets the bar the tool has to beat.
For the full set of current benchmarks, split by donor type, gift size and gift frequency, see Fundraising Metrics and KPIs.
The last row is the one most scoring tools cannot see, because it lives in notes, emails, and documents rather than in structured gift fields. More on that below.
The tools compared.
Seven vendors that either predict donor lapse directly or ship native retention scoring. Pricing is what the vendor publishes, not what a demo would quote.
| Tool | What it is | Lapse prediction approach | Published pricing |
|---|---|---|---|
| Dataro | Prediction layer on top of your CRM | ML propensity scores per donor, churn-risk and reactivation flags | From $15,000/yr plus $0.10 per active donor |
| Virtuous Insights | Analytics engine inside the Virtuous CRM | Predictive scores for churn risk, giving likelihood, upgrades | Not published |
| Blackbaud (ResearchPoint, Prospect Insights Pro, FPM) | Analytics suite for Blackbaud environments | Attrition scores and lifecycle trend comparisons | Not published |
| Bloomerang | CRM with native engagement scoring | Engagement and generosity scores, retention dashboard | CRM from $125 per month |
| Funraise Fundraising Intelligence | Analytics inside the Funraise platform | Churn prediction, revenue forecasting, AI trend explanations | Not published |
| Fundraise Up | Donation platform | Predicts recurring-gift cancellation risk at the payment layer | Not published |
| Gratefully | AI donor intelligence layer on your existing CRM | Flags at-risk donors with cited evidence from your own records, notes, and emails | Published on our pricing page |
Dataro is the purest prediction play: machine-learning propensity scores built from your own CRM data, with churn-risk lists and reactivation predictions. Its public results are customer case studies, including a reported 531 extra monthly donors retained through targeted churn-risk outreach at Greenpeace. Dataro publishes no validated accuracy metrics, but it does publish pricing: Essentials from $15,000 a year plus $0.10 per active donor, Growth at $25,000 plus $0.12, and Enterprise on application, read on their pricing page in September 2026.
Virtuous Insights bundles churn-risk scoring into the Virtuous CRM, blending first-party behavior with third-party wealth and demographic enrichment. Strong if you are already migrating to Virtuous; it is not sold as a standalone layer for other CRMs.
Blackbaud's tools serve Raiser's Edge and Blackbaud CRM environments, turning benchmarking data into attrition scores. The fit question is simple: if you are not in the Blackbaud ecosystem, this is not your shortlist.
Bloomerang's native scores are the most accessible entry point: every constituent gets an engagement score out of the box. Bloomerang reports that the top 20% of its customers see 47% first-time donor retention against an 18.5% industry average, a vendor-reported figure worth probing in a demo.
Funraise and Fundraise Up both attack the recurring-giving side: Funraise flags donors who usually give but have gone quiet, and Fundraise Up predicts recurring-gift cancellations at the checkout layer, reportedly saving 27% of donors who were about to cancel.
Gratefully takes a different approach, which deserves its own section, because the difference matters more than a feature list.
Scores vs evidence: two ways to predict a lapse.
Every tool above outputs some version of a number: a propensity score, a risk band, an engagement meter. Scores are useful for ranking a big file, and they all share one weakness. A score tells you who is at risk. It cannot tell you why, and it cannot see the risks that never touch a structured field.
Evidence-based risk detection starts from the other end. Gratefully reads the donor records, notes, and emails your team already keeps and builds a cited knowledge graph of every relationship. When Grace surfaces a donor as at risk, the answer draws on that evidence, the missed December gift, the question nobody answered in March, the site visit that was promised and never scheduled, with citations back to your own records. Instead of a bare number to trust, your team gets the context behind the risk, and the context usually is the script for the save. It is how Gratefully surfaces hidden revenue sitting in records you already hold, and it is why prediction and how Gratefully works are the same story: the intelligence layer that reads what your CRM stores.
What vendors will not tell you.
Nobody publishes accuracy.
Not one lapse-prediction vendor publishes precision, recall, or validation methodology for its models. Every accuracy figure in the market, DonorSearch's 81% repeat-donor accuracy, Bloomerang's 47% retention cohort, Fundraise Up's 27% cancellation saves, is vendor-reported and unaudited. Treat them as marketing claims to verify in your own data.
Almost nobody publishes pricing.
Of the seven tools in the table, only Bloomerang's base CRM price and Gratefully's plans are public. Everything else is request-a-demo. Budget accordingly: opaque pricing usually means negotiated pricing, and negotiated pricing usually means five figures a year at mid-size scale.
Seven questions to ask before you buy.
- What data does the model need, and how much history?If the honest answer is five years of clean gift data and you have two, the demo will not survive contact with your database.
- Show me a flagged donor. Can you show me why they were flagged?Score-only tools will show you a number. Decide whether a number is enough for your team to act.
- What happened in your validation testing?Ask for precision and recall, or in plain terms: of the donors you flagged last year, how many actually lapsed, and how many lapsed donors did you miss?
- Does it read our unstructured data?Notes, emails, and documents hold the relationship signals. Most scoring tools only read structured gift fields.
- What does the intervention workflow look like?A risk list that exports to CSV is a report. Look for tools that turn a flag into an owned, tracked next action.
- What is the all-in first-year cost?Platform fee, implementation, integration, and the staff hours to act on the output.
- Where does our donor data go?Whether PII is redacted before anything reaches an AI model, whether your data trains shared models, and what the audit trail shows.
Our AI Policy Pack has the full vendor-evaluation checklist for the last question.
What to do when a donor gets flagged.
Prediction only pays when the flag becomes contact. Prediction tools tell you who is slipping; for the donors already gone, here is how to win back lapsed donors step by step. A simple playbook:
- Step 1Check the evidence first.Thirty seconds in the record before any outreach: what changed, and is there an obvious cause like a completed pledge or a seasonal pattern.
- Step 2Match the response to the relationship, not the score.A first-year donor gets a warm, specific thank-you and an update on what their gift did. An eight-year donor gets a personal call from someone who can reference the relationship.
- Step 3Reference what you know.Mentioning the donor's own history is the cheapest proven lift in reactivation work. Generic we-miss-you messages read as list mail.
- Step 4Fix the cause, not just the symptom.If the flag traces to an unanswered question or a broken promise, the save is closing that loop, not sending an appeal.
- Step 5Log the intervention and the outcome.Next quarter's model, and next year's staff, are only as good as what you write down.
When you do not need prediction software.
You probably do not need a dedicated prediction tool if any of these are true:
- Your active file is under a few hundred donors. A monthly review of recency and giving trend in a spreadsheet will catch most drift, using the signals table above.
- You cannot staff the follow-up. Flags without capacity to act just generate guilt. Fix the acknowledgment process first; slow thank-yous are a lapse driver you can eliminate for free.
- Your data will not carry a model. If gifts live in three systems and notes live nowhere, invest in consolidation first. That said, this is exactly the situation where an intelligence layer that reads messy, unstructured records earns its keep, so the answer may be sequencing rather than skipping.
Frequently asked questions
See lapse risk with the evidence attached.
Gratefully sits on top of the CRM you already run and flags at-risk donors with cited evidence from your own records, notes, and emails. Most organizations are set up in under 60 minutes.