Every AI use case, mapped to the business.

32 use cases owners are being sold this year, sorted by where they act in a business of 5 to 100 people. Only 3 are proven in production.

Status
Pick a room to see its use cases
How we rate

Four statuses, one rule: rate the evidence, not the pitch.

Each entry gets a status for how proven it is in businesses of 5 to 100 people, and a separate rating for how strong the evidence behind that status is.

Evidence strength

StrongSeveral independent studies or large independent surveys.
ModerateAt least one independent study or credible survey, plus consistent case evidence.
ThinA handful of independent data points.
Vendor claimsOnly what sellers publish about their own products.

How an entry is made

Each entry is drafted from public sources: academic studies, regulators, independent surveys and reputable reporting. Every figure carries a numbered citation you can open. Anything a vendor says about its own product is labelled as a vendor claim.

Review and updates

Every source in this edition was opened and checked against the text in October 2026. Each entry is then signed off by a named operator and re-checked when the evidence moves. Until that sign-off, the entry is marked as a draft at the top of the page.

The full guide

All use cases at a glance

Field GuideSales & new businessAI lead scoring
In pilotsSales & new business

AI lead scoring

Also called predictive lead scoring, AI lead prioritisation, lead qualification scoring, propensity scoring

EvidenceThin
ClassAnalytics

AI lead scoring ranks enquiries by how likely they are to buy. It needs more closed deals than many small firms have. What the evidence shows.

ReviewDraft, awaiting operator review
Drafted27 October 2026
Fact-checkedEvery source opened, October 2026
Key takeaway

Predictive lead scoring can rank leads accurately when there is enough history, but evidence that it raises sales is thin. A 2024 review of 44 studies notes that little is known about its impact on sales performance⁠. A single-company case study reported 98% accuracy on 16,600 CRM records without measuring revenue⁠. The tools also need data and budget: one CRM requires at least 40 qualified and 40 disqualified leads⁠, and another reserves AI scores for a tier costing $3,600 a month⁠.

At a glance

StatusIn pilots
What it doesGives each new lead or deal a score for how likely it is to buy, based on patterns in your past won and lost leads, so salespeople call the best ones first.
Typical costUsually bundled into higher CRM tiers: one CRM includes it from its $105 per user per month enterprise sales plan, with full capacity at $150⁠, or $3,600 per month for another's top marketing tier⁠ (USD, vendor list prices, checked Oct 2026)
Data it needsHundreds of past leads closed as won or lost, with source, industry, company size and activity fields filled in the same way every time.
Setup effortMedium: switching the feature on is easy, but cleaning lead history and agreeing what 'qualified' means takes weeks.
EvidenceThin. One systematic review, a few single-company studies of model accuracy and vendor documentation; almost no independent measurement of sales lift.

What is AI lead scoring?

Lead scoring puts a number on each enquiry or contact so the sales team knows whom to call first. Traditional scoring is a set of rules someone writes by hand: ten points for a director's job title, five for opening a pricing email. AI or predictive scoring replaces those hand-written rules with a model trained on your own history of leads that did and did not convert⁠.

For an owner-led business the promise is focus. If two salespeople receive 300 enquiries a month, a good score tells them which 30 deserve a call today. The idea is sound; machine learning simply finds the pattern in data rather than in opinion. Mainstream CRMs, including the two we checked, offer it as a feature on higher-priced plans rather than as a separate product⁠.

How does it work?

The CRM takes past leads with their outcome, won or lost, and the fields known at the time: lead source, industry, company size, job title, pages visited, emails opened. A standard prediction model, such as a decision tree or one of its refinements, learns which combinations predicted a win⁠. Each new lead then receives a probability or a grade, and the model is retrained as new outcomes arrive.

The hard part is the training data. One CRM vendor will not start a model until at least 40 qualified and 40 disqualified leads have been created and closed in the chosen window, which defaults to two years⁠. Another needs at least 50 contacts, 25 converted and 25 not⁠. Those are minimums for the software to run, not for the scores to be good. In the published case study, 23,154 raw records had to be cleaned down to 16,600 usable ones⁠.

What it needs to work

Predictive scoring only works if your CRM holds a long, honest record of which leads became customers and which did not, captured the same way every time.

  1. At least 12 to 24 months of leads in one CRM, each closed as won or lost with a date rather than left open
  2. A minimum of roughly 40 won and 40 lost leads in the training window to switch the feature on, and ideally several hundred⁠
  3. Fields known at the time of enquiry, filled consistently: lead source, industry, company size, job title, region and product of interest
  4. Lead source tags that match your Google Ads and Meta Ads campaigns, so spend can be joined to outcomes
  5. A written definition of a qualified lead that sales and marketing both use
  6. A monthly check comparing each score band with its actual conversion rate

Most use cases in this guide succeed or fail here, on the data underneath them, before any AI is involved.

What does the evidence show?

The academic work mostly measures whether models predict well, not whether firms sell more. A 2024 systematic review of 44 studies starts from the point that little is known about lead scoring's impact on sales performance, even as it expects predictive models to replace traditional ones⁠. A 2025 case study at a small or mid-sized B2B software firm found a machine-learning model (gradient boosting) classified leads with 98% accuracy, but did not compare revenue against the old method⁠.

The few real-world outcome results are older. A 2017 B2B customer-acquisition experiment using social media data reported a 4.75% rise in prospect response rate, as summarised by a market research firm⁠. A Gartner survey of 227 chief sales officers (August to September 2025, released May 2026) found organisations giving sellers AI-enabled next best actions were 2.6 times more likely to achieve commercial growth⁠. That is broader than lead scoring, growth is undefined, and it is a correlation, not proof of cause. We found no independent study of AI lead scoring in businesses of 5 to 100 staff.

Predictive (AI) scoring vs rule-based scoring

Compared onPredictive scoringRule-based scoring
How scores are setLearned from past won and lost leadsPoints assigned by hand for job title, source and activity
Data neededHundreds of closed leads with clean fields⁠Only the fields you choose to score
ExplainabilityVaries; often a list of top factorsFully transparent
Cost modelUsually top CRM tiers⁠Included in some mid-tier CRM plans⁠
Best forFirms with steady lead volume and years of CRM historyFirms with a few hundred leads a year or fewer

Status and evidence

StatusIn pilots
Evidence strengthThin

Rated pilots: predictive scoring is available off the shelf in mainstream CRMs⁠, but the evidence measures prediction accuracy far more than sales results⁠. AI use of any kind is still a minority practice among US businesses: 19.8% in May 2026 Census Bureau data⁠, up from 5.4% in early 2024⁠. Evidence is thin because we found no independent study measuring revenue or conversion gains from AI lead scoring in small firms.

Where does it go wrong?

  • Too little history. A business closing a dozen deals a month may not reach a vendor's minimum of 40 qualified and 40 disqualified leads within three months⁠, and a model trained on a handful of wins learns noise.
  • Messy fields. In the published case study, the lead source and job title fields had outlier rates of 13% and 19%⁠. If salespeople skip fields or close lost deals months late, the model learns their habits, not buyer behaviour.
  • Self-fulfilling scores. Leads marked low get fewer calls, convert less, and so confirm the score.
  • Paying for a tier you do not need. Predictive scoring often sits in the higher-priced plans⁠.
  • Fixing the wrong problem. Research in Harvard Business Review found most companies respond to online leads far too slowly⁠. If enquiries wait a day for a reply, ranking them will not help much.

Numbers it should move

Quote conversionThe share of quotes that turn into orders.
Response timeHow long a customer waits for a first reply.

Ask the demo business

Northgate Supply is a fictional wholesale distributor of catering and packaging supplies: 28 staff, R46.6m in sales over the last year. Its data is generated for this demo, to 30 September 2026.

Fictional data
What share of our quotes turn into orders?
38%of quotes won in September

We won 38% of 1 909 quotes over the year. By rep over the last six months, the range runs from 30% to 46%.

OctNovDecJanFebMarAprMayJunJulAugSep38%
Thandi46%
Pieter35%
Aisha42%
Marco30%
How this was worked out

Quotes marked as won divided by all quotes issued, by month and by rep.

Answered in 1 msScanned 1 909 rowsFrom quotes
Want this on your own numbers, in the AI tools you already use?Book a demo

Where the data usually lives

HubSpotSalesforceZoho CRMPipedriveGoogle AdsMeta Ads

Where the records this use case needs are usually kept in a small business. It is not a list of recommended tools.

Frequently asked questions

How many leads do I need for AI lead scoring to work?

More than the software minimums, which start at 25 to 40 each of won and lost leads⁠. One vendor requires 40 qualified and 40 disqualified leads closed within a training period of three months to two years⁠; another needs 50 contacts, 25 converted and 25 not⁠. Those numbers let the model run; they do not make it accurate. A firm closing a few hundred leads a year is near the floor. Below that, simple hand-written rules are easier to trust and check.

How much does AI lead scoring cost?

It is rarely sold on its own. One CRM includes predictive lead and opportunity scoring from its enterprise sales plan at $105 per user per month, billed yearly, with full capacity in its $150 premium plan⁠. Another reserves AI lead scores for its top marketing tier at $3,600 a month plus a $7,000 onboarding fee⁠. Both are USD vendor list prices, checked Oct 2026. Rule-based scoring appears in cheaper tiers⁠.

Do I need a data scientist or IT team for lead scoring?

Not for built-in CRM features: you switch them on in settings and the vendor trains the model⁠. You do need someone who owns CRM data quality, makes sure deals are closed as won or lost, and checks each quarter whether high-scoring leads really convert more often. A custom model, like the one in the published case study⁠, does need data skills.

Is AI lead scoring allowed under POPIA and UK GDPR?

Generally yes, with care. Predicting people's behaviour counts as profiling under UK GDPR, and extra rules apply to solely automated decisions with legal or similarly significant effects⁠. POPIA section 71 restricts decisions with legal consequences or substantial effects based solely on automated profiling⁠. Choosing whom to call first is unlikely to reach that bar, but say in your privacy notice that you score leads. Check the ICO or Information Regulator; this is not legal advice.

Is lead scoring worth it if we already reply to every enquiry?

Speed comes first. Research in Harvard Business Review found most companies respond to online leads far too slowly⁠. If your team already replies to every enquiry within hours, scoring adds little until volume outgrows the team. It earns its keep when there are more leads than people to call and enough closed history to learn from⁠.

More in sales & new business

AI call notesIn pilotsEvidence: ModerateAI proposal draftingIn pilotsEvidence: ModerateAI outbound sales agentsUnprovenEvidence: Thin

Sources

  1. 1
    The state of lead scoring models and their impact on sales performance
    Information Technology and Management (Springer), via IDEAS/RePEc, 2024Study
  2. 2
  3. 3
  4. 4
    Configure predictive lead scoring
    Microsoft Learn, 2025Vendor
  5. 5
    Overview of the lead scoring tool
    HubSpot Knowledge Base, 2026Vendor
  6. 6
  7. 7
  8. 8
  9. 9
    What is ADM?
    Information Commissioner's Office (UK), 2026Government
  10. 10
    Section 71 Automated decision making - POPIA
    Protection of Personal Information Act 4 of 2013 (text via popia.co.za), 2013Government
  11. 11
  12. 12
    The Short Life of Online Sales Leads
    Harvard Business Review, 2011Study
  13. 13

This entry is educational. It is not legal, financial or professional advice, and it does not recommend any vendor. Figures attributed to vendors are their own claims.

Ask your own business. Get the true number in seconds.

Veld connects your accounting, sales and operations data to the AI tools you already use, and keeps it running, so every answer comes from your numbers.

Book a demo