Games Contact us
Games Contact us Finance Crypto Finance Fintech & Transfers Insurance Financial Reporting Banking Budgeting & Planning Auditing & KPIs Financial Planning Accounting Bookkeeping Cost Accounting Financial Statements Accounts Payable & Receivable Auditing Fixed Assets & Depreciation Accounting Software IFRS & GAAP Standards Marketing Brand Strategy Content Marketing SEO & AI Search Social Media Email Marketing Digital Ads TikTok Marketing & Shop Growth Hacking Marketing Analytics Pricing Psychology Brand Ambassadors Tools & Comparisons HR Compensation & Benefits Employee Engagement HR Strategy Recruitment & Talent Acquisition Sales B2B Sales AI in Sales CRM Systems Cold Outreach Pricing Strategy Pipeline Management Sales Enablement Sales Leadership Technology AI Tools & LLMs Cloud Infrastructure Cybersecurity Data Analytics Emerging Tech All β†’ Startup Corporate Governance Law Procurement Procurement: Sourcing Procurement: Vendor Management Procurement: Supply Chain Procurement: Contract Negotiation Procurement: Cost Reduction All Departments
Select Page

Applied AI: How Software Companies Put AI to Work

Kurums Applied AI Β· Series hub

Applied AI: how software companies put AI to work

A department-by-department review of the AI that CRM, HR, marketing, accounting and procurement software vendors now ship: what it does, how it is built, what it costs and whether the evidence says proven, emerging or hype.

5departments reviewed
40vendor AI products rated
40use cases scored
75sources checked
The verdict system

How do we decide whether AI in software is proven or hype?

Every review in the series rates each vendor product and each use case on the same three-step scale. The scale is about evidence, not ambition: a modest feature with years of measurable results beats a headline agent with a demo and a press release.

Proven

Generally available, used by a broad customer base, and backed by results we could trace beyond a single vendor slide.

Emerging

Shipping, but adoption is early or the evidence is thin, mixed or only vendor-reported. Worth a controlled pilot.

Hype

Announced, in beta or demo-ware, or the claim runs ahead of anything a customer can verify today.

β„–01 CRM & Salesβ„–02 HRβ„–03 Marketingβ„–04 Accounting & Financeβ„–05 Procurement

Use-case maturity by department

How many of the AI use cases we scored in each department are proven, emerging or hype.

CRM & Sales
431
8 use cases
HR
341
8 use cases
Marketing
341
8 use cases
Accounting & Finance
341
8 use cases
Procurement
422
8 use cases
The reviews

Which departments does the Applied AI series cover?

Each review appears here and on its own department page. New reviews in a department are added to the top of its card automatically.

CRM & Sales

β„–01

2 proven Β· 6 emerging Β· 0 hype across 8 vendor products

Read the full review β†’Sales department

HR

β„–02

2 proven Β· 6 emerging Β· 0 hype across 8 vendor products

Read the full review β†’HR department
Advertisement

Marketing

β„–03

2 proven Β· 6 emerging Β· 0 hype across 8 vendor products

Read the full review β†’Marketing department

Accounting & Finance

β„–04

2 proven Β· 6 emerging Β· 0 hype across 8 vendor products

Read the full review β†’Accounting department

Procurement

β„–05

3 proven Β· 4 emerging Β· 1 hype across 8 vendor products

Read the full review β†’Procurement department
Across the series

What does the series find about AI in business software?

Read side by side, the 5 reviews tell a consistent story. The AI that works is narrow, trained on a company's own history and checked by a person. The AI that is being marketed hardest is the autonomous agent, and that is where the evidence is thinnest.

FINDING 01

Narrow, data-rich tasks are the proven layer

Lead scoring, call analysis, transaction categorisation, invoice capture, spend classification and job-description drafting have years of deployments behind them. They share one trait: large volumes of labelled history and outputs a person can check in seconds.

FINDING 02

Agents ship everywhere; evidence lags

Of the 40 vendor AI products we rated, 11 are proven, 28 emerging and 1 hype. Most agent results available today come from vendor pilots and earnings calls rather than independent measurement.

FINDING 03

Pricing is moving from seats to meters

Credits, AI units, per-lead and per-action fees now sit alongside licences at Salesforce, HubSpot, Workday, SAP, Microsoft and others. Budget for consumption, set caps and check whether unused credits expire.

FINDING 04

Full autonomy is where the hype sits

Every use case we rated as hype is an end-to-end, unsupervised version of a job: fully autonomous outbound SDR (CRM); autonomous source-to-offer (HR); fully autonomous, self-driving marketing (Marketing); autonomous month-end close (Accounting); fully autonomous strategic sourcing (Procurement); one agent running all of source-to-pay (Procurement). In each department vendors still require a human to approve the final step.

FINDING 05

Regulation has become a product requirement

The EU AI Act treats AI used in hiring, promotion and performance decisions as high-risk, with obligations now scheduled for December 2027, and its transparency duties for chatbots and AI-generated content apply from August 2026. New York City already requires bias audits for automated hiring tools. Vendors that log decisions, explain outputs and keep a human in the loop are easier to buy.

Advertisement
The full scorecard

Which AI use cases are proven in each department?

All 40 use cases from the series, grouped by department. Open a department to see the verdict and the reason behind it.

CRM & Sales4 proven Β· 3 emerging Β· 1 hype
Predictive lead and deal scoringProvenShipped for years in Salesforce, Zoho, Pipedrive and Dynamics; needs enough won and lost history.
Contact and account enrichmentProvenProvider-backed data is mature; agent-led web research layered on top still needs spot checks.
Conversation intelligenceProvenCall recording, transcripts and coaching are broadly deployed, now bundled into CRMs like Pipedrive.
AI forecastingProvenUsed weekly by sales managers, though vendors publish no independent forecast error rates.
Email and sequence draftingEmergingEverywhere, but evidence of better reply rates than human-written templates is thin.
Automatic CRM data entryEmergingSuggest-and-approve updates work; silent auto-writes risk spreading errors through the pipeline.
Inbound qualification agentsEmergingGA from Salesforce, Microsoft and HubSpot; results so far are vendor pilots and case studies.
Fully autonomous outbound SDRHypeClaims of replacing SDR teams outrun evidence; vendor pilots show reply rates near 1.5%.
HR3 proven Β· 4 emerging Β· 1 hype
Job description draftingProvenShort output, always human-reviewed, built into most ATS and HCM suites.
Conversational schedulingProvenYears of frontline deployments with named customer results on time to hire.
Self-service HR answersProvenPolicy and pay Q&A ships in every major suite; value depends on clean source documents.
CV screening and gradingEmergingWidely sold, but high-risk under the EU AI Act and central to live bias litigation.
Skills inferenceEmergingUseful for mobility, but few vendors publish accuracy and profiles go stale.
Performance review draftsEmergingTime savings are vendor-reported; scoring employees triggers high-risk duties.
Payroll anomaly detectionEmergingAgents now explain variances, but gains so far are modest and vendor-reported.
Autonomous source-to-offerHypeEnd-to-end hiring agents are mostly roadmap; AI interviewers lack independent results.
Marketing3 proven Β· 4 emerging Β· 1 hype
Copy and image variant generationProvenUsed daily at scale; output is easy for humans to check before it ships.
Predictive segmentation (CLV, churn, send time)ProvenMature models in Klaviyo, Adobe and Salesforce; reliable with enough purchase history.
Ad bidding, targeting and budget automationProvenAdvantage+ and Performance Max carry most spend; results are reported by the platforms themselves.
Automated brand-compliance checksEmergingShipping in GenStudio, Canva and Jasper; few published figures on errors caught or missed.
Generative creative inside ad platformsEmergingWidely used by SMBs; incremental lift over human creative is rarely published.
Campaign-building agentsEmergingMost in beta or recent GA during 2025–26; results are mostly vendor-reported.
AI search visibility tracking (AEO/GEO)EmergingReal demand, but answers vary so much between runs that single scores mislead.
Fully autonomous, self-driving marketingHypeEvery vendor still requires human approval before launch; no audited end-to-end results.
Accounting & Finance3 proven Β· 4 emerging Β· 1 hype
Transaction categorisationProvenRuns at huge volume in QuickBooks and Xero; errors are visible and easy to correct.
Invoice capture & AP codingProvenStandard in SMB and spend tools; line-level coding learns from corrections.
Bank reconciliation matchingProvenHigh-confidence matches automated for years; split and partial payments still need people.
Expense policy auditEmergingRamp and Brex ship it widely, but accuracy figures are vendor-reported.
Bill & payment fraud flagsEmergingUseful signal checks; AFP finds only 17% of firms use AI against payments fraud.
Reconciliation prep & accrualsEmergingBlackLine, Microsoft and NetSuite agents are new; most went GA or preview in 2026.
Forecasting & FP&A insightEmergingNarrative drafts help; Gartner finds forecasting among the lowest-rated AI uses.
Autonomous month-end closeHypeBenchmarks show errors compound over months; every vendor still requires sign-off.
Procurement4 proven Β· 2 emerging Β· 2 hype
Conversational intake and guided buyingProvenShipping at hundreds of enterprises via Zip, Coupa and others; cuts request and approval time
Spend classificationProvenMachine learning has categorised spend for years; accuracy still needs checking on your own data
Invoice capture and matchingProvenHigh-volume, rules-based work; vendors report very high touchless shares with human exception handling
Tail-spend negotiation botsProvenWalmart's Pactum pilot showed deals with 64% of invited suppliers; narrow but documented
Contract extraction and redliningEmergingExtraction works well; playbook-based redlining is new and depends on clean clause libraries
Supplier discovery and risk agentsEmergingUseful alerts, but quality tracks third-party data feeds and network bias
Fully autonomous strategic sourcingHypeCycle-time claims are vendor-reported; complex awards still need category managers
One agent running all of source-to-payHypeAnnounced in 2026 with beta status and no published outcomes
For software builders

What can SaaS companies learn from how vendors build AI?

The series reads every product from two sides: the buyer deciding what to switch on, and the software company deciding what to build. One lesson from each department:

CRM & SalesKeep a human approval step

Pipedrive's Nova and Attio's suggested record changes ask the user to approve each update. That design builds trust and protects data quality, and it lets you collect approval data to improve the model before offering full autonomy.

HRLet humans set the criteria

Ashby's criteria-based labelling is easier to audit and defend than a single fit score, and its published FairNow audit doubles as a sales asset. Build explainability in from the first release rather than retrofitting it.

MarketingGovernance sells more than generation

Adobe, Canva and Jasper all compete on brand rules, approvals and compliance checks rather than raw output quality. Software companies adding generative features should build the review layer first, because that is what enterprise buyers pay for.

Accounting & FinanceExplainability is a feature, not a footnote

Ramp cites a source for every field and BlackLine attaches reasoning and confidence to each recommendation. In finance, users accept automation they can audit. Build the explanation into the interface from the first release.

ProcurementGround agents in your own network data

Coupa sells anonymised community spend data and Sievo trains on pooled customer data. A proprietary dataset makes recommendations harder to copy than a wrapper around a general model.

Across all fivePublish evidence buyers can check

The vendors that earn a proven verdict share baselines, sample sizes and named customers. Unqualified percentage gains are the fastest way to an emerging verdict.

Advertisement

Frequently asked questions

What is applied AI in business software?

Applied AI is artificial intelligence built into the products people already use for work, such as a CRM that scores leads or an accounting system that codes invoices. The series looks at those features rather than at general-purpose chatbots or research models.

Which departments does the Kurums Applied AI series cover?

The series currently reviews CRM and sales, HR, marketing, accounting and finance, and procurement software, rating 40 vendor AI products and 40 use cases. New departments and follow-up reviews are added to this hub as they are published.

Is AI in business software worth paying for in 2026?

Often, for narrow tasks with good data behind them: scoring, call analysis, categorisation, invoice capture and drafting. Autonomous agents are worth a controlled pilot with a control group and a spending cap rather than a company-wide rollout.

Why does pricing for AI features keep changing?

Vendors are moving from per-seat licences towards credits, AI units and per-action fees because agent workloads vary widely between customers. That makes costs harder to forecast, so model consumption at realistic volumes before you sign.

Did any vendor pay to be included?

No. No company paid for inclusion, position or verdict, and nothing in the series is a purchase recommendation. Prices and product names were checked against vendor pages and reputable reporting on 11 October 2026.

About this series

Kurums Applied AI is researched from vendor documentation, pricing pages, earnings calls, regulatory texts and independent reporting. Every figure is attributed and vendor-reported numbers are labelled. No company paid for inclusion or position, and the verdicts are editorial judgements about evidence, not recommendations. Reviews are updated as products and prices change.

Last Updated: October 2026 · Reviewed by the Kurums Technology editorial team.