# Modulai vs Walma

> Two top-rated AI Consulting agencies in Stockholm. Side by side: published rates, team size, service and industry coverage, and the curator's assessment of both — pairings are never sold.

Canonical page: https://www.aiagencies.eu/compare/modulai-vs-walma
Generated: 2026-09-20 · Curated by Gabor Kiss · No paid placements — inclusion, score, and rank are never sold.

## Facts
| | Modulai | Walma |
|---|---|---|
| Editorial score | 61.8/100 | 60.3/100 |
| Hourly rate | not published | not published |
| Team size | 10-49 | — |
| Location | Stockholm, Sweden | Stockholm, Sweden |
| Founded | 2018 | — |
| Last reviewed | 2026-08-05 | 2026-08-05 |
| Services | AI Consulting, AI Development, Generative AI, AI Agents | AI Consulting, AI Agents, AI Automation, AI Development |
| Industries | FinTech, Healthcare, Manufacturing, E-commerce | Public Sector, Manufacturing, SaaS & B2B |

## Our take on Modulai
Reviewed 2026-08-05 by Gabor Kiss against published evidence.

The most technically specific portfolio in this register, and the one where a reader with a machine-learning background learns the most.

The NovoNordisk work is a multi-agent retrieval system for clinical trial evidence that retrieves, analyzes and returns results in the right format while keeping traceability—the case index names the client, though the study itself describes it as one of Europe's largest pharmaceutical companies. For an unnamed American software company it built a Bayesian hierarchical logistic regression for small-business credit risk: a model choice, stated as such, in a market where most firms write 'AI-powered'. Fuelmatics needed a robotic arm to find a car's fuel lid in real time, so the detector was optimized to run on CPU rather than assuming a GPU budget. Optilogic got a custom agent embedded in its own supply-chain platform, turning plain language into SQL for users who would otherwise queue behind technical staff.

The published case index runs past a dozen engagements, and the engineering blog carries the same register: how to evaluate retrieval systems with synthetic data and an LLM judge, how a deep-research multi-agent system is built.

The gap is outcomes. Case after case describes what was built and stops, so the portfolio proves capability and never impact. Ask each reference what changed.

### Key strengths
- Names its models and its reasoning—Bayesian hierarchical regression, multimodal RAG, CPU-optimized detection—rather than describing capabilities
- Klarna and Novo Nordisk among named clients, which is a rare pairing of consumer fintech and regulated pharma
- Publishes method openly, including how it evaluates retrieval quality, so the engineering standard is checkable before you engage

### Good to know
- More than a dozen published cases and no outcome figure in any of them—this team documents how it works, not what it earned
- A 30-plus engineering team taking research-grade problems; confirm capacity and timeline before committing to a deadline

## Our take on Walma
Reviewed 2026-08-05 by Gabor Kiss against published evidence.

Four published engagements, and every one of them ends in a number: support response time down 65% with roughly three hours a day freed per case handler, forecasting accuracy up 40%, sales proposals from two hours to eight minutes with 80% less manual administration, internal service in a municipality roughly 50% faster.

That consistency is unusual. Most agencies publish one quantified case and four descriptions.

Behind them sits a product line—Noda, an enterprise AI platform built around Swedish data sovereignty, and Ocle—plus an EU AI Act advisory practice, which is a sensible pairing for a market where public-sector and regulated buyers are a large share of demand.

The gap is attribution. The four cases are anonymized by sector while ten client names sit elsewhere on the site, and the two are never connected. The register permits anonymized work when the method and outcome are described, which these are; a buyer should still ask which name goes with which result.

### Key strengths
- Every published engagement carries an outcome figure, and they are operational rather than vanity metrics—hours per handler, minutes per proposal
- Owns its platform, so the data-residency answer is architectural rather than contractual—relevant for Swedish public-sector buyers
- Runs an EU AI Act advisory practice alongside delivery, which most agencies of this size do not attempt

### Good to know
- The four quantified cases are anonymized and the ten client names attach to none of them—ask which is which
- No team size or founding year published; the organization number and address are there, the scale is not

## Which to choose
Choose Modulai for:
- larger team capacity for multi-stream or enterprise-scale programs
- disclosed specialization in FinTech / Healthcare
- broader service offering — also covers Generative AI

Choose Walma for:
- closer collaboration at a smaller team scale
- disclosed specialization in Public Sector / SaaS & B2B
- broader service offering — also covers AI Automation

## Questions buyers ask

### Which has more Manufacturing experience, Modulai or Walma?
Both agencies show documented Manufacturing work. Modulai has the broader industry stack overall, with disclosed experience in FinTech, Healthcare, E-commerce beyond their shared focus.

### Which scores higher overall, Modulai or Walma?
Modulai scores 61.8/100, 1.5 points higher than Walma at 60.3/100. Both portfolio quality and business credibility feed into the score — see our methodology for the full breakdown.

### When should I consider both Modulai and Walma?
Consider running parallel discovery briefs with Modulai and Walma if your project spans multiple workstreams, you want competitive proposals to compare scope and approach, or you're undecided between the specialization angles each brings (see the Best for cards above). Most engagements ultimately go with one — but the parallel-brief phase is a low-cost way to validate fit.

---
Source: AIAgencies.eu — the curated register of European AI agencies. Rate provenance: "agency-confirmed" = disclosed via submission or claimed profile; "directory-listed" = Clutch or two agreeing B2B directories (lone unverified sources are never written). Full method: https://www.aiagencies.eu/methodology · Rate index: https://www.aiagencies.eu/rates