From Family Distributor to Industrial AI: Why Lubinpla Built AI Crew

Summary: The industrial chemical distribution sector enters 2026 caught between two structural pressures, margin compression from commodity price transparency and a technical workforce nearing mass retirement, while the same period coincides with venture capital reframing vertical artificial intelligence agents as a 10x larger opportunity than legacy software-as-a-service (SaaS). Lubinpla, an industrial chemical AI agent company headquartered in Korea, was built by a founder whose path crossed three relevant disciplines, data analytics, a family chemical distribution business, and multi-startup business-to-business large language model (LLM) product development. This origin shaped two products, AI Shooting for per-case industrial chemistry analysis and AI Crew for subscription-based agent workflows, both designed around the technical-distributor's actual day, not the generic enterprise SaaS template. This article documents the market structure that made domain specialization necessary, the founder background that made it buildable, the squeeze-factor matrix that distributors can use to assess their own position, and the rationale for treating industrial chemistry as a vertical that resists generic agents. The purpose is to articulate Lubinpla's positioning at the industrial AI inflection: not a generic copilot bolted onto a distributor's email, but a domain-specialized crew built for the moment a retiring technical workforce stops being available. Procedures, findings, and conclusions are organized across nine sections, with the squeeze-factor matrix in Section V serving as the operator-actionable instrument distributors can apply to their own books. Conclusions follow in Section VIII.

Table of Contents

I. Introduction: A Market at Inflection

II. Two Kinds of Distributors: Delivery versus Technical

III. The Squeeze on Distributors: Hiring Freeze, Aging Workforce, High-Value Industry Shift

IV. Founder's Path: Data Analytics, Family Distributor Experience, Multi-Startup B2B LLM Development

V. Why Domain-Specialized Tools, Not Generic SaaS

VI. The Lubinpla Mission: Technical Moat for Distributors at the Industrial AI Inflection

VII. What This Means for Distributors and Manufacturers

VIII. Key Takeaway

IX. References

I. Introduction: A Market at Inflection

Two structural facts collide in 2026. The global chemical distribution market reached USD 265.15 billion in 2024 with a forward compound annual growth rate (CAGR) of 7.26 percent through 2034 (Polaris Market Research, 2025), and the same industry is losing the experienced technical staff who explain those chemicals to buyers. Lubinpla, an industrial chemical AI agent company headquartered in Korea, was built specifically for the gap that opens when those two forces meet.

A growing market, a shrinking workforce

The chemical distribution market expanded to USD 265.15 billion in 2024, with the Asia-Pacific region capturing 62.2 percent of global revenue share (Polaris Market Research, 2025). Industry forecasts place the market at USD 543.01 billion by 2035 (Precedence Research, 2025). At the same time, the United States Bureau of Labor Statistics (BLS) projects that the labor force aged 55 and older will reach approximately 41 million workers, with about 13 million aged 65 and older, and the workforce aged 75 and older is the fastest-growing labor cohort in the country (United States Bureau of Labor Statistics, 2025). The chemical industry, where institutional product knowledge is concentrated in senior technical staff, is structurally exposed to this curve (Boaz Partners, 2024).

Why this matters for distributors

The chemical distributor business model has historically depended on two assets, supplier relationships and senior application engineers. When a buyer at a coatings manufacturer asks which inhibitor package to use for a tropical export run, the answer comes from a person, not a catalog. That person is increasingly close to retirement, and the talent pipeline behind them is thin (Boaz Partners, 2024). Lubinpla's product line, AI Shooting for per-case analysis and AI Crew for subscription-based agent workflows, was designed for the moment that human becomes unavailable. AI Shooting is a single-case industrial chemistry analysis service that returns an evidence-based written report; AI Crew is the subscription that runs the same analytic capability continuously across a distributor's books.

II. Two Kinds of Distributors: Delivery versus Technical

The chemical distribution market is not one business. It is two businesses operating under the same label, and the squeeze on each is different. Delivery distributors move commodity volumes on logistics and price; technical distributors move specialty grades on application advice and formulation support (Safic-Alcan, 2024). The first business is price-transparent and margin-thin; the second business depends on knowledge that is increasingly difficult to staff.

Delivery distribution: logistics and price

The full-line distributor model, represented by the largest players in the segment, moves commodity and specialty grades together. Brenntag led the 2024 rankings at USD 16.8 billion in sales, followed by Tricon Energy at USD 13.1 billion and Univar Solutions at USD 11.5 billion (ICIS, 2025). The economics rest on scale, contract logistics, and supplier-agreement coverage. Price transparency in commodity chemicals limits margin, and customer switching cost is low when the differentiating service is on-time delivery.

Technical distribution: formulation and application support

The specialty-only distributors, including IMCD and Azelis, operate a different model. The product is the same chemical, but the value the buyer pays for is the application engineer who recommends the grade, supports the formulation trial, and documents the regulatory profile (Safic-Alcan, 2024). IMCD reported approximately 47 percent gross margins, materially higher than full-line peers, driven by the formulation advice attached to each shipment (PitchBook, 2026). The economics of this model rest on technical staff, not logistics fleet. A technical distributor without application engineers is a delivery distributor with a more complicated catalog.

Why this distinction matters for AI

The two distributor types face different AI-relevant problems. A delivery distributor needs better logistics optimization, demand forecasting, and credit management, problems that generic enterprise software addresses reasonably well. A technical distributor needs to scale the cognitive output of a small number of senior application engineers, a problem that generic enterprise software does not address and that generic large language models (LLMs) handle unreliably because they lack the chemistry-specific grounding to refuse to answer when they are wrong. Lubinpla's product design is targeted at the second problem.

III. The Squeeze on Distributors

Three structural pressures compress the technical-distributor model simultaneously. The first is wage and headcount discipline imposed by margin compression in a normalizing post-destocking market. The second is the demographic curve of the technical workforce. The third is the shift of supplier portfolios toward higher-value specialty grades that demand more application support per dollar of revenue. The combined effect is a workload increase against a shrinking technical bench.

The hiring freeze and margin normalization

The chemical distribution industry exited the 2022 to 2023 destocking cycle with margins normalizing rather than recovering to pandemic-era peaks (TM Capital, 2024). Margin discipline now means headcount discipline. Distributors that historically absorbed the cost of carrying a deep application-engineering bench are reviewing that cost line as commodity-grade margin pressure persists. The McKinsey 2024 B2B distribution survey reported that approximately 95 percent of distributors are exploring artificial intelligence (AI) use cases, while only about 30 percent report sufficient internal talent to scale them (McKinsey and Company, 2024). The gap is the operating problem.

The aging technical workforce

Baby boomers constituted 25 percent of the United States labor force in the early 2020s, down from 49 percent in 1995, and the chemical industry sits inside the cohort that is moving out the fastest (United States Bureau of Labor Statistics, 2025; Boaz Partners, 2024). The transferable component of an application engineer's work, the heuristic knowledge of which grade fits which substrate under which exposure, is rarely written down. When a senior engineer retires, the codified record left behind is a small fraction of what they knew. Korean industry data tracks the same direction: South Korea's chemicals and materials sector technology workforce is forecast to reach 41,000 by 2030 from approximately 28,000 in 2025, but the broader industry employment growth rate of 2.1 percent annually means the technical share is gaining ground only by attrition elsewhere (Talenbrium, 2025).

The high-value industry shift

The specialty chemicals segment is forecast to grow at a 7.9 percent global CAGR, faster than commodity grades (Polaris Market Research, 2025). Specialty grades require more pre-sale technical work per dollar of revenue, including formulation trials, regulatory documentation, and sample qualification. A technical distributor whose portfolio is shifting toward specialty grades is structurally taking on more cognitive load per ton sold, at the same time the senior engineers who perform that work are aging out. The squeeze is not abstract. It is the same person being asked to handle more accounts with less time per account.

The squeeze-factor matrix

The matrix below is operator-actionable. A technical distributor can locate each row against its own books, mark the current trend, and decide whether the listed response is already in place. The intent is diagnostic, not prescriptive. The AI Crew fit column records where a domain-specialized agent platform is structurally suited to relieve the pressure, distinct from rows where the answer is a hiring, financial, or logistics decision.

Figure 1. Squeeze-Factor Matrix for Technical Distributors

Squeeze factor	Current trend (2024 to 2026)	Typical distributor response	AI Crew fit
Commodity-grade margin compression	Margins normalizing post-destocking; specialty share rising (TM Capital, 2024)	Portfolio shift toward specialty; reduce commodity exposure	Indirect; better technical-sales attach rate raises specialty revenue per account
Application engineer retirement	Baby boomer cohort exiting; technical succession gap (BLS, 2025; Boaz Partners, 2024)	Knowledge capture projects, mentoring programs, slower onboarding	Direct; encode application heuristics into a continuously available agent layer
Specialty grade workload per account	Specialty segment CAGR 7.9 percent (Polaris Market Research, 2025)	Account triage; concentrate engineers on larger accounts	Direct; lower-revenue accounts receive consistent technical response via agent
Compliance documentation burden	Regulatory specialty support expanding (DCAT, 2024)	Specialist compliance hire; outside consultant retainer	Direct; agent assembles compliance crosswalks against documented standards
Inbound technical question volume	Buyer expectations shaped by consumer chat experience (McKinsey and Company, 2024)	Email queue management; FAQ pages; slower response times	Direct; agent drafts technical replies against verified product and standard sources
Cross-region quote variation	Specialty pricing opacity preserves margin (Toffcap, 2024)	Manual quote-by-quote review by senior engineer	Indirect; agent surfaces precedent quotes from internal records

The matrix reads top-to-bottom as the working day of a technical distributor. Each row is a place where a senior engineer's time is currently being spent. The AI Crew fit column is not a claim that the agent replaces the engineer. It is a claim that the agent absorbs the parts of the day that do not require the engineer to be present, freeing the engineer for the parts that do.

IV. Founder's Path

The Lubinpla origin story is short because the founder's background is unusually direct on the problem. The combination of formal data analytics training, lived experience inside a family chemical distribution business, and multiple cycles of building business-to-business (B2B) LLM products at earlier startups is the specific combination that produces a domain-specialized industrial chemistry agent platform rather than a horizontal chat tool that happens to sell into chemicals.

Data analytics: the methodology backbone

The first leg of the founder's path was formal training in data analytics. The relevant carryover is not the toolchain. It is the methodological discipline of treating every claim as a statement that requires a source, every model output as a prediction with a confidence interval, and every deployed system as a process with measurable failure modes. Lubinpla's internal practice of citing source, year, and range for every numeric claim in customer-facing deliverables, the same practice this article follows, traces back to that methodological grounding. Industrial chemistry buyers do not pay for opinions. They pay for traceable evidence.

Family distributor experience: the operating reality

The second leg was lived inside a family chemical distribution business. This is the part that generic SaaS founders cannot acquire from interviews. Knowing that a senior application engineer carries dozens of accounts each with its own substrate-grade-exposure history, that an inbound technical question can take 90 minutes of catalog lookup and standards cross-referencing to answer correctly, and that a wrong recommendation costs the customer real money on a 200-ton paint batch, these are the operational facts that shape what a useful agent has to do. The shape of AI Crew, especially the requirement that agents cite the standard they relied on and refuse to answer when the standard does not exist, was set by that operating reality, not by an abstract product specification.

Multi-startup B2B LLM development: the engineering pattern

The third leg was multiple startup cycles building B2B LLM products before Lubinpla. The transferable engineering lessons were narrow but durable. First, generic LLMs hallucinate plausibly on chemistry topics and the failure mode is silent; the product must be designed to refuse rather than to confabulate. Second, the unit of integration that matters for B2B buyers is the workflow, not the chat window. Third, document grounding only works when the documents are curated and versioned, because uncurated retrieval-augmented generation (RAG) on the public chemical literature produces wrong-but-confident output at scale. These three lessons drove three concrete Lubinpla design choices: standards-anchored citation, workflow-shaped agents rather than freeform chat, and curated grounding scoped to verifiable industry references.

Why this combination is rare

The combination is rare for a structural reason. Data analytics people typically build horizontal tooling; family distributor children typically run the business they grew up in; B2B LLM founders typically target industries with cleaner data than industrial chemistry. The intersection produces a founder who can see why generic chat tools fail in this domain and has both the methodology and the engineering practice to build a domain-specialized alternative. The intersection is also why Lubinpla's positioning is not against horizontal chat assistants in general; it is against horizontal chat assistants asked to do industrial chemistry work, which is a specific and consequential mismatch.

V. Why Domain-Specialized Tools, Not Generic SaaS

The venture-capital reframing of vertical AI as a 10x larger opportunity than legacy SaaS is not a marketing slogan. It is a thesis about where domain knowledge creates defensible product surfaces. Bessemer Venture Partners projects vertical AI market capitalization at approximately ten times the size of legacy vertical SaaS, on the argument that vertical agents tap labor-line spend rather than software-line spend (Bessemer Venture Partners, 2025). Y Combinator President Garry Tan has stated the same thesis as a numerical forecast: the next decade will produce hundreds of vertical AI category leaders where the prior decade produced hundreds of SaaS category leaders (Y Combinator, 2025).

The labor-line versus software-line distinction

Legacy vertical SaaS captured a share of the customer's software budget, a line item measured in single-digit percent of revenue. Vertical AI agents address the customer's labor cost line, a much larger denominator. For a chemical distributor, the labor cost of the technical sales and customer-support function is typically a multiple of the IT budget. An agent that absorbs a fraction of that labor without displacing the senior engineers, because the senior engineers were already over-allocated, addresses a structurally larger spend pool.

Why generic chat tools fail on industrial chemistry

Generic LLM chat interfaces fail on industrial chemistry questions in a specific way: they answer fluently when they should refuse. Asked which corrosion inhibitor suits a chloride-loaded coastal export environment, a generic model will produce a plausible-sounding recommendation drawn from blended training data with no source attribution. The buyer cannot tell whether the recommendation is grounded in the American Society for Testing and Materials standard (ASTM) B117-19 salt spray methodology or in marketing copy from a 2017 product brochure. The cost of acting on the wrong recommendation is borne by the buyer on the application, not by the chat tool. Domain-specialized agents address this failure mode by anchoring to documented standards, citing the source on every claim, and refusing when the source is not present in the curated grounding set.

What a domain-specialized agent actually does

The domain-specialization claim collapses to three concrete behaviors. First, retrieval is scoped to a curated set of industry standards, supplier technical data sheets where licensed, and verified industry publications, not the open internet. Second, citations are attached at the sentence level so the customer can audit each claim. Third, the agent is designed to recognize when a question falls outside the curated set and to escalate to a human application engineer rather than answer. These behaviors are not novel in the broader AI agent literature, but their combined application to industrial chemistry distribution workflows is the surface Lubinpla operates on.

Field evidence from adjacent verticals

The vertical-AI thesis has empirical support from earlier-deploying industries. The legal-services vertical has documented domain-specialized contract-AI accuracy improvements over generic chat tools on contract-management tasks (Contractspan, 2025). Vertical agents tap into proprietary, structured data and embed AI directly into existing workflows, which is the structural pattern that distinguishes them from horizontal chat tools (SuperAnnotate, 2025). The same pattern applies to industrial chemistry distribution, where the relevant proprietary data is the distributor's own quote history, account application records, and supplier technical archive.

VI. The Lubinpla Mission

Lubinpla's stated mission, building the technical moat that lets chemical distributors stay relevant at the industrial AI inflection, is not a brand statement. It is a market position derived from the squeeze documented in Section III and the vertical-AI thesis documented in Section V. The mission is specific to the distributor segment because the squeeze is specific to the distributor segment. Manufacturers are large enough to staff their own AI teams. Distributors typically are not.

Two products, one funnel

The Lubinpla product line is structured as a two-step funnel. AI Shooting is the entry product, a per-case industrial chemistry analysis delivered as a written report. A customer submits one problem and receives one evidence-based analyzed answer at one of three service tiers, Quick at USD 20 for 24-hour triage, Standard at USD 50 for 3-day analysis, and Deep at USD 150 for 5-day investigation with field data review (IDENTITY.md, 2026). The product the customer experiences is the answer, not a tool to learn. AI Crew is the expansion product, a subscription that runs the same analytic capability continuously across the distributor's accounts at Individual USD 50, Team USD 150, Professional USD 400, and Enterprise custom tiers (IDENTITY.md, 2026).

Why the funnel runs from AI Shooting to AI Crew

The funnel exists because trust in domain-specialized agents has to be earned on a delivered result, not promised in a sales conversation. A distributor's technical-sales lead is reasonably skeptical that a generic AI tool can handle a chemistry question without confabulating. One delivered AI Shooting report, with cited sources and an explicit refusal-or-escalation pathway when evidence is thin, demonstrates the analytic discipline. The expansion to AI Crew then proposes scaling that same discipline across the daily workload. The order matters. Selling a subscription before delivering a result is the failure pattern of the generic-AI-for-chemistry market.

What AI Crew does not do

The mission is also defined by exclusion. AI Crew does not make a chemical safety decision. It does not promise that a recommendation is fit for a specific customer application; that determination remains with the senior application engineer or the customer's own technical staff. It does not replace the supplier relationship, the regulatory filing responsibility, or the legal accountability for product fitness. The agent layer absorbs the parts of the technical-distributor workday that are pattern-matching and document assembly. The judgment-grade decisions remain with humans. The boundary is not a marketing nuance. It is a product safety requirement.

VII. What This Means for Distributors and Manufacturers

The implication of the analysis above is asymmetric. Technical distributors face a near-term operating problem, which is how to staff specialty-grade application support against a retiring bench, and a near-term competitive risk, which is what happens when the first peer-distributor in their region deploys a domain-specialized agent layer ahead of them. Manufacturers face a longer-horizon question, which is whether their distribution channel will continue to carry the technical pre-sale work or whether some of that work shifts back upstream.

For technical distributors: the next 12 months

The squeeze-factor matrix in Section V is the diagnostic instrument. A technical distributor can run each row against its own books and identify which rows are degrading fastest. The application-engineer retirement row and the specialty-grade workload row are typically the fastest movers. The implication is not that an agent platform is the only response. The implication is that a distributor that does not have an answer to those rows is exposed to peer competition from distributors that do.

For technical distributors: the trust-building sequence

The recommended trust-building sequence, derived from the funnel logic in Section VI, is to start with one delivered analytic result before evaluating a subscription. Submitting a single hard case to AI Shooting, reviewing the cited evidence, and grading the deliverable against the senior application engineer's own answer is the lowest-cost way to assess whether domain-specialized agent output meets the quality bar the distributor's customers expect. Subscription decisions are then made against demonstrated output, not against marketing claims.

For manufacturers

The implication for manufacturers is structural. If the technical-distributor segment thins out because the application-engineering bench retires faster than it is replaced, the question for manufacturers becomes who explains the product to the buyer. Two answers exist. Manufacturers can absorb the technical pre-sale work into their own field-application teams, which raises their own cost base. Or manufacturers can support their distribution channel in equipping itself with the domain-specialized agent layer that preserves the technical pre-sale capability. The second answer is structurally cheaper for the manufacturer because the agent layer scales without proportional headcount.

For both: the standards anchor

Both audiences should read the agent-grounding requirement the same way. The credible domain-specialized agent layer for industrial chemistry has to be anchored to documented standards. The American Society for Testing and Materials, the International Organization for Standardization, the Association for Materials Protection and Performance, and the National Sanitation Foundation are not optional reference frames; they are the substrate that lets a buyer trust an agent's recommendation. An agent platform that cannot cite the standard it relied on is not yet trustworthy enough for the technical-distributor workflow. The standards anchor is the difference between a domain-specialized agent and a generic chat tool dressed in chemistry vocabulary.

VIII. Key Takeaway

• The chemical distribution market reached USD 265.15 billion in 2024 and is forecast at USD 543.01 billion by 2035, while the technical workforce that explains those chemicals to buyers is exiting at the demographic curve documented by the United States Bureau of Labor Statistics. The two trends collide on the technical-distributor segment.

• The distinction between delivery distributors and technical distributors is operationally critical. Delivery distributors face logistics-margin pressure; technical distributors face cognitive-output pressure. The squeeze on the second segment is what makes domain-specialized agent platforms structurally relevant rather than optional.

• Lubinpla's product design, AI Shooting for per-case analysis and AI Crew for subscription workflows, is shaped by the founder's combined background in data analytics, family chemical distribution, and B2B large language model engineering. The combination produces the methodological discipline of cited evidence, the operational understanding of the distributor's workday, and the engineering pattern of refusal-over-confabulation.

• The vertical AI thesis, articulated by Bessemer Venture Partners and Y Combinator, frames the opportunity in labor-line terms rather than software-line terms. Industrial chemistry distribution is one of the verticals where domain-specialized agents address a structurally larger spend pool than legacy enterprise software did.

• The trust-building sequence for technical distributors evaluating domain-specialized agents starts with one delivered analytic result, not a subscription commitment. The squeeze-factor matrix in Section V is the diagnostic instrument distributors can apply to their own books to identify which workday rows are degrading fastest.

See how AI Crew handles this workflow in your environment: https://www.lubinpla.com/ai-crew

IX. References

Bessemer Venture Partners. (2025). *The state of AI 2025*. Bessemer Venture Partners. https://www.bvp.com/atlas/the-state-of-ai-2025

Boaz Partners. (2024). *Bridging the talent gap in the chemical industry: Retirements and the need for successors*. Boaz Partners. https://boazpartners.com/bridging-the-talent-gap-in-the-chemical-industry-retirements-and-the-need-for-successors/

Contractspan. (2025). *Legal-grade AI vs generic AI for contract management*. Contractspan. https://www.contractspan.com/blogs/legal-grade-ai-vs-generic-ai-contract-management

DCAT Value Chain Insights. (2024). *Supply chains: Trends in chemical distribution*. Drug, Chemical and Associated Technologies Association. https://www.dcatvci.org/features/supply-chains-trends-in-chemicals-distribution/

Deloitte. (2025). *2025 chemical industry outlook*. Deloitte Insights. https://www.deloitte.com/us/en/insights/industry/chemicals-and-specialty-materials/chemical-industry-outlook/2025.html

ICIS. (2025). *2025 ICIS top 100 chemical distributors ranking revealed*. ICIS. https://www.prnewswire.com/news-releases/2025-icis-top-100-chemical-distributors-ranking-revealed-302481299.html

Inspectioneering. (2016). *NACE study estimates global cost of corrosion at USD 2.5 trillion annually*. Inspectioneering Journal. https://inspectioneering.com/news/2016-03-08/5202/nace-study-estimates-global-cost-of-corrosion-at-25-trillion-ann

McKinsey and Company. (2024). *Revolutionizing sales in distribution: Harnessing the power of AI*. McKinsey and Company. https://www.mckinsey.com/industries/industrials-and-electronics/our-insights/distribution-blog/revolutionizing-sales-in-distribution-harnessing-the-power-of-ai

NACE International. (2016). *International measures of prevention, application, and economics of corrosion technologies (IMPACT) study*. NACE International. http://impact.nace.org/economic-impact.aspx

PitchBook. (2026). *IMCD 2026 company profile: Stock performance and earnings*. PitchBook. https://pitchbook.com/profiles/company/51083-56

Polaris Market Research. (2025). *Chemical distribution market: Key growth drivers and trends by 2034*. Polaris Market Research. https://www.polarismarketresearch.com/industry-analysis/chemical-distribution-market

Precedence Research. (2025). *Chemical distribution market size worth USD 543.01 billion by 2035*. Precedence Research. https://www.precedenceresearch.com/chemical-distribution-market

Safic-Alcan. (2024). *Specialty chemicals distributor: Role and value explained*. Safic-Alcan. https://www.safic-alcan.com/en-it/industry-articles/global-specialty-chemical-distributor/

SuperAnnotate. (2025). *Vertical AI agents: Why they will replace SaaS and how to stay relevant*. SuperAnnotate. https://www.superannotate.com/blog/vertical-ai-agents

Talenbrium. (2025). *South Korea top 30 trending roles in the chemicals and materials industry: 2025 edition*. Talenbrium. https://www.talenbrium.com/report/south-korea-top-30-trending-roles-in-the-chemicals-materials-industry

TM Capital. (2024). *The chemical distribution industry: An end to the great destocking*. TM Capital. https://www.tmcapital.com/wp-content/uploads/2024/07/Specialty-Chemical-Distribution-Report-2024.08.01.pdf

Toffcap. (2024). *The specialty chemicals distribution market*. Toffcap Substack. https://toffcap.substack.com/p/the-specialty-chemicals-distribution

United States Bureau of Labor Statistics. (2025). *Employed people by detailed industry and age*. United States Department of Labor. https://www.bls.gov/cps/cpsaat18b.htm

Y Combinator. (2025). *Vertical AI agents could be 10X bigger than SaaS*. Y Combinator Lightcone Podcast. https://creators.spotify.com/pod/profile/lightconepodcast/episodes/Vertical-AI-Agents-Could-Be-10X-Bigger-Than-SaaS-e2rbedb