In 2015 I co-founded CompanyHouse, a LegalTech attempt to digitise company registration and corporate-change filings in Iran. The premise was simple and, I still think, correct. Registering a company here requires an expert because the process is deliberately opaque. The expertise is the product — and the bottleneck.
We built what that premise implies. We modelled the registration specialists' tacit knowledge into a guided, question-based digital workflow that generated the filings. Company-type selection, data entry, document generation. We worked with legal experts to convert rules, decision points and exceptions into product logic. On paper it was a clean piece of product design.
It ran for a year and closed. It was my first serious experience of productising specialist knowledge, and the lesson it left me with is the one I have used most since: productising expert knowledge is a knowledge-extraction problem long before it is a software problem.
Why extraction is the hard part
An expert can describe the path. What they cannot easily describe is the exception — the moment where they look at a filing and something is off, and they route around it without registering that they have made a decision. That is where the value sits. It is also the part that is invisible to the person doing it, because to them it is not a decision, it is just Tuesday.
So the test of whether you have productised anything is not whether the described path works. It is whether the undescribed one does. Until you can model the questions the specialist asks and the exceptions they know, you do not have a product. You have a service with a login screen, and the specialist is still the bottleneck — now with a worse interface.
The same problem, five more times
Once you have named a pattern you start seeing it everywhere, and most of my work since has been some version of it.
At Bamana, co-founded in 2020, the gap was between expert knowledge and a decision at 11pm on a Tuesday. Parents drown in parenting advice and starve for parenting guidance. The intervention was not content — it was structure: taxonomies, need classifications and content pathways that turn specialist knowledge into something a tired parent can act on. The extraction step was organising what specialists know around what a parent has to decide, rather than around how specialists file it. It shipped and iterated over four years.
At Mehrabani and Soha the tacit knowledge belonged to social workers. Charitable giving fails on both ends: donors cannot verify need, and social workers cannot prove outcome. Trust collapses in the middle. The judgement that a need is real sits in a person's head, and it does not travel. So we built the Mehrabani Card — one object carrying the verified need, the beneficiary's story, the donation and the outcome — with Soha underneath as the verified-needs data layer. The part I am most confident about is the least glamorous: we designed the operational workflow social workers actually use. If the extraction layer is a burden, it does not get filled in, and then there is nothing to productise.
At WritingChex the expert was an IELTS examiner. A candidate outside a major city has no examiner, no feedback, and no way to know why their writing scores what it scores. We built an AI feedback engine that simulates the exam, analyses the response and returns specific, actionable, examiner-shaped feedback, then designed the improvement loop around it. It reached MVP validated with real submissions and first paying customers — the first proof that people would pay for machine feedback. It wound down in 2025. I was honest with the team about where the model could not be trusted, which is the CompanyHouse lesson applied earlier: the exceptions are where the value is, and a system that is confident on the exceptions is worse than no system.
Ketabno inverted the problem. The knowledge to productise was not an expert's — it was what makes reading stick for a teenager. Teenagers do not stop reading because books are bad. They stop because reading is solitary, unrewarded and invisible to their friends. So the product carried the social layer: story journeys, points, medals, narrator characters, quizzes. It launched in roughly three months.
Where it finally worked properly
Shalize is the clearest case, because the tacit knowledge there was operational rather than professional. A physical gold and silver business was running on WhatsApp messages, memory and trust, and volume was climbing faster than the process could hold it. Memory does not hand over.
We rebuilt the operating system end to end — enquiry, order, invoicing, payment verification, fulfilment, follow-up and buyback — and put a CRM under it. The real launch was not the website. It was sales having to stop working from memory; CRM adoption was the launch. Daily physical volume moved from 0.5 to 10 kg/day, and the business reports 4,000+ verified purchasing customers and 1,000B+ Tomans in gross transaction volume. Those figures are company-reported, from the Tamin Tejarat Espahbod Khorshid portfolio, not audited, not revenue and not net margin. I would rather say that than let them do work they have not earned.
Mokaab is the companion note. I architected a full investment platform — buy, sell, hold, buyback, pricing, invoicing, delivery and account journeys — with the buyback obligation designed in from day one rather than bolted on. It never launched: the parent group's priorities moved to physical operations before the platform shipped. But the architecture and the market model survived and were folded into Shalize's operating system. Extracted knowledge outlives the product it was extracted for. That is the argument for doing the extraction even when you are not sure the thing will ship, and it is why I list the failures — a portfolio without them is one you cannot calibrate against.
At Emkan the same move applies to policy. Territorial development is discussed in language no product team, no investor and no citizen can act on. So the work was translating it into structures a product team could actually build: development missions, economic zones, sectors, library, gallery. For Invest Iran, the investment-promotion agency platform, the tacit knowledge was how a serious investor is actually handled — architected into a full investor journey, from opportunity discovery and competitive-advantage assessment through expression of interest and organisational engagement to request management and aftercare.
The test is simple and unkind. If the expert goes on holiday, does the thing still work? If the answer is no, you have not productised anything yet. You have documented a bottleneck.