Short answer: Explain deep tech from the customer’s problem inward. Write down the problem in your customers’ words, turn it into one sentence a non-specialist can repeat, and put three layers under it: how it works, what proves it, and the specifications an engineer will check. Give every claim its rung on a proof ladder (lab result, pilot, independent validation, certification) and its test conditions. Both audiences read fast: investors spend an average of 4 minutes and 10 seconds on a pre-seed deck before deciding on a meeting (DocSend), and in our review of 30 funded B2B startup websites only 13 were clear within 5 seconds. Medical, environmental and AI claims also carry legal risk, so every number needs a source and an owner.
This guide is for founders of robotics, energy, materials, biotech, space and industrial AI companies preparing a round, a pilot or a first enterprise deal.
Why is deep tech hard to explain?
Two readers on every page
A deep tech company explains itself to two kinds of readers at once. The technical evaluator (the engineer who runs the pilot, the safety reviewer, a fund’s technical adviser) reads for the mechanism, the test conditions and the limits. The investor, the executive buyer and procurement read for the problem, the market, the timing and the risk. DocSend’s data on pre-seed decks shows 4 minutes and 10 seconds per deck on average, 33 seconds on the company purpose and 77 seconds on the product section, and DocSend advises testing the purpose sentence on non-technical people.
Founders tend to write for the first reader, because that is the reader they used to be. Economists call this the curse of knowledge (Camerer, Loewenstein and Weber, 1989): people who know more struggle to predict what less informed people will understand.
Long cycles and several rounds before revenue
BCG’s investor guide to deep tech found that deep tech investments take 25% to 40% more time between funding stages, from seed through Series D, than other tech investments, and that more than 80% of deep tech ventures build physical products. One explanation has to carry the company through several rounds while the proof moves from the lab to pilots to contracts.
Enterprise buyers add readers. A pilot of a new sensor, robot or material passes through operations, safety, quality and procurement, and most of those people see only what the champion forwards and what is on the website. In a Gartner survey of 632 B2B buyers, 69% reported inconsistencies between a supplier’s website and what its sellers told them.
More deep tech decks in every partner’s week
According to Dealroom’s European Deep Tech Report 2026, deep tech took 32% of all European venture capital in 2025, up from 15% in 2015, and US deep tech funding grew 92% to $144bn. With that many deep tech decks in a partner’s week, the word “breakthrough” no longer sets anyone apart.
Claims carry legal weight
Regulators read medical device advertising to decide what a device is for, the FTC’s environmental guides cover B2B sales, and the SEC has charged a founder over false AI claims to investors (see step 7).
What we saw on DACH deep tech websites
Our study of 30 funded B2B startup websites, 15 from Poland and 15 from Germany and Switzerland, found only 13 sites clear within 5 seconds. DACH sites scored lower than Polish ones on that dimension (3.1 vs 3.6 of 5), and many DACH deep tech sites opened with the vision instead of the application, often a three- or four-word line with no category. A defence-tech startup with one of the best-looking sites (design 4 of 5) scored 4 of 10 overall: no company details, no team, no proof of any kind. A space company mentioned a regulatory licence and a space-agency contract only in blog posts. Proof separated the best sites from the weakest more than any other dimension (a gap of 2.9 points on a 1 to 5 scale).
The method: seven steps from interview to website
Step 1. Start from the customer’s problem, in their words
Before you write a line, talk to the people who buy, use and check your technology, and plan each group separately: an engineer and a CFO want different things from the same explanation.
| Group | Who | What to ask | What you get |
|---|---|---|---|
| Buyers | Whoever signed or sponsored a pilot | What made you start looking? What did you compare us with? What did the pilot have to prove? | The problem, the trigger and the real alternative |
| Users and operators | The people who run the system day to day | What did you do before? What still slows you down? | The workflow and the words buyers use |
| Technical evaluators | Engineers, safety, quality and security reviewers | What convinced you? Which claim did you doubt? | The proof each claim needs, and the claims to drop |
| Investors | Current investors, investors who passed, their advisers | How did you explain us to your partners? What could you not explain? | The one-liner, the category and the risk the deck must answer |
A systematic review by Hennink and Kaiser found that studies using empirical data reached saturation within 9 to 17 interviews, mostly with fairly homogeneous groups and narrow questions. Plan that range per group and close a group when new calls stop adding triggers, alternatives or objections.
Ask about what happened, not about what people might do one day, and write down their exact phrases. “We find cracks after they have already cost us a shutdown” tells you more than “predictive maintenance is important”, and it often becomes the headline.
Step 2. Write the one-sentence explanation
Use this shape: [Product] helps [buyer] [achieve outcome] by [mechanism in plain words]. Add one proof line underneath. The sentence names the category, the buyer and the result. Leave out “revolutionary”, “next-generation” and “platform”; they describe your ambition, and the buyer is looking for their problem.
SkyInspection, a client of ours, flies drones over bridges, wind turbines and refineries and uses AI to find damage in the footage. Its old site sold “99.9% detection accuracy” and “60% cost reduction”, neither with a source. The new positioning reads “from drone inspection to failure prediction”, and the buying argument is that a defect you predict gets fixed during a planned shutdown instead of at 2 a.m. An operations manager can repeat that without knowing how the model works.
Test the sentence before you design anything. Show the first screen or the title slide for 5 seconds to someone in the target role and ask what the company does. If they repeat the architecture but not the consequence, the sentence is still too technical. If they get the consequence but do not believe it, the proof is too thin.
Step 3. Run the analogy test
An analogy borrows a picture the listener already has. It helps when the picture carries your real mechanism. Form Energy explains its iron-air battery as “reversible rusting”: while discharging, the battery breathes in oxygen and turns iron into rust, and while charging, an electric current turns the rust back into iron. The picture is close to the actual chemistry, so a listener who reasons from it ends up roughly right.
Analogies mislead when they map only the outcome or add abilities your technology does not have. “A quantum computer tries every answer at once” sounds clear and is wrong, which is why Scott Aaronson heads his blog with the line that quantum computers “won’t solve hard problems instantly by just trying all solutions in parallel”.
Give the analogy and ask the listener to predict one thing about your product from it: the cost, the speed, how it fails or what it cannot do. If the prediction is wrong in a way that matters, change the analogy or say where it stops working (“like X, except that…”). Keep analogies out of specifications and claims; those need units and test conditions.
Step 4. Layer the explanation
Each layer answers the next question a reader has, and each should make sense without the layers below it.
| Layer | Question it answers | Website | Pitch deck | Sales materials |
|---|---|---|---|---|
| 1. Headline | What does it do, for whom? | First screen | Title and problem slides | First lines of the one-pager and the email |
| 2. How it works | Why should I believe it can work? | Section under the hero, with one diagram | Product or technology slide | Middle of the one-pager |
| 3. Proof | Has it worked outside your lab? | Strongest proof next to the hero, case studies | Traction or validation slide | Pilot report, case study |
| 4. Specifications | Can an engineer check it? | Technical page, data sheet, papers | Technical appendix for the fund’s adviser | Data sheet, safety and compliance documents |
Investors usually read layers 1 and 3 and pass layer 4 to an adviser. Buyers read layer 1 and forward layer 2 to colleagues, and their engineers go straight to layer 4.
Step 5. Show it before you describe it
The product section of a pre-seed deck gets 77 seconds in DocSend’s data, among the longest of any section. In our website study, about a third of the sites (9 or 10 of 30, depending on the reviewer) showed no real product in the first two desktop screens.
Four kinds of visuals do most of the work:
- Before and after in the customer’s workflow: what the team does today, what changes and where your technology sits.
- A system diagram with five boxes or fewer: inputs, your component, outputs and the systems you connect to.
- Scale: Form Energy describes battery modules grouped in enclosures about the size of a shipping container, so a reader who has never seen a grid battery can picture one.
- A demo with real data: footage, sensor readings or model outputs from a real test, with the conditions in the caption.
Label renders and simulations as such. A procurement reviewer who later finds out that a photo was a rendering will doubt the rest of the page.
Step 6. Climb the proof ladder
Deep tech proof comes in rungs, and each rung supports a different claim. The EU defines nine technology readiness levels (TRL), from “basic principles observed” to “actual system proven in an operational environment”, and the EIC Accelerator asks applicants to show at least TRL 5, validation in a relevant environment.
| Rung | What it proves | How to phrase it | Typical TRL |
|---|---|---|---|
| Model or simulation | The principle is plausible | ”Modelled”, with the assumptions | 2-3 |
| Lab result | It works under controlled conditions | Result, conditions and sample size | 3-4 |
| Prototype in a relevant environment | It works outside the lab | Where it ran and for how long | 5-6 |
| Pilot with a named customer | It works in a customer’s operation | Customer, duration, what was measured | 7 |
| Independent validation | Someone without a stake confirms it | Who tested it and how, with a link | Any |
| Certification or authorisation | A regulator or notified body accepted it | The exact scope of the approval | 8 and up |
| Repeat orders | Customers pay again | Customers, renewals, expansions | 9 |
Say which rung each claim stands on. “Detects defects with high accuracy” tells an engineer nothing; “detected [X]% of defects in lab tests on [sample] under [conditions]” can be checked. Workado advertised its AI content detector as “98 percent” accurate, and the FTC says independent testing found 53% on general-purpose content, because the model had been trained or fine-tuned on academic text.
For investors, the ladder is also the plan. Commonwealth Fusion Systems describes the research behind its machines as work that “retired key risk” on the path to commercial fusion. Say which rung you are on, which rung this round buys and which risk that removes.
Step 7. Check every regulated claim
Medical devices. Under 21 CFR 801.4, the intended use of a device can be shown by “labeling claims, advertising matter, or oral or written statements”. A website headline or a sales deck can widen what the FDA sees your device as being for. In the EU, Article 7 of the Medical Device Regulation prohibits advertising that ascribes to a device functions and properties it does not have.
Environmental claims. The FTC’s Green Guides “also apply to business-to-business transactions” and cover claims “in any medium” (16 CFR 260.1). Marketers should not make unqualified general environmental benefit claims such as “eco-friendly” (260.4), and substantiation “often requires competent and reliable scientific evidence” (260.2). “Carbon-negative” on a slide needs that evidence.
AI. In March 2024 two investment advisers, Delphia and Global Predictions, paid $400,000 in total penalties to settle SEC charges over false statements about their use of AI. In April 2025 the SEC charged Albert Saniger, founder of the startup Nate, alleging that he raised over $42 million by telling seed and Series A investors that the app completed purchases with AI and without human involvement, while contract workers entered many orders by hand. The FTC’s Operation AI Comply began with the line that there is “no AI exemption from the laws on the books”.
Keep a claims register: every number and comparison, where it appears, its rung, source and test conditions, and who approved it when. When a number changes, you change it everywhere at once.
Investor version, buyer version and website first screen
The facts stay the same for every reader; the order and the depth change.
| Investor deck | Buyer materials | Website first screen | |
|---|---|---|---|
| The question they bring | Can this become a large company, and what has to be true? | Will this fix our problem without new risk to our operation? | Is this for me, and what is it? |
| Lead with | The problem, who pays for it, and why now | Their problem in their words, and the outcome | One sentence: category, outcome, buyer |
| Technology depth | One mechanism slide plus a technical appendix | Enough for the champion to defend it internally: integration, safety, maintenance | None on the first screen; layer 2 directly below |
| Proof that moves them | The rung you have reached and the rung this round buys | A pilot at a similar company, certifications, references | The strongest single proof, next to the hero |
| Risk to answer | Technical, market, regulatory and financing risk | Operational, safety and procurement risk | Whether the company is real: team, legal entity, contact |
| The ask | A meeting, then the data room | A pilot with written success criteria | One honest next step |
The buyer version needs material the champion can forward without you in the room. The website serves both, so after the first screen give investors, buyers and engineers their own paths. Our London guide to branding and pitch decks shows how to build the investor narrative from customer interviews, and one story for pitch deck, website and sales deck shows how to keep the three consistent.
Examples from public deep tech companies
We describe what these companies publish and add no numbers of our own.
Form Energy (long-duration batteries). Its technology page moves through the layers in order: multi-day storage, the “reversible rusting” mechanism, what is inside a module (including a non-flammable electrolyte “like the electrolyte used in AA batteries”), and the physical scale. It sets the product next to the alternative buyers know, lithium-ion batteries that provide energy for a few hours at a time, and presents itself as their complement.
Commonwealth Fusion Systems (fusion). CFS explains its technology as three steps with one job each: high-temperature superconducting magnets as the enabling technology, SPARC as the demonstration machine meant to produce more energy than it consumes, and ARC as the first power plant meant to deliver electricity to the grid. An investor can see which risk each step removes.
Proxima Fusion (fusion, Munich). Proxima answers “why now” in one sentence on its technology page: stellarators “were exceedingly difficult to design in the past, until the advent of supercomputing”. It calls its demonstration magnet hardware that “will retire key technical risks” and presents its programs as a path from Alpha, a demonstration stellarator, to Stellaris, its concept for a first commercial power plant. The plasma physics sits further down for the readers who want it.
In all three, the physics sits one or two layers below the headline.
Checklist
Before you send the deck or publish the site, check that:
- You have interviewed buyers, users, technical evaluators and investors, and saved their exact words.
- The one sentence names the category, the buyer and the outcome, and people in the target role can repeat it after 5 seconds.
- Every analogy has passed the prediction test, and you say where it stops working.
- The four layers are separate, one diagram shows where your technology sits in the customer’s process, and renders are labelled.
- Every claim has its rung, its test conditions and a source, and medical, environmental and AI claims have passed regulatory or legal review.
- The strongest proof sits next to the hero and on the traction slide.
- The deck says which rung the round buys and which risk that removes.
- A technical appendix or data sheet exists for the fund’s adviser and the buyer’s engineers.
- The claims register matches the website, the deck, the one-pager and every language version.
- The website shows the team, the legal entity and a contact, and offers one honest next step.
Worksheet
Fill in one row at a time. Leave a row empty rather than guessing: the empty rows show which interviews or tests are still missing.
| Field | Your answer | How to check it |
|---|---|---|
| The problem in the customer’s words | Quote it, with the role of the person who said it | |
| What they do today instead | Name the alternative: a manual process, an incumbent supplier, doing nothing | |
| Why now | A change in regulation, cost, technology or the customer’s situation | |
| One sentence | 5-second test with someone in the target role | |
| Analogy, and where it stops working | Prediction test | |
| How it works, in three sentences and one diagram | An engineer confirms it is correct, a buyer confirms it is clear | |
| Strongest proof and its rung | Source, conditions, date | |
| Next rung, and what the round or pilot buys | Milestone, date, budget | |
| Specifications and where they live | Data sheet, technical page or appendix | |
| Regulated claims and who approved them | Entry in the claims register | |
| Investor ask, buyer ask, website next step | Each is one concrete action |
How dp.vision works on this
dp.vision is a brand, web and product studio for funded B2B tech companies. Every project starts with interviews with our client’s real customers, and the explanation, the deck and the website are built from what those buyers say.
For SkyInspection, which inspects bridges and wind turbines with drones and AI, we audited every number the company wanted to publish and kept only those an outside source could support. We rewrote the positioning around the move from finding damage to predicting failures, put the flight authorisations and inspection standards where procurement looks for them, and made a pilot inspection the single call to action. The client reports 23 inbound demo requests in three months and 2 enterprise pilots after the new positioning and site.
Our rate is $85-100 an hour, with one fixed price agreed before work starts (how much dp.vision costs). Other studios that work with deep tech are in our list of deep tech branding agencies. For an outside read on your deck or website, get in touch.
Frequently asked questions
How do you explain deep tech to investors?
Start with the problem a paying customer has, then show what your technology changes, how far it has been proven and what the next round pays for. DocSend found that investors spend an average of 4 minutes and 10 seconds on a pre-seed deck before deciding on a meeting.
How technical should a deep tech pitch deck be?
Technical enough for a fund’s adviser to check, and layered. The main deck covers the problem, the mechanism in plain words, the proof and the plan for removing risk; specifications and test data go into an appendix.
Should a deep tech website speak to engineers or to buyers?
To both, in layers. The first screen tells buyers and investors what the product is, what it changes and for whom, and engineers find the technical page one click away. In our study of 30 funded startup websites, only 13 were clear within 5 seconds.
When do analogies help, and when do they mislead?
They help when the picture carries your real mechanism, as with Form Energy’s “reversible rusting”. They mislead when they add abilities you do not have, as with “a quantum computer tries every answer at once”. Ask a listener to predict one thing about your product from the analogy and check the answer.
What counts as proof before revenue?
In rungs: models, lab results with their conditions, a prototype in a relevant environment, a pilot with a named customer, independent validation and certification. Say which rung each claim stands on. The EIC Accelerator asks applicants to show at least technology readiness level 5.
Which deep tech claims carry legal risk?
Medical, environmental and AI claims. Under 21 CFR 801.4, advertising can establish a medical device’s intended use, the FTC’s Green Guides apply to B2B environmental claims, and the SEC has charged firms and a startup founder over false AI statements. Keep a register of every claim, its source and who approved it.
How many customer interviews do you need before writing the explanation?
Plan them by group: buyers, users, technical evaluators and investors. A systematic review by Hennink and Kaiser found saturation within 9 to 17 interviews in fairly homogeneous groups. Close a group when new conversations add nothing new.
Sources
- DocSend, Building your pre-seed pitch deck: https://www.docsend.com/blog/what-to-include-when-building-your-pre-seed-pitch-deck/
- BCG, An Investor’s Guide to Deep Tech (2023): https://www.bcg.com/publications/2023/deep-tech-investing
- Dealroom, The European Deep Tech Report 2026: https://dealroom.co/reports/the-european-deep-tech-report-2026/
- Gartner, press release of 25 June 2025: https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-sales-survey-finds-61-percent-of-b2b-buyers-prefer-a-rep-free-buying-experience
- Camerer, Loewenstein and Weber, The Curse of Knowledge in Economic Settings, Journal of Political Economy (1989): https://doi.org/10.1086/261651
- Hennink and Kaiser, Sample sizes for saturation in qualitative research: a systematic review of empirical tests (2022): https://pubmed.ncbi.nlm.nih.gov/34785096/
- European Commission, Horizon Europe Work Programme 2025, General Annexes (TRL definitions): https://ec.europa.eu/info/funding-tenders/opportunities/docs/2021-2027/horizon/wp-call/2025/wp-14-general-annexes_horizon-2025_en.pdf
- European Innovation Council, FAQs on the EIC Accelerator: https://eic.ec.europa.eu/eic-frequently-asked-questions/faqs-eic-accelerator_en
- eCFR, 21 CFR 801.4, meaning of intended uses: https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-801/subpart-A/section-801.4
- eCFR, 16 CFR Part 260, Green Guides: https://www.ecfr.gov/current/title-16/chapter-I/subchapter-B/part-260
- EUR-Lex, Regulation (EU) 2017/745 on medical devices: https://eur-lex.europa.eu/eli/reg/2017/745/oj/eng
- SEC, charges against Delphia and Global Predictions (2024): https://www.sec.gov/newsroom/press-releases/2024-36
- SEC, Litigation Release No. 26282, SEC v. Albert Saniger (2025): https://www.sec.gov/enforcement-litigation/litigation-releases/lr-26282
- FTC, Operation AI Comply (2024): https://www.ftc.gov/news-events/news/press-releases/2024/09/ftc-announces-crackdown-deceptive-ai-claims-schemes
- FTC, order against Workado (2025): https://www.ftc.gov/news-events/news/press-releases/2025/04/ftc-order-requires-workado-back-artificial-intelligence-detection-claims
- Form Energy, battery technology: https://formenergy.com/technology/battery-technology/
- Commonwealth Fusion Systems, technology: https://cfs.energy/technology/
- Proxima Fusion, technology: https://www.proximafusion.com/technology
- Scott Aaronson, Shtetl-Optimized: https://scottaaronson.blog/
