Autonomy on trial: What I saw in the Autonomy Wars

Another break from the Post-PC/techie world. As I'm sure you've noticed there has been something of a kerfuffle around HP/Autonomy this week. I have to confess I have some first-hand knowledge of the company, so I thought I'd report on what I saw at the time:

Autonomy strikes back from the dead

Mike Lynch: Making the news for all the wrong reasons.

Autonomy is one of the most contentious companies I've covered as an equity analyst. The market was always in two minds whether it was a real innovation-driven tech story (a rarity on this side of the pond), or an M&A-led roll-up with a healthy dash of questionable accounting. CEO Mike Lynch always believed he was cleverer than the rest of us, and with his PhD in mathemetical computing, millions in the bank and blue-sky pitch about Bayesian inference theorem. And who were we analysts to dare argue?

His coup de grace was to squeeze a $12bn bid out of HP which was rich by even the rarified standards of tech M&A (11x sales; 24x EBITDA). And note this wasn't just any old bid - HP paid up in cash (hint to CFOs: If you're a distressed buyer and the seller has an information edge, probably best to share the downside by paying stock), and the deal was forced through with exquisite timing despite the concurrent defenestration of its architect, Leo Apotheker. Once Autonomy disappeared into HP's gaping maw we assumed it was a done deal. Mike had finally proved, with his $800m payout, that he was cleverer than the rest of us.

Of course on Tuesday Autonomy struck back from the dead, prompting a $8.8bn write-down at HP and furious accusations of accounting improprieties.

The case for the prosecution

In a nutshell HP is alleging three things:

1. Low-margin hardware sales (c10-15% of group revenues) were booked as software licence sales.
2. Licensing deals with resellers were used to bring forward revenue from future periods.
3. Licensing deals with resellers were used to create revenue where no ultimate end-user existed.

(By way of background, I previously wrote about the benefits of a classic software business model here. Some of the stuff discussed below relates to that)

A few things to note.

Higher software revenue impacts profitability

Booking low-margin hardware sales as software has two impacts on Autonomy's financials. Firstly it inflates gross margin, as hardware typically has a lower gross margin (20-40%) attached to it than software (80-90%). However bear in mind that it in the long run it does not change the overall operating margin of the business, as costs associated with the hardware will still be taken further down the P&L (apparently here in the sales & marketing line). Software investors tend to focus on operating margin much more than on gross margin, as the majority of costs in this business model lie in S&M and R&D.

There's a catch though. One way this misallocation could affect profitability is if the phasing of the costs for software is different for the costs of hardware. For example if you sell a hardware box the cost of manufacturing that box is going to be booked as a cost at time of sale. However if you sell software the R&D cost associated with that might be capitalised and taken over the life-cycle of the product. On, say, a five year view the impact would be neutral but it reporting more software sales mean higher margins in year 1 and lower margins in year 5.

Software licence growth is a big deal for your P/E rating

Secondly showing higher licence growth is beneficial to the valuation (e.g. P/E ratio) the market attaches to a stock; investors tend to prefer software businesses showing fast-growing software revenues as that is seen as a lead indicator for growth elsewhere in the business (particularly in ultra high-margin maintenance revenues). Or to put it another way, if Autonomy started to show licence declines, the share price would have gotten trashed.

This is serious.

Channel stuffing seems strange at a software company

As are the accusations over reseller sales. Historically Autonomy would sell its software via both its direct sales effect and via resellers (typically IT Services companies or systems integrators). HP's second accusation is effectively that it engaged in "channel stuffing" - selling more inventory to resellers than they were selling on to end-users. In effect this means products which wouldn't be sold to end-users until next quarter could be booked as revenue this quarter.

Note that channel stuffing can be legitimate. Hardware companies, for example, report sales into the channel as revenue, not to end-users (e.g. Apple's reports a sale when it ships an iPhone to a retailer such as Best Buy, not when a punter buys it in a Best Buy store). They would typically engage in channel stuffing ahead of a busy period, such as the holiday season.

However there are a couple of funnies here. The first is Autonomy's own revenue recognition policies laid out in its annual report (see page 51)

The group enters into OEM and reseller arrangements that typically provide for fees payable to the group based on licensing of the group’s software to third party customers. Sales are generally recognised as reported by the OEM or reseller and is based on the amount of product sold.  Sales are recognised if all products subject to resale are delivered in the current period, no right of return policy exists, collection is probable and the fee is fixed and determinable.

The key phrase is "Sales are recognised if all products subject to resale are delivered in the current period...". If that means delivered to the end customer then Autonomy shouldn't have been booing channel sales as revenue; its not completely clear if this is the case, but the fact the clause also mentions right of return policy (again something which would be attributable to the end customer) implies that it is.

The other strange thing is that discussion of inventory and channel stuffing as a whole is highly unusual for a software company, simply because there is no physical product to stuff into the channel. Software companies sell licences to software code, typically downloaded or delivered on a 5c DVD. If Autonomy really was selling high-margin software it is very strange there would be enough product to legitimately stuff into the channel to have a meaningful impact on results.

However it gets worse.

Fictitious revenues is the cardinal sin (but quite easy to do)

The third allegation is the most serious - that deals were created with resellers to create revenue where no ultimate end-user existed. That is more than channel stuffing - that is outright fraud. In the great scheme of accounting shenanigans, manufacturing revenues by creating fictitious customers is a) the cardinal sin and b) surprisingly easy to do in a software company where there is no physical product to manufacture and ship.

The evidence: Seven things I observed in the Autonomy Wars

At the moment HP hasn't shown us the direct evidence which underpins their accusations. However over the years at Autonomy there was a great deal of circumstantial evidence that something was amiss. Nothing that could be proven in a court of law but a number of things which said "this doesn't quite smell right".

1) Low deferred income & high accounts receivable

See this blog article for a good snapshot of the situation at Autonomy. In a nutshell receivables (money owed for services already rendered and booked as revenue) were high and deferred revenues (money received for services not yet rendered and booked as revenue) were low. This was an inversion of the normal situation at software companies where receivables were typically low (cash paid upfront for licences rather than being collected later) and deferreds high (cash collected upfront for maintenance and support yet to be rendered - SAP is a great example of this).

2) Mixed cash conversion

Cash conversion is the proportion of accounting profit (net income) converted to cash profit (operating free cashflow). This sort of relates to the first point as high receivables generally mean low cash conversion. Normally cash conversion is very high (as I said in my earlier blog post, software companies have little physical plant or inventory tying up money so are high cash generative). At Autonomy, if my memory recalls (I no longer have my old model with me, alas) it was mixed. Sort of in the 70% range (I think). Not the worst I've seen but not what you expect from a top-class software model.

Note that there can be legitimate explanations for weak cash conversion. In particular if a business is growing fast (which Autonomy claimed it was) it would be reasonable for capital to be committed to support growth, either for vendor financing or to support a sales & marketing push. However if it continues to be weak when growth slows, particularly, that is a warning sign.

3) Unusually high operating margins

The third thing that was a little strange was Autonomy's high operating margins, typically 40%+. A classic software business model is generally capable of doing 20-30% operating margins no sweat. 40% is possible, but normally only in mature businesses with large cash-cow installed bases and lower marketing costs (Oracle and Microsoft are classic examples). Autonomy claimed to be having it both ways - but growing fast but also producing mature-company margins. I can't think of another software company which was reporting a similar growth/margin profile. Meg Whitman's comments that Autonomy's underlying margin was actually 28-30% sounds a lot more like reality.

4) An acquisitive track record

Autonomy was a perennial acquiror. It was a classic "roll-up", conducting a series EPS-boosintg acquisitions (Interwoven, Verity, Zantaz etc) each time growth appeared to be slowing. This had several impacts on the numbers. For one thing the consolidation of the balance sheet of each target (and the taking of merger provisions) obscured the numbers, making it very hard to track true organic growth.

For another it enabled Autonomy to acquire new customers to cross-sell into - costs which would have flowed through the P&L as sales & marketing costs if the customers had been recruited organically. There is nothing illegitimate about this, but I always thought that to reflect the true nature of the business Autonomy's margins should have been looked at including acquisition of goodwill (typically taking 500bp of of that 40%+ operating margins) as it was effectively a S&M cost.

The other funny thing about Autonomy's M&A was that they were remarkably quick at integrating acquired products into their IDOL platform. Without fail they would happily report a couple of quarters after acquisition that all new products were integrated into their suite - this is surprising given the complexity of the engineering task. As this Forbes article alleges the reasons was that products weren't actually integrated.

5) A persecution complex with analysts

The fifth strange thing was their bizarre relationships with analysts. Autonomy's various run-ins with critical analysts were well documented even before the HP deal. From my recollection I cannot recall seeing a company so obsessed with what analysts thought of it. In particular each quarter there would be a "review of analysts models" slide which, though not labelled as such, was seen as tacit guidance. Autonomy was the only company to take this step, rather than simply including a formal guidance statement.

I think there was something of a vicious circle to Autonomy's relationships with the analyst community. The more people wrote bearish notes on Autonomy the more the company would circle their wagons and the more bearish the bears would get. This wasn't helped by Autonomy's quarterly reporting cycle (unusual for a UK company). Given the volatility of quarterly results (especially for a smaller companies) there was always going to be some "funny" in the numbers each time for people to jump on. Sometimes this was legitimate, sometimes this was just noise. Of course if Autonomy had reverted to a more sensible half-yearly reporting structure that would have just made people howl that they were hiding something!

At the end of the day  good companies are generally ones which look after their business and then let the share price take care of itself. Autonomy seemed to take the opposite view.

6) A dangerous internal culture

The sixth worrisome sign was the internal culture of the company. Autonomy struck me as having a dominant CEO with little formal constraints on his power. CFO Sushovan Hussain did not strike me as someone who, when push came to shove, would stand up to Mike.

Also, like many software companies, there seemed to be a Darwinian sales-driven culture (see the anonymous comment at the bottom of this Reuters story for (what may be) a eye-popping account). That is not to say that sales-driven cultures = accounting fraud, rather that sales-driven cultures tend over-promise on the topline, and be incentivised on a rising share price. This creates fertile conditions to try and cover it up when sales (inevitably) fails to deliver and threatens to derail the equity story.

7) A strangely static tech pitch

Autonomy's technology pitch: "I know more Greek

symbols than you. Now please buy my software"

The last thing was Autonomy's technology pitch, which was basically "We have a black box which revolves around Bayesian Inference and Shannon's Information Theory. It searches stuff". The funny thing wasn't the pitch though - black box companies aren't great (Warren Buffet memorably never invests in anything he doesn't understand), but they do exist and flourish. Just look at Google (hint: I doubt its search algorithm relies much on tracking html linkbacks to websites anymore).

The funny thing was that the pitch never changed, right from when I first came across Autonomy in 2000 to when it was finally taken out. Now imagine a major tech company who's pitch hasn't changed in 12 years. You can't. There isn't one. Twelve years ago Apple was a desktop computer company with a sideline in media players. Now its a phone company with a sideline in desktop computers. But Autonomy never changed.

Now none of these factors were a killer in and of themselves - I've seen them all occur in isolation at many well-run and highly profitable companies. However together they were all a bit... Strange.

The case for the defence (and other logical fallacies)

Of course Mike Lynch isn't the sort of guy to keep quite about this. He was busy making the rounds of TV studios yesterday putting his case. In a nutshell:

1. He admits there was some hardware reselling involved, but this was openly disclosed and only c2% of revenues.
2. The Autonomy numbers were audited by Deloitte and subject to due diligence by KPMG plus a horde of associated bankers. If there was something going on they would have spotted it (amusingly, Meg Whitman has been making the same point to justify why she wasn't responsible, except saying if there was something on they should have spotted it).
3. Autonomy was fine when it was bought and has been wrecked by mismanagement post-acquisition, hence the write-down.
4. HP is using this as a smokescreen to hide problems in its own business.

Thinking about these, I think Lynch's first point has some merit. However the others are logical fallacies, plain and simple.

Bundling hardware is legitimate - the question is how much was going on

I think Mike has a point about hardware sales. Autonomy did talk in the past about bundling software and hardware, particularly with its IDOL Search Appliance, a $1m+ box which combined Autonomy software on a dedicated, high-speed server. Given the numbers Lynch has talked about I think this product is what he is alluding to.

Bear in mind that bundling hardware and software is a perfectly legitimate practice - it's what tech companies call selling a "solution". Everyone does it to some degree. Arguably Cisco, for example, is simply a software company which bundles its programs on generic low-margin router hardware and sells them for a premium. Similarly SAP has been blowing its trumpet for the last two years about HANA - its bundling of an in-memory database on a commodity hardware server.

It sounds like the debate isn't that it was being done, but the magnitude. If hardware sales were 10-15% of revenues as HP alleges, that would not be something I can remember then talking about before.

His other arguments are, however, red herrings.

Throwing Deloitte under the bus - the fallacy of the Appeal to Authority

With regard to the due diligence whether it was spotted or not by Deloitte, KPMG or the man on the moon is a reflection on their competance, and has no bearing on whether it actually took place. This is a logical fallacy known as the Appeal to Authority: "Deloitte says its so therefore it must be so".

HP's subsequent behaviour - the fallacy of the False Dilemma

With regard to what HP did or did not do to Autonomy after acquisition again that is not the issue. The question is what happened before HP bought Autonomy. HP's (undoubted) management ineptitude is another matter. This is a logical fallacy of the false dilemma: HP wrecking Autonomy post-acqn, and Autonomy have dodgy accounts are not mutually exclusive events.

HP's smokescreen - the fallacy of the Appeal to Motive

Finally saying HP is using this issue as a smokescreen is a logical fallacy known as the appeal to motive, where you impugn an opponents reasons rather than his reasoning.

As I said before we do not have the full evidence, but I don't think Lynch's case adds much to the defence argument. Anyhow let's see what HP come up with. And let's not underestimate HP's unending ability to shoot itself in the foot (on that note I actually saw someone using a Palm Pre the other day. Wonders never cease!).

(PS if you want to understand more about formal and informal logical fallacies, I can thoroughly recomment Masen Pirie's How to Win Every Argument: The Use and Abuse of Logic. Essential reading for anyone who's ever been involved in an online flame war!)

Lessons to be learned

So what are the lessons to be learned from this mess? Very early days but here are the things that jump out at me.

The dangers of shorting bad companies
Be careful shorting dodgy companies when there is a bid out there. This sounds sort of counter-intuitive (surely you'd want to sell the bad companies?), but don't forget that up until this week if you had bet against Autonomy on the back of these you'd have had your face ripped off. Every time the bears got too dominant the company would pull out (manufacture?) a blow-out set of results or another acquisition to muddy the waters. And eventually they found (what they thought) was the ultimate get-out and found an even greater fool (HP) to sell the business to.

Ironically the right investment thesis on Autonomy (the shares) as opposed to Autonomy (the company) was "buy the stock, because Mike is clever enough to pull it off".

Another good example was actually a former competitor of Autonomy's, Norwegian company FAST Search & Transfer. This was another one that always had questions asked about its numbers (culminating in an accounting restatement in 2007). However in the end they got a $1.2bn bid from Microsoft before the whole story could come out (I sort of remember MSFT taking a write-down on that one too, but a quick scan of the 09/10/11 10-Ks didn't throw anything up).

Another blow to Europe's tech hopes

It hasn't been a good few years for European tech. With the implosion of Nokia and Alcatel-Lucent we now have very few global tech leaders. Off the top of my head SAP, ASML, ARM and perhaps Dassault Systemes. Capgemini in IT Services (does that count as "tech"?). Most of the mid-sized names are slowly getting picked off - in the last year or so Autonomy, Logica and Misys in the UK alone.

What's more worrisome though is that there are few companies coming through with the to potential to become $10bn (or even $1bn) tech players. Most of them sell out before they get to that stage. I was sad when Autonomy was bought out, because at the time it had the best shot of any of them at making it big. And it would be particularly galling if accounting issues were proven, after the examples of FAST and iSoft in the last few years. What is it, investors may ask, about European software companies and dodgy accounting?

HP are a complete bunch of lemons

There are a lot of I-told-you-so's flying around this week, but the plain truth is there was reasonable doubt over Autonomy before HP came in. The seven points I listed above were all transparent to everyone before HP turned up with a bid at 11x sales. What was even more galling was that after the defenstration of Leo Apotheker they had a golden opportunity to walk away from the deal (the only barrier being a trifling break-up fee) but instead pressed on with it.

Even so given Autonomy's history you would have thought they would have hired the SAS to do the due diligence. At the very least they should gotten a team of forensic accountants in and pulled all the historic customer contracts. Now that's no guarantee (contracts can be faked, as we found out at the recent iSoft trial), but it should have least given an inkling of the proper licence/hardware split.

In this case, I'm afraid, there is no excuse.